AI GTM STAGE

Welcome everyone, and thank you for joining us today. Our program will begin shortly. Please make your way to your seats. As a courtesy to our speakers and fellow attendees, please take a moment and silence your devices. Welcome back to day 3 of SASTR Annual 2026. Please take your seats. Our program will begin shortly. Once again, welcome Amelia Larue and Jason M. Lemkin.

Woo!

I don't see any hero video. Didn't roll?

For me?

Yeah.

Oh, I don't have a video.

You said there was an intro video.

No, just our voice of God.

Oh, I see. Voice of God. All right. Thanks, early birds. Thanks for being here. Hopefully, if this has been a good event so far, thumbs up. You can give us a thumbs down. Woo! Okay, not a lot of thumbs down, but quick, we should have done a slide just so you know on the agenda for today. It'll be tighter in focus 'cause it's day 3. Some are tired, some are rolling late. We are gonna go through as much detail as we can on our agent stack. We've talked a lot about it at this event, but we're gonna go into as much detail as we can, show you backends, show you how they really work on the other side. Really, we could do an hour on each, and maybe we'll do that over the course of the year on our live webinars. We're going to do as much as we can in this session, just to really give you a sense of how they work. Maybe it's our 401 class. Right after this, we'll have the head of all marketing for Google Cloud here, right, with Amelia, Sarah Kennedy. And it's the number one most on fire hyperscaler of everything. So they'll talk about why. If you want to know about macro, what what the hell's happening with hyperscalers. I think it'll be a unique person and time to have here.

So just come to the summit stage. You'll see me walking around.

Okay, so she'll be at the, which stage? She'll be over there. Yeah. That stage over there. Yeah. We'll have, okay, after that we'll have the CEO of Gamma too.

Yeah, so you've got two options after this one. So I'm gonna be talking with Sarah Candy over there on the summit stage. You can, anybody can come, it's open. And then here we'll have Adam from Owner.

Adam from Owner. Yeah, Adam from Owner crossing— has an AI CMO, how it works. They've just crossed a billion in revenue for SMBs and restaurants with their Agentic CMO. So Adam will be here, it'll be great too. That's some good competition between Google Cloud and Adam, right? Uh, then after that we'll have two— Gamma and Klaviyo at the same time.

Exactly.

Then you'll have another choice at 11, at 11, Founder Brunch, Founder VC Brunch.

It's three choices.

At 11, you— there's a founder— if you, if you want to fundraise or you're an investor over in the VC VC lounge, uh, where is it? That way. That way. Um, there'll be a founder VC meetup at 11, and then there will be— Amelia will be with the CEO of Gamma talking about how they scaled, which will be pretty interesting. And I'll be here. The CEO of Klaviyo, uh, you know, one of the biggest public leaders, um, in B2B for e-commerce, will be here sharing how they built their agents. Andrew's great. Andrew Beilack is one of the greatest founders, so that's a tough one, right? Then we'll do the pitch competition. Yeah, for AI startups. And then after lunch, then we'll close out with an AMA with me. You can leave, you can have a cocktail, but I'll literally take all your questions to the extent I can help until I pass out in the afternoon. So that's kind of a tradition. But with that, I'll let, hopefully you can process that, you can see it on the website too. Well, I'll let Amelia kick it off with, we'll do a deep dive. We won't do all of our 20 agents, but it looks like we're gonna do about 7 or 8 of the ones that really make a difference.

Yeah, I tried to force rank maybe the top ones that we use every day. So I had, I actually used Gamma last night when I was tired after the VIP party. Party to make me a little agent card stack, if you will. So we got to put the screen on. 20 agents that we've vibe coded. Some are vibe coded, some are not. So I'll explain the differences. And they've been used over a million times. And collectively, our agents have had multi-millions numbers of interactions. And so I thought it'd be fun to just make a little— everyone's been asking us what our stack is this week. And so I thought I would make them as Pokémon cards so that you can also collect them all. If you want, catch them all. Okay, so a few of them here that we'll go through today. But just to give you a quick run through of some of our top agents that we use every day. So 10K, you've heard us talking about, hopefully we came to the session on Tuesday where I walked people through how to build your own AI VP of marketing. 10K is our VP of marketing that we use. If you did miss that session, it's on the livestream, so you can go back. I do a whole hour on how to deploy your own VP of marketing. Qubi, who we now call Qubi, is our AI VP of customer success. He's handled all the sponsors you guys see here today with some human interactions, of course, but he's done a lot of the heavy lifting to help the sponsors and our customers that we're loading into the SaaStry Annual this week. There's Annie, who you guys all know because that's the SaaStry Annual website. She actually turned into an agent the more that we did with her. She has the parking pass app. She has a few other things she can do. And so started as a website, turned into an agent. We'll explain how that works next. Amelia AI, which you can also demo over there if you want. That's our inbound agent that runs off qualified from Salesforce. She— we've had her since like last summer, I think. And she is the one that is driving all of our inbound sales, handling all of our attendee questions like you guys have been asking her for, like Wi-Fi and all that stuff. Where's this? Where's that? That's Amelia AI. So she's part seller, part customer support, part marketer. She does a lot of things. Then there's AgentForce. I made it look like a ghost because we use AgentForce primarily to follow up with leads. That our sales team never wanted to follow up with. So I call that one— you can call it Agent Forest. I sometimes call it King Boo because he's a ghost. So that's Agent Forest reviving our dead leads that we never followed up with. The two other ones we use for sales as well. So Ava, that's Artisan. You guys have all seen Ava on their booth this week. We use her for a mix of both cold and warm outbound. And then Monica, who's newer to our staff, we use it for completely cold outbound to prospect. And companies that look and match our ICP. So it's real, Monica's really good at taking lookalikes for companies that you already have and that you've already closed, and then just automatically filling your pipeline with companies that look exactly like that. We do have other agents we've deployed. I don't know how much you wanna touch on these, Jason, but I just wanted to show the top ones.

Yeah, we can keep going. If you haven't seen it though, you can, and we'll update it with all the content from today and this week. But if you go to sastr.ai/agents, Agents. That's continually updated with everything here, the best sessions, all the decks, all the slides. It's good today, but by next week, it'll have everything from here organized for you. So you can come back to that.

Cool. So let's start with 10K. If you missed any of our previous demos on what 10K does, because he does a lot of the heavy lifting for us. So one thing I wanted to preface, because we just showed you our stack, some of the ones that we built ourselves, we built on Replit. So I'll call out in these slides what it's built on, what's kind of powering all of our different go-to-market agents. But one kind of common theme I noticed as I was building out this deck and just thinking about our stack of AI agents, if you will, because folks were just asking us like, hey, what's in your stack? What are you guys using? Is a lot of these did not start as full-blown agents. It's just something where we wanted to replace something. I wanted to replace our Notion dashboard because I got tired of copy-pasting things from like our different data sources. And so 10K really started as just a dashboard back in January. That was his thing. I just wanted to hook up to Marketo and Salesforce and really simplify life. And it really was just a dashboard for a few weeks. But then 10K evolved, if you will, and now does a whole lot more. He owns the number. He helps us track our daily revenue across all of go-to-market. He helps with our forecasting, which is super important and helpful for something like this where you have really time-sensitive, maybe releases or product things that you guys are doing as equivalents. He knows all of our campaign performance. Now. So he's tracking all of our marketing campaigns in real time. He pushes out our, our daily marketing ideas. We, we've been talking about this week, our top 3 ideas that we get every day for SaaS comes from our AI VP of Marketing. Um, and it's interesting too, I did a little test comparison this week where on Tuesday I rebuilt our AI 10K VP of Marketing on Replit, and then yesterday I did a little bit live with Elena on Lovable. And actually the ideas that came out with were were different. It's like the LLMs were so different that it came out with 4 different ideas.

So you tried to rebuild 10K on Lovable as much as you could, right? And what, what was different?

So the, the biggest, like, the dashboard still, you know, kind of looked similar. Like, there were elements that were the same, but the underlying brain was different because the, the biggest test for me was like when I gave it to Replit, I think the top 4 ideas it gave was like more email-based. And then the ones that Lovable came out with were more like Okay, you need to do ads, you need to spend on this, you need to do—

oh, the idea is that the seeming brain— we got to be careful not to anthropomorphize AI, but the idea is that the lovable version— we're different than Replit.

Exactly. Yeah.

Yeah.

So I gave them both the same spec and the same inputs.

You know, we could run it. We should run an experiment. We'll follow up. We should get them both running for real and have them debate and argue their 3 ideas a day. They each come up with 3 ideas a day. It's too many. Let's have them compare and contrast, because for reasons that we probably are not smart enough to understand, 10K is very biased on email marketing, right? And the dataset. And the Version Unlovable is very biased on advertising.

Advertising, brands. Yeah, it was a lot more branded. Sounds a little like Elena. I know.

That's pretty interesting that they came to different conclusions.

Yeah, it is funny. I was like, it actually kind of matches the marketing style of those internal teams. And I don't know. Maybe it's not intentional that however their internal teams are using it for marketing is like, oh yeah, Replit's nerdier and Lovable's more marketing.

Yeah.

So it's a little funny.

Yeah, it is. We'll, we'll thread some stories in this about how it is surprising how seemingly agents with seemingly similar datasets and LMs come to different, good, but different conclusions. I'll share a story from this morning later. The one thing I would add to this, Emilia, if you were here on the first day in the morning for the 201 session, Emilia gave you a recipe. She gave you the skills. We'll post it online. You can build your own 10K. We don't wanna duplicate duplicated here. But the one thing I would add, I think Amelia hit on it, the simplest way you can get some of this done, like the most basic version of this, and Amelia will go deeper, but this headless Salesforce idea that's in the media, if you haven't tried it, it's really effin' powerful, okay? Salesforce just launched something called Headless Salesforce if you're on Salesforce, and some of that is actually a product that doesn't even exist yet. It's on MuleSoft, which they bought years ago, and a bunch of stuff that most folks are never gonna use. You're not using— who's in the room using MuleSoft? Okay, nobody. Okay. But most of you are using Salesforce. The Salesforce API is really good. So if you do nothing else, spin up a Replit or Lovable or v0 from Vercel and tell it, and then just tell it you want to go into Salesforce and build— I want a type of dashboard or analytics I don't get today. Okay. And we'll show you, like, for example, for me, I wanted, I want real-time visibility into attendance and ticket sales here every hour. That just doesn't work in Salesforce, right? So that actually took two APIs, Bizible and Salesforce. But my point is, now I can interact with Salesforce data in a way I never could. And this is— you can get a quick— ours started off as a dashboard for tickets and attendance, but it could have been a Salesforce dashboard for sure. You do not— if you really want to do something cool, just go to— go to— you can do it in Cloud Code too, but I actually think Reputlovable or these are easier, these connect. And just say, connect, connect me to Salesforce with their connectors, and I wanted to do this and watch some magic happen. They already work headless and agentic. You can get 10% of what we do in an hour and maybe do something really cool that you would never be able to do with Salesforce on its own because these agents are so powerful.

Yep. So just a little bit of the backend of 10K. So I did a version of this for each of our agents that I showed as our top agents. So 10K was built on Replit. His first commit was in January of this year. So he's really not even— he's only 4 months old, actually. But it's funny, I asked each of the agents how many commits we've done. So this one has almost 1,000, which is a lot. 7 to 8 commits a day, so that's between myself and Jason. And then for fun, I put how many lines of code each one of them was. But the more of the— and then there's, there's— 10K has the most external APIs wired into it. So this is what we mean when we say, you know, headless Salesforce is we're using Salesforce just directly via the API without logging in directly to Salesforce. You know, I, I still, I still log into Salesforce. Jason doesn't have a Salesforce login, and I still have to, like, you know, use the Authenticator app when I go, go move things around. Um, that's not in the API yet, but, um, so that's, that's what we mean by headless, if you guys are wondering what, what all that means. Bizabo, so that's what we use for ticketing. That's how you guys all bought your tickets. And then Marketo, which was our marketing automation platform, um, it's still hooked up to this for now. And then if Slack, that's how we get our daily, every day when we get our daily marketing ideas, we get an email with it, but we also get a Slack update. QB also has Slack, so QB will Slack us a morning customer report essentially of, hey, here are the accounts you need to follow up with, here are some of the sponsors who maybe didn't finish what they were supposed to be doing, here are people that are overdue, and then it'll also Slack us and say, hey, here are some folks in your customer base that we think you need to follow up with 'cause maybe they haven't done haven't done anything, or they're super behind on things, or they're approaching a deadline that they haven't quite met yet. So it can do a lot of that automated for us, and so I really like 10K and Slack. And then you'll see Clerk in there too, but those ones at the top are really, I think 1 through 6 are kind of the big ones for 10K. And then we talked about this a little bit, but I wanna just force rank kind of like the top 3 things I think 10K does for us. He had a different list of what he thinks he does for us the top 3 things well. But yes, number 1, he is still a dashboard, but he is a living, breathing dashboard. So we can talk to 10Kates. So to Jason's point, he can just talk to any of our data from Salesforce, Marketo, Bizibo, and he can ask a question of like, how many VCs are coming to SaaStr? How many CMOs do we have coming to the summit today? What sessions are maybe looking a little bit light? Should we move them around, add a new speaker? And so we can just talk to our data every day. I think an interesting point here is, I don't know if you guys heard the session with Denise yesterday from Snowflake, but it's also on our YouTube channel. She was talking about how she has this bold statement that the dashboard is dead, but really it's because she has dashboards, but she has agents. She just talks to her agents. And for her at scale, having her, she's got multiple versions of 10K. For her, the biggest thing at scale was that she doesn't have to argue anymore between sales and marketing what the number is. The number just is the number. For us, it's still, I think, prevalent across— for your guys' scale to have something that is, yes, a dashboard, but it's one that we all agree on because it's directly connected to the API. And so there's never any nuance or people lying about the numbers or maybe pulling it from the wrong dates or trying to make it look better than it is. The numbers are just the number because it's just pulling it directly from the API. And so I think that was a good insight that she brought up that I hadn't I hadn't really thought about previously. Do you want to talk about the re-engagement campaigns from 10K? I think this is kind of your email, the Replo one.

Well, I think I hit that with the Amjad session. So if you were there, I don't want to be too duplicative. I don't have too much to add. I guess it's just a reminder that Amelia said when she had the lines of code in the commits, it said 6 to 7 commits a day. And anyone that's a developer engineer here might roll their eyes and say you shouldn't be doing 6 to 7 code commits a day. It's like a bad sign. But we're not— we are tech from the perspective of Replit. We're doing 6 to 7 commits from the perspective of us as The boss of the agents, or the agents of the boss of us. We're just improving the application each day, right? Um, and, uh, so I guess just the lesson from this email, when last week 10K started to create better emails than humans, is just how much if you keep engaging with the agents every day, they will get better and better and better. And that's the key, whether it's just to train something like Artisan or Monica or something, or qualified. If you train them, they will get better and better every day. Or if you have agents that are somewhat more alive, like 10K and QBR, if you talk to them more every and every day, their context windows will get richer and richer and they will perform better. And then somehow last week, they will create a better human out of the blue than a human can. And when I asked Amjad, CEO of Replet, how he did it, he didn't quite know. When I asked their head field engineer, he didn't quite know. But that's, that's the key, is the more time today The whole open cloud thing is really dangerous, that you could— not just that it has security issues and breaches and can leak your confidential information, but I think it also confused people that an autonomous agent requires no work. It can go off on its own. The number one lesson we've learned is the more time you invest with the agents, just like most humans, the better they get. And just like most humans, you don't really need to go over the basics with your agents if they're well if they're performing well. If they're not performing well, you have to go over the basics again and again and again. But even your best people, the more— if you can just spend a little bit of time with them each day or each week, they will get better when you start to mind meld with them. And it's literally the exact same with agents. And that's a little bit different than the narrative on people that aren't doing it, right? You think it's the opposite.

Yep. Okay. So QB, we've— I realized we did a workshop a while ago on how we we built QBee. Um, but he is our AI VP of Customer Success, so we can do a fresh one after this where we walk through how we built QBee, kind of what he does now. Because when we did that webinar or podcast a while back, it was like he was— he didn't have all the features and functionality he has now because we, again, we keep shipping, um, lines of code into him. But he, again, he started as, you know, 10K was just a dashboard, QBee was just a project management tool. So we had an antiquated project management tool we were using for customer support, basically, to manage all of our sponsors, to get information from them. And because SaaS and events are kind of niche and weird, no project management tool was great at this. They were all out of the box, kind of just you get what you get. And there wasn't a lot of customizations we can do. And so it required a ton of human follow-up, manual emails, calling people, texting people. It was a lot of manual labor outside of the project management tool that we had in years past. And so we started this as a vibe coding project of, I just wanted to save us some time, maybe get some, you know, it has saved us quite a bit of, of, of budget by building him versus like having humans follow up with these folks all the time. And so he, he really turned into a sponsor, what I call now like self-service agent, where he can manage questions, he can intake, um, logos, he can intake websites, he He can remember all these things about our different customers. And he really has become not only a self-service agent, but also a personalized asset management. So he's collecting all the— it used to be a pain in the butt for us to get everybody's logo so that we can put them on the sponsor wall, put them on all the signs here. It was honestly annoying because then people email you files. Sometimes they're too big. But now you just upload it to our agent. It's much better. But the better part of QBee now is even though he started as just just a dashboard, quickly I realized, okay, this agent's pretty good to the point where he can send each of our, you know, there's almost like 150 sponsors 'cause we have some non, like food sponsors too. And he can email them personalized, right? This is something that no human can do. Like no human CSM wants to have 100 accounts. They all wanna have like 5. And so that was the magic of Qubi was that he knows all of our sponsors cold. He knows all their assets. He knows their logos. He knows what they do. He can, you know, he uses whatever cloud to research about them. Like he knows more about our sponsors than like probably some of your best CSMs know your customers.

Can we, can we try to ask QB some questions we never tried before?

Let's do it.

Let's see what happens.

Okay, QB. Here's QB.

It's QB. Okay, let's ask QB. This is your show, right? Let's ask QB.

Yeah.

What sponsors are most at risk of not renewing?

Okay.

No idea what he's going to say. Never asked. But QB has a lot of data. Let's find out. You probably had something else you were going to ask, right?

No, I know. I think it's a good thing to ask today.

It might fail the test or it might surprise us. Let's find out what QB has to say.

He's thinking. But while he thinks, I'll just show you guys around QB a little bit if you've never seen the front. You can go to SaaSForSponsors.com. You can play around with QBee. He does like you, you'll hit a wall eventually if you're not a sponsor of ours, 'cause you'll have to log in so that you don't get their data or anything weird like that. But you can see like he started as a dashboard, so he's still, you know, he's still our project management tool here. He's got all the information about the events. You know, there's certain sponsor load-in times that they had here. This is where they all picked their booth on the floor that you see here now. These are all the things he did originally, which again, These are all things a human had to do. And he created these nice overlays for our maps, which our designer used to have to do, but now he can work on the stages. And so there's a lot he can do. But now he's— so here's his assessment of—

Who's at risk?

Who's at risk? Artisan, I mean, that's a fair one. He's flagging that they never really logged into him. So he's flagging—

That's a signal. It is a signal. They never talked to me. They had a customer service ticket. Success, they never talked to me.

They never talked to QB, but they did talk to me a lot. So, so he lacked the context, right? Yeah, yeah, yeah.

Okay, flag that one.

Salesforce, interesting. He says they were incomplete and they went dark for a little bit. So, you know, that's a good flag. But again, I talked to the Salesforce team and so does David a lot. Google Cloud, same thing. Like, these are, you know, folks we— it's funny, he flagged all the people that we, the humans, spend more time on than the There's a few others here. Pylon.

Well, Pylon complained the most to him.

Okay, I see.

That's why he said it, 'cause I asked him before, 'cause Quby reached out to all the, instead of a human reaching out to him, Quby reached out on his own, talked to them, asked what their issues were and how he could help. And Pylon said their logo was wrong and a bunch of other things. They complained a lot. So he still gave them a medium. I can't quite read it from here, but a medium sounds right. It's not a high risk.

Right?

We fixed the logo before it went live, but their amount of frustration was a single— was visible, yeah. That a human might not know how to rank properly or hide from.

For sure.

Or ignore, right?

Yeah, for sure.

Yeah. Any other interesting ones that he thought?

Lightfield in Monaco never did their VIP nominations. That's a, I mean, that's a good, that's a fringe thing that we do for folks. We let our, we let our sponsors nominate some of their customers to come to SaaStr. He's saying they never did it, but they are, you know, They're some of our top sponsors. And so maybe it's, it wasn't a high ROI for them.

Okay. Ask QB one last question and move on. You can do it. I know it'll take a second. You don't even need to be in power mode, but keep it in there, right? Who, what sponsors did you talk to that seemed happiest?

Okay.

It's the flip side, right? Let's just see what he says. 'Cause he talked to them.

That seemed, he'll forgive my typos.

This is why I think traditional customer QBS is just dead. They won't talk to 120 people and like all they want is a commission on the renewal. They'll hide issues. They won't surface issues because it doesn't help them, help them hit their number or goal any longer. So I think QBS is better. Who's that? Is this who's the happiest? Is that what he says?

Yeah. So he says there's no explicit this is great signals in the chat yet, but here's who I think have the highest signals of satisfaction. Vidvan, because they wanted to post more on social. Launchpad just returned a lot to the chats. Gamma seems pretty happy with what they're doing. They never pushed back, never escalated. That is a good signal, actually. Then he lists out a lot of sponsors all at once. Asked once, got the answer and left pattern. So that's a good sign too. They got what they needed from QB. And so he's saying that those folks are probably fine because I got them what they needed. And they went on their merry way.

Yeah.

Yeah. So pretty good.

I think this— I know what you guys think. I think this is pretty good. And again, this, this, this agent and agent's kind of a loose definition, but this agent's less than 90 days old, right?

Correct.

Yeah.

And again, it started off as a just dashboard to track assets, almost a pre-AI workflow. There's a lot of asset trackers if you're in the creative industry or do anything. Now we can talk to it and it can identify issues. And I think that analysis was pretty good. We'd never done it before. That was— I just made that up with you. I thought it was pretty good. I would say top 15% of CSMs I've ever worked with already. And you can already see looking at, okay, that's good, but it lacks some of the context of the human conversations that we had. And so we'll probably— Amelia and I will probably ramble, think about that, and we'll talk with Quby later. How could we integrate some of that context?

You just need my email.

Well, for sure. A simple thing is to hook up to your email. Yeah, we could hook up to, sorry, what's the application where we listen on your phone?

Oh, Whisperflow.

Whisperflow, any of these things that have APIs, maybe we just hook them up and actually if they have an API, it's really easy. Yeah, it goes and figures out what the best emails are in Whisperflows in 10 or 15 minutes and maybe we ask again in a month. Maybe we'll do it on the agents on our podcast and we'll see how it compares because we've never done it before, but I would already give that like a B. Yeah, like it missed a whole, it only has the context it it missed the human side, but for the digital stuff, pretty good. It was pretty good in remembering all the chats it had with maybe 500 or 1,000 people, force ranking the issues and identifying them better than a human hiding from issues. So that's QB.

That's QB. He's got an interesting API stack that differs from 10K. So they have a lot of crossover. But noticeably too, he's got, you know, he's got Clerk. Clerk is number 1 on the list because that's what we use for single sign-on. And so that was one I've talked about it previously. It's gotten easier in Replic now that it's native, but it used to be a little hard to actually hook that whole thing up. And I'm not, again, I'm not a technical person. It was pretty hard for me to figure out single sign-on, but I got it done in a weekend and now it's even easier. And so that one has remained top of the list. 'Cause you gotta imagine, you know, people are inviting other people to interact with Qubi and they can just add it on their own. They can catch up, they can see what other folks are doing with Qubi within their org. And so that one makes the top of the list for Qubi.

Yeah, it's a pretty simple stack to get that much value. It's not even logging into Salesforce, is it?

No, it's funny. When I, it's funny, when I went through it, I was like, all this time I really thought QB had Salesforce, which we'll hook up next, so he can write some of this information back to it. But he doesn't need that. That answer he just gave you was without even having any Salesforce data. So it'll get even better. All right. We already talked about this, so I'll move on to the next one. So Annie is somebody you all know. 'cause this is our Saastrannual.com. Did not, again, did not start as an agent, much like QB and 10K. Started as just a replacement 'cause last year's Saastrannual.com was on Squarespace, and we were just limited by how much we could do on Saastrannual.com 'cause we're not developers.

We're just limited to Squarespace. You can't vibe.

You can't vibe.

Yeah. All you can do is put images and some videos. It wasn't, it wasn't enough this year, so we just put— I quickly built a V1 on Replit. It's okay, it has a little— some purple gradients, but, um, but we could get it to do whatever we wanted to do, right?

Correct. Yeah. And so again, started as a website, uh, I don't even remember— oh, I guess it's on the next slide when we did it. But now it does a lot for us, right? So I asked actually our SaaS website, which we just call Annie because annual, um, how it would describe itself as an agent. Because I was like, okay, 10K is very clearly— we call it our AI VP of Marketing. We call QB be our AI VP of Customer Success. I said, Annie, what would you call yourself? Yeah, and it said it's our AI event producer hybrid because it's hybrid of human and what's the hybrid? I think hybrid of producer and also event manager, technical. Yeah, event manager it said it was, and it was also our technical producer because it has all the website and the agenda stuff.

It's always interesting when you, when you ask somebody, including agent, what they want their title to be.

Yeah.

Sometimes I remember asking Amelia years ago what title she wants. She said VP Marketing. I said, okay, that's your title. Came right out of her mouth when she started. Now she's Chief Officer. But Andy came up with quite a title.

It really did.

Quite a title. Yeah.

I was like, you're going to confuse people.

I'm not sure that that spec is on Jobvite, but I like it. Yeah.

I was like, are you trying to say that you did some of the physical stuff? Yeah, that didn't quite work out. But yeah, it was a little funny. But it does, you know, again, it runs. I was like, okay, but to your credit, you are running the website, you're running the agenda. You are sending a lot of, like, the attendee newsletters you guys hopefully all read this morning. That's all coming from Annie.

Yep.

And, and then obviously I— it kind of became agentic when I added this now kind of famous parking passes app, which was previously— if you don't know the story, for you guys to get a parking pass to come to SaaStr, we had to have a human on our team split up like the 5,000-page PDF. So what I did is I first, I just used Cloud Cowork to split up the PDF, add all the addresses to the bottom of the PDFs, and then I hooked it up into Replit, um, in our existing website at the time and said, hey, can you help me create just some agentic workflow so that you can automatically send a parking pass from this long doc of PDFs that's now been separated to people who need it? So when you guys request a parking pass, you just put in if you're an attendee, sponsor, speaker, how many days you need, and then it sends you your parking pass automatically versus you you guys having to— previously it was a form fill. Last year, fill out the form, wait for one of us to send you your parking pass, and that took a while. It's also in the background something else I think it's doing that I don't give it enough credit for is because I've hooked it up to Vector, it can see our active website visitors and run targeted campaigns to folks based on what they're doing on their website. Qualify does a bit of that too, so there's a little— funnel action there, but this is Annie, built on Replit. First commit back in November. That's when we swapped off of Squarespace. He, she, or she, Annie, has the most commits out of all of our agents and the highest amount of commits per day. And I think that's because we were just, you know, Saastrannual.com became kind of a living, breathing agent and kind of copilot for the event as it calls itself an event producer versus it just literally being a website that was static.

Can we do it? Can I ask Andy some questions?

Sure.

You know, it's interesting on the side, it's interesting, these, um, 46,000 lines of code, it doesn't really matter how many lines of code you have. I think again, a lot of engineers and maybe critical of it on X, but I wrote a SaaStr post literally 2 weeks ago about 10K to that. It was $257 a month and I think 18,000 lines of code. Now we're adding apps inside of it, micro apps, and now they're 46, 50,000 lines of code. A lot. There's clearly some slop that has gone in if you've gone from 18,000, 16,000 lines of code to 45,000 a week. But it's also part of adding more and more apps in one scenario. That probably eventually will slightly break, but it is, it is, it's just as an aside, it is a lot of, that's a lot of going from 18 to 45 in 2 weeks is, if it kept going at that logarithmic or exponential rate, we would have a larger code base than like Salesforce or meta. Maybe I'm exaggerating. Um, Annie, well, because I— let's see how well it does.

Okay.

Um, write me a follow-up email to everyone at SaaStr AI 2026 trying to get them to come to 2027.

Okay.

See how it does.

Since today is the last day and you are You are on stage right now.

What did you say?

Oh, I just told Annie, write me a follow-up email for SASRA Annual '26 to come to SASRA Annual '27 since today is the last day and you are on stage right now. Smiley face. I don't know. I like to talk nicely to my agents unless they fuck something up. Okay.

I'm more direct than you.

I know.

Let's see what's in it. Let's see how good it is.

Okay. It said subject line. That's a wrap. It said it wrote it as you.

Okay. Which we didn't ask it to do.

I didn't ask it to do that. It is interesting. Like I did, I ran this experiment the other day with like a final email to our database and I asked Annie to do an email to the database and then I asked 10K to do an email to the database and see which one did better. It's funny, they were actually kind of equal in like their conversion rates, but one of actually 10K wrote it as Jason. And Annie wrote it as like, you know, one of those fully designed HTML emails. But the content was— the Jason one was more like punchier. The Annie one was a little bit funnier. So it's a little funny now that now Annie has swapped to writing it as Jason. So that's a wrap on SaaStr. Lock in 2027 today at the low price we'll ever offer. Preheader: I'm writing this from the stage. 2027 tickets are not live yet, but they will be later today. And they'll never be cheaper than they are right now. I'm writing this from the stage at SaaS or AI Annual, about to do the closing. So forgive the brevity. 10 days, 10,000 AI leaders, the CEOs of Klaviyo, Gamma, Webflow, Rubrik, Leap, all on one campus, the Deploy Summit, 6 AI workshop tracks, the pitch comps, and more first meeting introductions than any week of the year. Thank you for being here. None of this works without you. Now the ask. Come back next year.

It's pretty good.

It's pretty good.

It's not a 10 like that crazy Replit one where he pulled every attendee, but it's like, it's better than 95% of humans would do. And it did it in minutes. And you can see the context it has. It has the agenda built into it, right? So it's able to pull the speakers that we had this year. It knows the numbers of folks that are coming, knows the type of tickets. Someone that joined your team on their marketing team last month would be unable to write this email in a week. They wouldn't know that unless, I mean, unless they could access vibe the data themselves. It's pretty good. I think that passes the test, right?

I mean, you guys are gonna get that email later.

It might be this email. And then Annie, and then I'll share a story. Annie is, so Annie's hooked into Bizibo and Resend.

Anything else?

Did you have a slide?

Yes, so she said, oh no, do I?

I don't think you do. It's not hooked into Salesforce.

It's not hooked into Salesforce. But she is hooked up to Marchetto to send like the attending news reporters. Marchetto too?

Okay.

For now.

So Annie has a different set of contacts. I'll tell you a quick story. Surprised me. I think it's interesting. So this morning we were coming to get here and I said to Amelia, everything's been, I think it's been great. Congratulations Amelia, great job. I think it's an amazing job.

Thank you.

But there's been so much going on. I think, I thought we forgot to remind folks I need folks to come to the Founder VC Brunch at 11 in the VC lounge over there. I said, she's right, you're right, I forgot. I'm like, can I ask, can I ask 10K to do it real quick? 'Cause it's risky. I'm doing it literally in 5, I'm sending an email in 5 minutes to 1,000-something people. Yeah. But she gave me the permission because it's not high risk. What if the email's a little bit off? Like, it's to you guys. It's not the end of the world. But what's interesting, but she says, no, ask Annie 'cause Annie sends better emails like you saw. So I said, okay, go to Annie. Annie, and I said, Annie, please look at everyone coming that's either a VC, a founder, or CEO, and invite them to the brunch. Annie's like, I don't know how to do that. I only see 17 people coming that are VCs or CEOs.

I don't know why.

Annie, Annie, Annie, how could you? You have all the data. How can you think, oh, there's only 17 founders, CEOs, VCs? And she was— I should have screenshotted it. She was very direct that she's unable to do this. Like, she has— she's like, oh, you know what she said? You're gonna have to upload me a spreadsheet for me to be able to do this job. But I'm like, you have access. So it's a reminder that, you know, agents, they get confused. The context is great, but— so Annie had the great enough context to write a really good email with all the speakers and type, but it didn't— it couldn't remember that it had access. It couldn't put together.

It's okay.

So then I go to 10K, who Amelia told me not to use, but 10K has access to even more at Salesforce now.

That's the most.

And it's like, no problem. It did a deep analysis. It went through 10,000 records here in minutes. It pulled up all the founders, VCs, came up with a list, and then it said, hold on, I confused Lightspeed CRM with Lightspeed, a venture firm. Those are not 55 VCs, those are SDRs. I'm removing them from the list. Prepared the list and said, can I send you a sample? Send me a sample. It was pretty darn good. It was about— it was almost as good.

Okay.

And, um, it said, can I send it all to 1,500 people? Then it researched and said, oh, we've never used this version of the API. It send to do a mass send. I researched it, it'll work fine. Can I send it? I said, yeah, send it. I approved it. Great. It was a good meal. But, but Annie used a prohibited email to send it. An email where you're not allowed to use from the sending. It's in the memory. It's in the rules. Broke a rule for multi years. Now, if you hired a marketing manager, they might use an email. They will. They might use it. They might send email, a gung-ho SDR might email people they're not supposed to email. And I will add this later, and I said, "Annie, how could you, this is in the core memory, how could you prohibit it?" She's like, "There's no excuse. I did not, I forgot to read the memory," Annie said. "I forgot to read the memory." She was waking up, yeah. And Annie said it's worse than that, because this is irreversible. When you send an email from the wrong address, I cannot undo it, she said. My job is to call in the architect. And the way Replit works is Replit runs on Claude. But when you have something complicated, it actually brings in OpenAI 5.5. It calls to the architect for a proof. It brings in a different model to look at something, take another look at, see and see if it's an idea. Calls the architect. It said there is no excuse. This is exactly the thing I'm supposed to call the architect to and I didn't. I'm terribly sorry. There's nothing I can do about it. What should I do? You want me to block block anybody that emails you?

Oh no!

She started to scramble for ideas. It's pretty interesting, one, that Annie couldn't do it even though it was, Amelia thought it was better for regular email, couldn't find it. 10K did it, but because of the, and it was partly my fault, because of the pressure of time, I wanted to do this in an Uber coming here too fast, I didn't slow it down a little bit, but when I started on this journey, this would have seemed like a huge huge issue to me. That's unacceptable.

Well, first it wouldn't have gotten done.

Yeah, we got— but when I started on this journey, it would have bothered me because I'm like, how can you send to an email that is prohibited? But you got to step back. A human could have made this mistake too, right? So this is why you're gonna get better and better and better. Ryan, to me, slow it down just a little bit because these agents are so productive. Annie could have sent 1,000— not just 1,000 emails, but 1,000 different emails.

Yeah.

By the time we started this session, yeah, there's no way I could have reviewed and interacted. And so not only do they come to different outcomes, you've got to spend more time with them because they will inadvertently in goal-seeking cut corners. And it just— and it's just— no, I'm sorry, there's no excuse. It's clearly you're not allowed to use this email address and I use it anyway. So a story of the day. But humans do the same thing.

Our agent is forgiven. Yeah. Okay. I'm going to talk a little bit about Amelia AI. And also I think my Pokémon card is Very cute. But, um, all right, so Amelia AI, I launched last summer, post, post last Saastr before this Saastr. Um, I wanted to solve some of the inbound issues we were having on then our Squarespace website, um, which was, you know, we had a lot of folks and engagement on our Saastr, just saastraannual.com, um, but we weren't able to really fully capture a lot of it. I couldn't really do anything with it to book a meeting. It was similar to a parking pass, unfortunately, like you had to talk like tell our team in a contact form what sponsorship you wanted, how much budget you had, and who we should follow up with. And then I would personally round robin it as quickly as I could, which sometimes was not as quickly as I wanted to be. And then one, we had slightly more folks on the sales team, and then I would round robin it to one of our AEs. And then our AEs would take a slight delay, and then they would follow up. So it'd be like a kind of like 2 or 3 day process. If you inbound into our website. So then we, we added Amelia AI, which runs off Qualified from Salesforce now, um, and she just does this automatically. So we, we did spend a lot of time with my FTE at Qualified, um, Cam, um, training her, like going through again the nuances. Okay, because it's for SaaS, we have two different customer buckets, right? There's all of you guys, so all of our, our kind of like self-serve customers, if you will, that can just buy a ticket off the website. But then there's all the small potential sponsors and current sponsors that are also hitting the website, coming to the website looking for information. So we built all that nuanced context back into Qualified when we first launched it. So Qualified, I would say I put the most amount of training into. It does also scrape, um, sash3annual.com in real time, sash3.com in real time. So it's every day it's updating, crawling all of our websites. So that's why Amelia AI is so good. Like, some of it is context I've built into her. So Anytime I push a release to 10K QB or Annie, I also try and make sure I push it to qualified if it's not already there. If it's something on the backend, I try and make sure I put it into her memory so she has it too. And that's why Amelia AI is pretty good because she's got a knowledge, she's got maybe one of the biggest knowledge bases of all our agents.

I didn't know it's the most trained agent.

It's the most trained agent. So that's why it's pretty good. But you can see that like not only she's, she's, she's a pretty good rep, Amelia AI, She's booked a lot of meetings for us. She's had 2.2— 2.2—

slow down for a minute because it's so important. Yeah.

The—

and everyone, everyone should have an agent on your website automating inbound. I'll get to that in a minute. There's no excuse. I'm shocked how many folks I've talked to that are running AI startups that just have a contact me form. Okay, I'm going to come back to that next. But by adding essentially adding AI to an old school Calendly or Chili Piper type link, we booked 614 good meetings out of that.

Just for, just for this annual.

Yeah. So 614 meetings and our average ticket size is about $85K. So that's pretty, pretty good, pretty high ROI to book 614. Now obviously they didn't all close or we'd have $60 million worth of sponsors here instead of $10 million next year. But that's, that's pretty That's pretty efficient agent, right?

Pretty efficient.

Pretty good, right? And almost zero complaints booking it this way, right?

No, no. I mean, the only complaint was like sometimes, like she, so she'll write, right now she'll round robin to her reps or myself, but primarily David. And so it's weighted heavily to book, it'll book a lot, most of the meetings with David, 'cause that's how I programmed her round robining. And then some of the meetings with me, like if I happen to be free, or if it's kind of categories I'm good at, closing deals, which is like, you know, the gammas, the—

It weighs that?

It weighs that, yeah, yeah. It weighs like the— it weighs our Salesforce data of like, okay, what does Amelia close the most of type of sponsors? What does David close the most of? Which David has a big mix 'cause he just has a lot of the customer base. But mine's pretty more nuanced. So it kinda weights that. And then it'll say, okay, for a while it was giving me all the agentic companies 'cause it was like, oh, David has a lot of like old school companies. So it was giving me all the agenda companies. It was like, give Amelia AI, real Amelia, all the agenda companies. I was like, no, no, no, that's gonna be everybody, dude. You gotta stop weighing against that. And so it was pretty funny for a while. So we opened it up, but she'll weigh against that and kind of book our meetings accordingly, which is pretty fun. And then it's funny, I think I got a— now that Slackbot is hooked up, it was like, hey, you should get back in the round because I'm off the round robin for the last like 2 months. It was like, hey, do you want to put yourself back on the round robin and qualify? I was like, maybe in 2 more weeks. But it was, it was pretty fun.

I want to ask you about this. Is there anything worth seeing itself? Do you have a— this is what we're going to look at live or not?

In qualified? Yeah, I can. I can.

But so let me just— sorry, just to go back to Imran. Let me just understand that slide because this is new to me too, guys. 675 torching meetings booked right now. 2.25 million 2.25 million sessions on Annual.com and Saastr.com together, probably?

Just Annual?

Just Annual?

Just Annual?

Just the website for this event had 2.25 million sessions?

I think so. Maybe it's Saastr Annual too, or Saastr.com too.

I don't think so. And 402,000, so interactions is probably individual chats with the agent? Correct. 402,000 chats the agent handled. That's pretty effective. I wanna— let's have Amelia show it if there's anything to see. Just— but the point I wanna make is, to me this seems like the most captain obvious thing to do. I'm shocked that 100% of people don't do it. Replace whatever you have, at least on the part of your websites where you're trying to interact with prospects and leads, and have an agent that qualifies them, that's extremely well trained. It's so much better for the prospect and the customer to talk, okay, should I sponsor SaaStr? Whoever, I'm, uh, I'm, um, I'm Monaco. Is there anybody like me doing AI sales agents here? Yes, there's, there's 10 other folks doing like this, right? How— here's my ICP, here's what I want, here are the agents. It should be able to answer all your questions honestly without drama, give you all the data that you want that is fresh, it's not dated, it's not last year's prospectus or data. And then it can decide with some intelligence who to send that lead to in real time, and they can schedule the meeting in instantly with the agent. Like, it's so— you saw the metrics. We could— we have 2.5, 3 people. We could never hit the number of metrics that you saw, the 2 million sessions, 600-something appointments. We would need like 3 BDRs that would like quit every 3 months. So I know this is like— Qualified is not the only vendor. There's plenty of other vendors. Just buy somebody that does this, train the agent, and I can't imagine you won't have a lift versus a crappy chatbot.

Yeah. Two things I'll just briefly show on Qualified. So this is her Contact Studio just for annual. So if I, if I just show you guys the top toggle here, if you can see it, there's, there's annual, there's London, which it was on, and then it has a slightly different one for annual in person, which is the demo counter over there because we just wanted to compact her memory a little bit for Saastr annual. We were like, she doesn't need to have all of saastra.com in her brain for Saastra annual. So we, we kind of toned her back for the in-person one so that you guys would have better interactions with Amelia AI when you ask her about more like venue-specific things versus things on saastra.com. Com. But this is her brain. You can see she literally was updated this morning automatically. So she's pulling in sessions that we updated on the website so that again, if you ask her a question and I made a change on Annie with our Saturated Web site, Qualified already knows this. So she's gonna give you the right answer if you say, hey, where did this session move to? What's the agenda for today? She can give you that right answer, which is pretty cool. And then another cool thing it does is it does do a lot of email campaigns for us. So you guys probably got some of these email campaigns, but you'll see here she's got two basic, like, I don't know if you call it top of funnel, middle funnel, whatever funnel, but she's got two triggered campaigns here that the agent will do automatically based on what you do with her either on the website, whether you chat with her or not, she has these two top performing trigger campaigns as I call them, so one's for sponsors, so if you visit the sponsor page and you don't quite fill it out, but we know who you are either in Marketo, Credo or Salesforce, the agent will send you a follow-up and say, hey, do you want to book a meeting with David? Here are some similar, um, some similar companies, um, and the, that are current sponsors of yours. Here's the meeting router to book the meeting. It also will exclude anybody who's a current sponsor so that if they were just browsing the website because they're a sponsor, maybe don't send them, hey, do you want to sponsor? Because they're already a sponsor.

So if you don't finish, if you don't finish the flow and ask to talk to a human, but you started The agent will follow up. It'll find similar customers, lookalikes. It will make sure you're not already a customer, and then it will follow up on its own with you.

Right. Yeah.

Yeah. No human can do that.

No. And then similar thing for tickets. Maybe I reversed them there in my mind, but similar thing for tickets. So if you hit the website and we know who you are and you don't buy a ticket, but we know through either like, you know, website tracking, data anonymization, you're already in our sales force, it will email you a VIP code, which a lot of you used to come to SaaStr. So it sold hundreds of thousands of dollars of tickets by either you— if you talk to Amelia AI on the website, she gave you a code, and then if you didn't use it, she followed up with you. And if you're on the website and maybe you didn't interact with her, but it's been a couple days and now you're visiting the website again, she would just send you a VIP code to try and get you to come to SaaStr. So, uh, pretty good.

It doesn't— it's a niche thing. One interesting thing it does that I think is hard for humans is it actually— and again, this wouldn't work with enterprise deals, but it automates discounting.

Correct.

So there are most folks, one way or another, came here with a discount. We would like not to do it, but it's such a piece of human psychology, we can't help it. If you buy a ticket tomorrow, you're gonna get by far the cheapest price ever. Do everybody a favor, buy a ticket You get tomorrow, it'll be half price. So that's an inherent discount. But no matter what the price is, we have so much data over the years says it's better to mark it up 20% and give people a 20% discount. We hate it as founders. We hate it. We have the data to prove you have no choice, right? It's so, so— but what we can do is rather than like humans forget a code or make shit up, if you ask for it, the agent just gives you the right discount and it can change over time going up to that. There's no reason to give people a discount today. The 3 people that will buy a ticket Why don't you take it today? There's no reason, right? And it just— because you can't— we'd all love to not have discounting, but to truly guardrail it so it's automated, it's kind of a gift because it removes friction from the sales process.

And it's—

again, I don't think humans are able to do this. Like almost all, especially sales reps, when they have no ability to close a deal, they'll instantly discount again. When they smell— when a human smells a deal going away and they already gave you 20% off. The next answer is, well, will you buy my product for 25% off? Are you listening? You don't have a core integration I need. How about 30% off? I think I'm going to buy our competitor. How about 34% off? And it doesn't move the needle. It creates a lot of issues. So you can create guardrails here. I didn't get this for a while till I watched what Amelia AI was doing. It's super efficient because it removes all the drama from discounting, right? It's almost like a pseudo CBO. CPQ system, but it does it in real time because you have rules, right?

Yep. Just a couple of things I wanna touch on, 'cause I don't wanna get Adam up here so he can do his demo.

Yeah, we're running outta time.

Okay, I'll talk a little bit about AgentForce. I showed it a little bit with Adam Alfano, who was here yesterday, who's the president of SMB. This is actually the 1.0 AgentForce Builder. There's a better one now that we're on, 'cause Lindsay just upgraded me, but I think the goal here of AgentForce Builder force. Um, what I've learned this week actually was I think a lot of folks are like, okay, it's Salesforce, so it's gonna be really hard to deploy an agent. Like, all the other things I had to put into Salesforce were kind of hard. So like, how easy can it be to get my agent force up and running? And there's kind of a nuanced quick answer I'll give you, which is one, how we use our agent force, which is for ghosted leads, is a very bounded use case right now. We will use it for other things, but right now, King Boo's really good at AgentForce, of reviving leads we never followed up with. Some of them are still older leads, some of them are still newer leads where he's doing more of a re-engagement campaign of leads who maybe said no to me and David, and he's going back out to them later to be like, hey, do you wanna just come to Saastr, see what it's like for next year? And so again, he's, his primary goal is to follow up with folks that didn't give us a yes. But I really like AgentForce because to that point now, one, it's gotten a lot better since we launched it back in October. There's a 2.0 builder. It has the highest open rate. And why is that? Just quickly, I'll touch on, is because it has all of our Salesforce data. And so it's got a lot of— well, now also it has all of our qualified data since Salesforce Bot qualified and Momentum, which we were using as separate agents. Now it's got all of that in it. And so it's got the most context automatically. Everything you guys just saw—

I kept asking Amelia, is this one integrated Salesforce. Does this— remember the agent force? The advantage is if you run on Salesforce, it'll be true when HubSpot finally launches agents, but it has every— it has it all.

Yeah.

So if you can do it in agent force, maybe do it in agent force, right? If you can, because it's gonna— if you're on Salesforce, because it just has the maximum context.

It really is the path of least resistance, and I know the team over there at the booth will help you get set up today if you want them to. So it really— again, it just it just has a lot of context. You don't have to move things from one place to other. If you were already using things like Qualified or Momentum, it just got even better, honestly, kind of overnight. And so it was a gift that they bought those agents we were using.

So are we out of time on this one?

I just want to hit Artisan and Monica real quick. Okay. Artisan. How do we use Artisan? So Ava, who you guys have seen in the booth too, we use her for slightly warm outbound. And so a lot of campaigns that are not in qualified, not in agent force. We listen, we have a big go-to-market motion here. I'm realizing we have a lot of different agents that do a lot of different nuanced things. We use Ava for what I call slightly warm outbound. So these are folks where, you know, we have past, past sponsors, past customers, past attendees, if you will, between attendees and sponsors. And I would go into things like, okay, if their email is still valid, cool, we can load it into Ava. She can interact with them. If they've moved on, she can help find like the right contact information, whatever. And she can also build lookalikes, which it's pretty good at. And then she can help us, you know, I use this primarily for us on the ticket side. So if you have something that's a lower ASP, you can use something like an Artisan to do kind of, as I call it, lukewarm outbound as a sales agent. 'Cause she, again, she's also hooked up to the website, so she knows everything we've done. I get really segmented with Ava, so I give her super specific campaigns like, here are alumni of Saastr Annual 2024. Here's what was different about 2024 specifically versus now 2026, so that when you email these people, you have that specific context to use in your emails when you're doing their follow-ups.

Yep. And one insight I had to synthesize all of this, and now we're just about out of time.

Yeah.

So I'll go very quick. But I was having a conversation with an AI CEO last night that that their own product's kind of at the bleeding edge. And, and also as head of marketing, and synthesize everything. They're like, oh, this doesn't work, we haven't gotten this to work. And I went through one simple heuristic with her, with both of them, the head of marketing and the CEO. For your— another way to think about this, you have your A leads. Your A leads are so hot a human's gonna follow up within 60 seconds. So if someone comes in and says, I want to buy a million-dollar sponsorship or half million, even your laziest rep is out there at the at the mall or watching movies is going to like respond to that email in 60 seconds. Okay, I got a million dollar budget, I'd like to sign today. Don't, don't put an AI on those accounts. Put it on the B leads. The B leads are the ones that have, that have some signal, some score, but they're not worth a human's time in your organization. Every organization of size, humans will just not follow up with the B leads. The C or D leads may or may not have gold in them. It's a topic we'll run out of time on. but put something like Artisan on the ones already in your database that you have contacts with that have some score, some signal, but a human— it's not worth a human's time. That's $500K of gold for us. Okay, this is not a massive business or even a business, but we need the $500,000 so you can, you can eat. Okay, without $500,000, be bring your own sandwich. Okay, so the Artisan got us an extra $500,000 to the event. That, that's the B leads. And so I want that Let me do one more on the break. But if you think about it that way and you train it, you realize it's going to work because you have B leads in your database.

Yeah. And I have given artists a lot of content. We can talk about it more later. I can, I can maybe squeeze some. I can help answer questions during the AMA later. Okay, Monica, this is our last one, our newest one. So Monica, who's also here, references— I kind of did a little copy, copy paste error there. But Monica is the the one that you guys can see here. This one I put on cold outbound. So even though Artisan can do cold outbound, the reason I like Monica for it is, I mean, technically we're not, we're not their ICP because we have this giant agent stack. Sam will be the first person to tell you this, but I like it because I basically gave it the best sponsors across all the years and it said, Amelia, I got it. So it took all of our ICP data, all of our Close One history. We did have to export started from Salesforce, which is a little nuanced. But, and then Monica just automatically found lookalikes, booked a bunch of meetings, closed some pretty big logos still for SaaS in a very short amount of time. So this is the one that I think idles the least amount of time out of all of our agents because it is self-filling. It'll just keep going out to different ICPs. It'll keep filling your funnel. And so I like Monica for that reason because I really don't have to think about it, which is quite nice. So that's why we added to our stack. So though it's not, I know we're not their ICP and it's not perfect, but I really like it.

Yeah, it's a, we're running out of time and we'll do another follow-up deep dive, but the, the Monaco one is cool because it fills its own funnel.

Yeah.

If you can get it to work, if you can train it, if they think you're the ICP, it is the only tool we right now that, that constantly fills it with lookalikes. It's just doing a clever use of lookalikes. And as you dig into these things, pretty good. Just like B leads are good for SDRs, some of the, some of the art underneath is just getting good at lookalikes. Likes. Like if your sponsor is, is, is Oracle, is Lightfield like them? If your sponsor is Salesforce, why isn't HubSpot here? They should be here. But it's not that hard to figure out that since everybody but HubSpot is here, they should be here, right? And so maybe the team followed up with the wrong person at HubSpot. So Monaco will say, we'll go out and figure out, okay, who is the right person at HubSpot to talk to? And, and what it did a real instantly good job of— then we got a break. That deal may or may not close, but it instantly identified a really good buyer and gets you a meeting. Okay, so this whole lookalike thing, it's actually not that. A lot of this stuff, if you, you're gonna realize in a year, is not that complicated. Okay, it's just clever. And what you saw today when you looked at the backends of a lot, you saw forms that look like other forms for web. Like it's the, there's some clever wrappers and guardrails and things, but you can do all of this yourself. You can do it all yourself. So thanks everybody. We got a great session with Adam from Ownr on this stage, right? If you want to learn how to have an AI CMO get you a billion dollars of customer revenue, I would say for this one.

All right.

Thanks, everybody.

Please welcome the CEO and founder of Owner.com, Adam Guild.

Let me take you back to 3 years ago because the past 3 years has been A huge roller coaster at Owner. 3 years ago, Owner was crushing it. We were building this Shopify for restaurants thing that was working. It was a website builder, online ordering system, marketing automation system for restaurant owners. And from a growth perspective, it was going great. We were exceeding the triple, triple, double, double, double trajectory that the SaaS lore says eventually leads to building a unicorn. We'd raised more than $89 million in capital from all of these incredible investors. Things seemed great, but beneath the surface we were facing an existential risk starting in early 2023. That risk was the ChatGPT moment that was cascading throughout the tech industry and putting huge pressure on us in two different ways, each equally terrifying. The first way was all of the AI-native startups that are being born in early 2023, doing things that previously weren't possible, like vibe coding website generation and AI phone answering and ordering for restaurants. And the risk that they posed was that they were growing like wildfire and directly could disrupt our product. The other risk was coming from maybe even more terrifying direction because all of the massive companies in our space, publicly traded companies, had taken notice of our momentum and were as aggressively as they could trying to stop it, throwing hundreds of engineers on trying to clone our product and then distributing it into their customer bases of hundreds of thousands of similar customers. It became obvious that unless we adapted in this moment and took advantage of it, that we were going to get crushed from both of these directions, that we were going to slowly bleed out and die, and that we wouldn't have a business in a few years. The hard part wasn't realizing that. The hard part was all of the smart people around the table that disagreed. Major investors saying, "Adam, this is CEO thrash. We can't just invest millions of dollars in this new product direction and take people off of this thing that is working. It's working well. It's growing efficiently. We can gradually layer these in over time." Or, Team members, product managers, and engineers saying, Adam, I know 100 customers personally that are asking for the product to do this thing. And if we're going to take people off of that to build this new thing that hasn't even worked or been proven, it's just going with the AI fad. How could we possibly do that? Or the industry experts, the folks that had spent their lives and careers in our industry of restaurants. That were saying, Adam, if there's one thing I know, it's restaurant owners, and restaurant owners are terrified of AI. If you start to pivot your company in that direction, you're going to alienate your customer base. It was brutally hard, but we pushed through all of those reasons not to. And if we hadn't, then this wouldn't have happened. This glorious tweet from the godfather of SaaS and the king of SaaS AI, Jason M. Lemkin. Jason M. Lemkin, one of our incredible early investors and board members, on the first of this year, talking about how we grew faster in 2025 than we did in 2024, and how more than 83% of our new customers, up from 0% just 2 years prior, started their journey by using our AI products. Talking about the caliber of talent and momentum that we've been able to accrue that we wouldn't have otherwise. So that is what the next 30 minutes is going to be about. About all of the ways in which we had to reimagine our product and our business to adapt to be AI native in the way that we build product, the way that we operate, and the way that our whole customer journey works. I'll also touch on psychologically all of the challenges that came from making these sweeping changes across the business. And to make it extra helpful, I'm going to be demoing for you the 5 most impactful things that we built. Both the product and how that changed and all of the internal systems that do things like automate the busywork for all of the builders in engineering, product, and design. What it looks like to have an AI-native finance function or sales function and the tactical things that I learned from the psychologically hard challenge of convincing a team to move in a completely different and unproven direction. So let's dive in. I'll take you back to the moment in early 2023 that we got up the courage to go all in on this direction and push through the industry experts saying it was a bad idea. That moment was at the International Pizza Expo there, where I'd gone in very excited, showed up early to do my usual thing, showing these restaurant owners the Shopify for restaurants that we'd built, the website and online ordering product. And I show up that day amped, very early, get my laptop set up. Right before the showroom doors open, pizzeria owners start coming in. And as that happened, and pizzeria owners start walking in to the showroom floor, the first few passed me by, and then one, he cranes his head, he looks at something behind me, and he walks up to the back of our booth and pulls out his phone. I'd never seen this happen before. So I walk up and introduce myself, "Hey, I'm Adam, how can I help you?" And he was scanning something that one of our PMs brought to the show as an afterthought. It was this poster that was describing one of our upcoming products, which at that point was a terrible MVP, very embarrassing. But it basically said, "Analyze everything that is broken about your restaurant online and fix it with AI," with a QR code to scan so that they could try using this experience. And this guy, Joe, was fascinated by this. He was so excited by the idea of using AI in his pizza business pizzeria. This 55-year-old pizzeria owner from Pennsylvania was really excited by AI. And what was so interesting about this to me is it was the exact opposite of what I'd heard from the industry experts and even what I'd heard in customer discovery interviews 3 months prior. What I realized in that moment is that the ChatGPT moment hadn't just taken the tech industry by storm, it had also been used by all of these small business owners. They'd read about it and they'd gotten excited by the idea of using AI as an advantage in their businesses. Over the course of that day and talking to all of these pizzeria owners, it was the number one most common thing that they wanted to know about, even though it had nothing to do with the vast majority of our booth or collateral, what we were there to demo. I was getting questions like, how can we use AI to drive new customer discovery? How can we use it to drop our labor costs and become more efficient? And it was that day that we got the courage to go all in on this direction. And actually neglect some burning fires in other parts of the product in order to do it. So to give you a demo of what specifically changed, in early 2023, we were 100% sales-led growth. It was an inbound motion. People would book a demo and then have to talk to a salesperson, an onboarding specialist in order to buy the product and then get started with it. And we reimagined that whole journey in an AI-first way, asking ourselves, how much easier and better could this become for the restaurant owner now that we have generative AI? What it led to is this experience that you might have seen went super viral on X about 2 weeks ago. We got more than 2 million views on the tech community announcement of Grader. But what wasn't on X is that in the restaurant world, this has been growing like wildfire for the past 2.5 years because the way that it work— oh, not hardwired into the Wi-Fi. Classic. Fortunately, my hotspot might— The law of live demos always strikes when you need it to not to most. I don't actually have my phone on me, so I hope that the hotspot from backstage activates.

Woo!

How about now? I've got another option.

That's not gonna go.

Do you have any other one?

No, no, I mean, do you have any other on the live?

The live demo gods are taking their vengeance on us for not praying first. The live demo gods always strike when it is least convenient, of course. Okay, refreshing.

I'll stay here until you get it.

Okay, thank you. It seems like the internet is not yet connected. Should I go into Wi-Fi settings? I've got another one just in case.

I'm gonna pop—

I'm gonna switch over.

Ready? I have another.

Okay, refreshing. I appreciate that very much. We could try. Might as well.

It's not the dongle.

It's not the Wi-Fi.

It's losing the internet at the base.

Yeah.

Maybe we use my own hotspot because I think— yeah. If you could bring up my phone. Thank you so much, sir. Very kind of you to offer. Our VP of security has things locked down like Fort Knox, so I worry if I'm on a foreign—

Yeah, it's fine.

OK.

Okay, let's see here. Okay, it worked. It's a much more impressive demo when you have internet, I promise. Okay, so way the whole thing works is instead of having to go through a salesperson and an onboarding person to build out this new website and online ordering system for the restaurant, we asked ourselves, can we use AI to automate all of this and just have the restaurant owner type in the name of their restaurant and then have the AI figure out everything that is screwed up about it online and then fix it for them as the restaurant owner. So the way it works is you type in the name of your restaurant, which in this case I'm typing in a restaurant that I saw last week, really needs a new website. And within seconds, the AI jumps in and starts crawling every place it can find that restaurant on the open web, pulling in all of their nearby competitors so it can show them how they compare and contrast to their competitive landscape, going into their Google Business Profile and scanning to make sure they're following all the right settings, just descriptions and keywords that drive Google discovery, reading all of that restaurant owner's reviews so it can figure out what customers love about that restaurant and rebuild the experience with that in mind, doing a photo quality analysis, an AI photo shoot where it's not only ingesting all of the photos that it can find, but then upscaling the images, increasing the lighting conditions, the color, making it look much more professional. So that it goes from this website before, which is hilariously barren, to say the least. It's a very confident restaurant because the photo is just a picture of napkins on the website that says, "Welcome to Thai Thale." Uh, so this is the before, to give you some color. Here were all the things that were wrong with the report before, uh, where images were missing alt tags on the website, bunch of other SEO things were screwed up. It's checking like 90 different factors in SEO and CRO. And then in a matter of less than 5 minutes, it went from that before to this after. Which initially blows the restaurant owner's mind because that video they've never seen before. What we've done in that 5-minute period is we not only did an AI photoshoot for the restaurant of upscaling all of their images and making them look much more professional, but then we added motion with VO3 and created this compelling initial video. In addition, we've built out all of that restaurant's content to perfectly reflect what people on Reddit, Instagram, Facebook were saying about that restaurant. And make sure that it represents the value proposition well. It's got spotlights around the most popular dishes. The Thai sampler is one of them. Photo gallery, details on the cocktails, sake, and the bar, so that in less than 5 minutes and for free, a restaurant owner can have— go from a really bad initial website that is actively hurting their sales and creating a bad initial first impression to a state-of-the-art amazing website And it isn't just a flashy demo. This has had a profound impact on our business. It is why more than 83% of our new customers are starting their journey by using this product. It is why we grew faster in 2025 than we did in 2024, and a huge part of why we are now accelerating as we're about to cross $100 million in ARR. Thank you for the person that clapped. So what are the most important lessons learned on product from going through this experience of building a product to disrupt our prior thing and reinventing the customer journey? Lesson 1 is that in order to benefit from agents, you've gotta build an extremely opinionated product because the thing that makes agents magical from the perspective of a user is this idea that it is the first time that software is able to drive the outcome that that user wants without them having to manually intervene and try and figure out how best to configure the software in order to get the outcome. The time savings and fewer expertise-requiring insights that are required to engage with the product. That is the thing that makes that Thai Thale website so powerful from the restaurant owner's perspective. It's not just that it looks prettier, it's that when they activate it, it quickly grows their sales. By increasing their conversion rate and SEO discovery and having the best practices that we've learned from powering thousands of restaurant websites and the experiences of tens of millions of consumers that are using those websites. Interestingly, this also connects to the second major takeaway on product, which is that in order to insulate yourself from the dreaded getting screwed over by Claude effect that is terrifying so many startups right now, That opinionation ends up providing that insulation. What I mean by that is if we were to go in today to Cloud Code or Codex and say, build me a beautiful new website for Thai Thale, it would already build a much better website than what they had before. And yet the website that it would build is based on the corpus of all of the restaurant websites on the internet and all of the other UI best practices, but it is missing the insights around exactly what converts components on that restaurant's website are correlated with sales growth. Exactly what it is about the restaurant's call to action or placement of the navbar or the rest of the homepage content that causes it to show up on the top of Google. And that means that through having this very opinionated product approach where we enforce that all of the restaurants are following the same system and same best practices, that we've got this unique data advantage in being able to consistently drive that outcome for users and this moat that ChatGPT or Claude can't exactly touch in order to deliver the outcome for our users. Which brings me to lesson number 3. In the old world of SaaS, the way we thought about high-quality products was we were measuring things like daily active, weekly active, monthly active users in the dashboard. That was a sign that our user base was engaged and that the product was working. And now I think that closer to the opposite is the signal of the product really working. Meaning that if agents are powerful because they drive the outcome for the user without their constant manual intervention, then every time that somebody's logging into our website builder to manually intervene with the way that it is setting up their restaurant online, it means that we have failed. It means that we haven't created an experience where the software drives the outcome because the user is the one that is trying to solve the problems that were created by the software, and that instead what we should aim for is an experience that is as low-touch for users as possible, where they don't need to log into the dashboard and constantly intervene with what is happening with their business, but instead the software just knows what the best practices are and is constantly executing them on their behalf. Okay. Now that we've covered the reimagining of our product, it's also worth covering the way that our product development and engineering process changed as a result of this product change. This one actually starts with Will, who is a genius builder, a one-in-a-million unicorn who's an excellent product mind, design mind, and engineer in one person, and the first person that we had on building Greater. As Greater took off, Will found himself suffering. Suffering because the thing that energizes Will most in the world is building awesome products, being deep in the codebase, and continuously making things better. But as the thing took off, we started putting more and more engineers and product folks and design folks in that part of the company, and Will found himself having to incur all of these coordination and communication costs in managing a bigger team, doing things like keeping Linear up to date and Notion and daily standups, and all these things that we've just kind of accepted as part of the product development lifecycle to keep the team aligned, but that are really a huge cost on the very most leveraged ICs in the company. No great builder wants to do the builder busywork of staying aligned with the team and showing up to the standup meetings. They all just want to build. So he asked himself, could we apply the same line of thinking that just led to this product breakthrough in the way that we actually develop product? And so Owen was born. Owen is our internal agent that does 90% of builder coordination work. Owen is constantly listening to GitHub, Slack, Notion, Linear, Google Meets, grabbing the transcripts from Gemini to understand the work that is already being done by builders and keep the team aligned automagically so that Will no longer has to show up to the meetings and ask people for status updates and Slack them back and forth to understand the status of any given project in leading this part of our platform, but instead just has all of that intelligence in real time at his fingertips. Here's what that looks like. Will can generate a report from Owen in any given week that shows him exactly what got done on any part of the team. It shows the demo links of things that people built, what the updates are on the project that is in action, the most relevant dashboard that are relevant to the specific metric any given product initiative was trying to move, and the best comments of the week for comedic value. Then you can go into the Today tab and see in real time exactly what each member of the team is working on. This is a live demo. We can see here that Chaz is working on reviewing website marketing copy and taglines. He's one of our designers. That Oscar is engaged with the Project Launch project. That Matt— is resolving a POS call scheduling conflict. This is real-time data so that he no longer has to do stand-ups or Slack people or tap them on the shoulder to try and figure out how things are progressing, but instead just has a real-time understanding of that and can click into any given person or any given project to see what the real-time status is. This is saving us a huge amount of time. And another awesome thing that it can do is automate all of the really boring front-end work. What I mean by that is if you look at that screenshot, Owen is constantly listening to Slack, where when somebody submits a bug or some visual thing that seems off, like in that case, he's showing a screenshot of our Agentic chat and saying that the bullets look too small. Previously, that would require somebody creating a ticket in Linear and then a frontend engineer taking up that ticket, finding the part of the codebase that controls the size of the bullets, and updating it. Across a massive product area, there's all sorts of things that come up constantly, and it is taking the time of some of the most leveraged people in the company. So he asked himself, could Owen actually do this? Could Owen call Claude Code, which has full visibility into our codebase, and just immediately when a Slack message is posted from a product person, a design person, a support person reporting a bug, jump in and do the first draft PR? So now Owen does. You can see that he immediately responded and linked the PR that he wrote, which immediately fixes the a visual issue. This saves a tremendous amount of time across engineers and keeps them focused on the highest leverage engineering work versus the very boring and monotonous front-end fixes that come up. Another breakthrough on this front of leveraging our builders came from our brilliant co-founder and CTO, Dean. Dean found himself with a similar version of the problem that Will was facing. Dean is also an amazing builder and hates unnecessary meetings and stakeholder alignment stuff and trying to distill context from big groups of people. A lot of you probably relate to that. So, what Dean asked himself is, is there a way to get the signal that is currently costing me hours every month where I have to interview our customer support folks and our sales folks and our customer success folks and read all of the different tickets of things that have been submitted by customer-facing teams and talk to customers customers in order to understand in this lossy way what it is that would be most impactful to build in any given sprint. And that is what led to the Product Insight Command Center. The Product Insight Command Center is constantly pulling insights from every place that our customer-facing interaction occurs— Salesforce, Intercom, Momentum, Talkdesk, all sorts of others— and then grabbing those transcripts. And anytime it notices a prospect asking for a feature that we don't have, or somebody reaching out for customer support as a result of a bug or something that is missing, it develops this dashboard that gives Dean immediate and real-time visibility on where the biggest opportunities are to take friction out of our customer journey or to improve the product in order to drive our revenue. Here's what that looks like. This is specifically the support intelligence tab. Okay, good. I'm glad we didn't get locked out of our system. I was worried that our new security thing was going to cause that, but it worked. You can see in real time in any given time period what the support cases were that were submitted by customers, and then click into any given category, like in this case wanting to understand what the 83 tickets were related to delivery issues, and get all of the real examples with real quotes from customers that unpack the problem so that it saves him from having to do all of these meetings that are an imperfect way to get the right answer of talking to support and sales and customers. Uh, he still talks to customers a lot for a different reason, but it enables him to have that immediate clarity on where the biggest opportunities are to improve our product and improve our overall business by adding the things that people are requesting. He can also click into any of these cases and in real time it gives him a detailed understanding of why it got logged the way that it did. and this visualization of in any given week what the primary reason for reaching out to support is. It does the same for revenue, but I'm not gonna show that because it's gonna leak way too much sensitive stuff. But that's what the market signals tab does. It's like all of the different sales conversations that have occurred and the reasons why deals were won or lost based on how frequently the prospect cited any given feature for their decision. In addition to, having our more technical teams, traditionally engineering, product, and design operate as AI native. We've also started to operate as AI native in finance and sales. And this next example is one of my personal favorites because it saves me a tremendous amount of time and occasionally embarrassing moments. It came from our brilliant CFO, Will, and Mira on the finance team, who about a year and a half ago started operating in cloud instead of Excel as our primary financial artifact. And what that enables is the situation that many of you are probably very familiar with, where you're catching up with an investor and they ask an extremely detailed question about a metric and exactly what ladders up to it. And it's not something that you would just have at the top of your head about how the Q3 rule of 40 compares to the Q4 rule of 40, or the different ways in which CAC is changing over time. It requires you saying, great question, I'm gonna check with you and get back to you, versus being able to just spit it off on the top of your head. So what Will built by having our financial model and all of our most important financial artifacts in Claude is the ability to just query our financial model with the types of questions that you would get. As an example for this, I'm gonna click in and give it this prompt. Oh mighty Claude, I'm currently on stage screen sharing you to an audience of thousands of people but want to show off your power and what kind of growth we can expect in May. Without sharing anything beyond what I asked for, because I will kill you if you leak our financial model right now, can you please share how much net new ARR we added in May and the rule of 40 and burn multiple of that month last year and the seasonal GPV lift? Also help me visualize it with a graph. This is the most simple query that you can give it, but it's going to spit out the right answers in all all of 10 seconds, answer exactly what the metrics were that I asked for in May of last year, and create this visualization that you can imagine how powerful it is when you're asking more complex questions so that you can immediately give that insight to your investors, to your team. Or I use this when we're going through FP&A discussions and budgeting discussions because I can immediately have the answers at my fingertips versus having to do a bunch of back and forth talks with the finance team or to pour through the Excel model myself, and it saves at least 30 minutes per occurrence. You can see in addition to spitting out the metrics, gave us the exact answer that we asked for, which is the seasonal GPV lift that May payments volume sees versus April payments volume historically. That's AI-native finance. Another game changer is making our sales team and go-to-market team AI-native. Few epic things to demo on that front. Credit for these go to our awesome CRO Kyle, who you might have seen yesterday, our go-to-market AI lead YG, and our awesome VP of Biz Ops Jonathan Shankman. Also, YG is also awesome, and I don't know why I just used that descriptor 3 times. Kind of awkward. But what they built is not. What they built is amazing. They identified a problem that you're all very intimately familiar with, I'm sure. Reps are losing a huge amount of time every day. In manually doing research before every demo in order to show up understanding the prospect's needs and the current situation and not waste a bunch of time with discovery questions so that you can get right into the meat of the benefits of the product. In addition to saving a bunch of rep time, the other thing that they were solving for is our buyer restaurant owners are extremely busy and overwhelmed and skeptical. So if you don't show up extremely prepared, there's a good chance you're getting hung up on or not taken seriously, which is why our reps were spending so much time on this. 20 to 30 minutes before every demo. In addition, it was hard to quantify and qualify restaurant leads in real time if they wanted a demo within a couple hours. So what they built is the following: AI PCR, pre-call research, and eGPV. The way it works is the moment a lead is submitted in our system, booked through our website, or added in by a sales rep, deploys a bunch of agents to research everything that that rep would do manually. Not only running the greater report that we saw for Thai Thalee detailing all of the different issues with their existing online presence to drive problem awareness, but also pulling in things like who is the nearest most successful customer to this restaurant that can be shared as social proof, or EGPV, estimated gross payments volume. Within about $250 now, we're able to triangulate around how much in gross payments volume any restaurant is going to drive on our platform. Which is a major driver in being able to prioritize the highest value leads and automate qualification so that it's not just up to the BDR going through a detailed qualification question. It saves both the rep and the prospect a huge amount of time. And over the past 3 years or so of implementing and continually fine-tuning this system, we've seen a more than 90% increase in call volume and rep time with customers. And they're basically out of having to do all of this less extremely monotonous research work ahead of each opportunity. We've also seen close rate increase by a really substantial amount. So that solves the practical problems of reps, saving them a bunch of time before every demo. It's also worth talking about one of the game changers in solving the emotional problems of sales reps. And that emotional problem that I'm referring to is intimately familiar to any sales leader or founder in this audience. Which is that it is brutally, brutally hard to be a salesperson. Even when you have a great product, you're just getting rejected constantly and reaching out to people who don't respond. And so the insight from our marketing folks, Hannah and David, each superstars, was if we could automate any time a customer says something positive about our product and share that with our sales team and our broader team, it's going to increase the amount of confidence and conviction that a salesperson is able to show up with because they're constantly being fed examples of the positive impact that their product is making. So the way that it works is we have Momentum hooking into a bunch of different places where we have all of our customer-facing interactions in the post-sales part of the journey, support and success. And anytime a customer raves, which is multiple times a day, the exact quote from the customer is shared with the team. The salesperson that signed that customer up can see, wow, this customer, Majid, said that he's driven more than $150,000 in additional sales and he's never had issues. Like, they're getting this constant form of reinforcement of the belief that they can have in the product, which makes it emotionally that much easier as a salesperson to push through all of the brutally hard parts of sales, of getting rejected constantly, because you're constantly getting reminded of the impact of what you're doing. What you're selling and how it's helping customers. So that's a bunch of the most tactical stuff that we built internally and how we reinvented our product. It's also worth talking about, reflecting back on the past 3 years, what I think was most important for driving all of these changes and this adaptation as CEO. Managing the psychology of it all. First major takeaway. Is that leaders must lead by example through these changes, especially the CEO. A failure case that I see in friends and peers is saying scary shit to their team, like, if you don't 10x your productivity with AI, you're out of here. Or similarly extreme statements, but then not actually doing much with it themselves. And I think that the most effective way to get the team to operate as AI native is is to personally and constantly operate that way and demonstrate to the team what it looks like and how much more value we can add for customers. I am not a super technical CEO. I was once a mediocre at best script kid engineer building Minecraft servers, but I've not written production code in owner history until recently where I heard a problem from a customer This has happened a bunch of times over the past 2 years, but I'll show you a recent example. And then built something over 6 hours on a Saturday to solve that problem that today is used by hundreds of customers. The specific problem that I heard came from Juliana Vasquez, owner of Somos Oaxaca, where she described this very frustrating situation where she spent $2,000 a few months ago on a commercial photo shoot to take photos of all of her menu, and it took half a day of her time, a bunch of money. $2,000 is a lot as a restaurant owner. And then she added new items to her menu for spring. And she said, "Adam, I feel so embarrassed because I can't afford to get that photographer back here. It's $2,000. And when I upload these new menu items to the Owner system with my iPhone photos, they look so unprofessional and bad. And I almost don't even want to add them to my menu at all. Is this something Owner can help with?" And I said, "Hold that thought, Juliana. I'll get back to you this weekend." Conversation happened on a Friday. By Saturday afternoon, I had Owner photographer. Built out. Own a Photographer enables Juliana and other restaurant owners to upload the image that they want fixed. In this case, it's this blurry iPhone image. This is her real food photo that she wanted help with of tacos. You can see that, like, the background is kind of blurred out, which blurs out the beans. Doesn't look great compared to all of her other photography. So she uploads that, she selects a style. In this case, we're going to go with the style that that commercial photographer produced. Much nicer plating, better background, higher resolution. And in a matter of less than 30 seconds after clicking generate photo, a bunch of models get called under the hood that describe each photo in very precise detail, feed it into Nano Banana with a set of anti-prompts to make sure that the food doesn't look uncanny valley and fake, which is a big problem. So that she goes from having a photo that initially was taken taken on an iPhone, not a great impression, to a photo that matches the exact style that previously cost her $2,000 and has a much better rate of being added to cart and represents her brand better. Okay. Hopefully Nano Banana is not compute constrained in the way that Claude has been lately and this genuinely takes 20 seconds. It did in all of the practice demos. So let us pray to the live demo gods again. Don't know how being on a hotspot might slow this down, but we will see in a moment. It's a pretty jaw-dropping demo, so I hope it works. The point is, uh, instead of just saying to the team, ship more stuff for customers and 10x yourself with AI, I, I have to personally be doing that too, uh, in order to make— oh no. Okay, well, it's a good thing I recorded a backup Loom, knowing the law of live demos, of me doing this last night. Uh, so I'll just take you to the punchline, went from a photo that looked like that before to a photo that looked like that after, preserving the exact realistic qualities of the before, but also making it perfectly match the style that cost her $2,000 to produce. A lot of leaders are saying things like, adapt or die. If you don't 10x yourself, you're out of here. This type of thing. And I think you have to embrace that through your actions, through the way that you show up, through the way that you personally build. This is one of 5 things that I personally built in the past 2 months. And lead with that sense of urgency. Don't just say it, but lead with it. And the scary stuff tends not to work. Also, another wrong conclusion that I see is this idea that now that people are so much more leveraged and impactful and efficient, asking their team and themselves the question of how many fewer people do we need now? To execute the original vision. And I think that is the exact wrong conclusion to come to. That the right conclusion is asking ourselves, how much more could we build now? How many more of our customer needs can we meet? And how many more people can help contribute to accelerating what previously would have taken 10 years to build over the next year or two? That more high-agency people, more builders, is actually the right answer to come to. And that Jevons paradox, this idea that as a resource becomes more efficient, it becomes rational to use more of that resource, ends up being very true with high agency people. Because AI unlocks the golden era for high agency people. This is the most awesome time to push harder and expand ambition, not to shrink and ask ourselves, how do we do the original plan with fewer people? I think that's a really depressing and wrong conclusion to come to that doesn't take advantage of this opportunity. That bet that we made 3 years ago is one that we are as aggressively as ever doubling down, tripling down, quadrupling down on, even sextupling down on in some ways. It is the most exciting time in history to build. And at Ownr, we're just getting started. Thank you. We've got about 10 minutes for questions if anybody has any that they want to jump in with. Feel free to just walk up to the mics on either side of the room. Hey, great presentation and love your energy. Thank you. A lot of caffeine this morning. How many attempts you had before before you came up with a winning idea?

Or did you follow any process to come up with like how to pivot to AI?

Just to make sure I'm hearing the question, 'cause it's kind of loud with ambient noise. Is the question how many pivots did we have to find the idea?

Yeah, how many ideas do you went through to come up with this one winning idea?

So many and so many failures. The first 3 years of the company's life in 2018 to 2021, we were building something that got completely destroyed by the pandemic. It was this website builder that helped restaurants drive dine-in and reservations in a reservations module, which then in April of 2020 became completely irrelevant. So we had to adapt and rebuild the business around the online ordering thing. And then in terms of the AI transformation, if that's what we're talking about the past 3 years, it was immediate that we came to this conclusion. Because the conclusion came from asking ourselves, now that generative AI can do so much of what previously required us to manually intervene for our users, to have an onboarding specialist setting these things up, how much faster can we make the time to value and how much more effective can we make the product if all of that gets automated? And so it just led us to developing that. The hesitation that we had was not that we didn't think it was going to technically work, but we were hearing from the customers that we'd interviewed in late 2022 and the industry experts. Restaurant owners are terrified of AI. They don't want this. If you build this, you're going to alienate your customer base. But it was that conviction that we got at the show that then caused us to go all in on that direction.

Thanks.

Yep. Good question. Okay. You, sir. Yeah.

Thank you, Adam.

Two questions.

First of all, with the AI, it looks like you could do much more.

Right now on the same time, how not to lose the focus? Because you could accept like to take the different small parts, but make it enterprise ready, it's not like so easy, and then build to the enterprise product. And second topic, how you put your AI cost in your current subscription code? Because with additional AI tools for the customers, you have extra cost. How you put it into your subscriptions?

Yeah, excellent questions. Just to repeat them for the audience. First one was, how do we think about the idea of focus and getting great in one area when we're building so many different things for customers and hearing so many different things and probably identifying a bunch of customer needs? And then the second one was, how do we think about pricing all of these AI features basically? Okay. First one is a very interesting one because I think that the right answer has completely changed over the past 3 years? The right answer previously in the world of SaaS was laser focus, become best in breed, build an amazing product in one specific area, and stay focused there because that is your competitive advantage against bundling. What I think has massively changed is that so many fewer people are required to build excellent products. Like our website builder product goes head on with Squarespace and Wix and basically wins every time. We've already taken thousands of people off of those website builders. And even though they have a team of north of 1,000 engineers working on their website builder, we have a team of 7 on ours. So what that means is that with the right amount of talent density, the right amount of founder DNA, and the right level of judgment in the product and design process, that the benefit, the alpha of having a point solution and having to invest all of your resources as a company in making that point solution amazing is no longer there because the same thing can be achieved with a much smaller team of people. So one of the really unusual things about Owner is we've got about 110 people across engineering, product, and design, and about 35 out of 110 are founders, folks that have founded companies as CEOs and CTOs, in many cases raised millions of dollars, gone through YC. They bring a ton of founder-level intent agency and ownership over the parts of the product that they own. And that enables us, as we scale, to appoint these extremely high-agency capable founders, like Will that you saw there and a bunch of others, to own specific domains with small elite talent dense teams and to meet more and more of our customer needs. Of course, this eventually logically, uh, we can't just grow to meet every customer need at the same time. So the way that we think about, uh, prioritizing things like enterprise readiness, one of the examples that you you cited that could draw our focus or expanding into other verticals is that we first have this duty and obligation for the people that we've defined as our ideal customer profile, independent, takeout-heavy small restaurants in the US, to build amazing products to meet their needs that are related to the problem that we're here to solve, making their first-party ordering and first-party website and guest experience amazing. And that until we are a clear APOC, not plus on each dimension of that, we're not going to start doing things like building for the enterprise or expanding into other verticals. And so keeping the team focused on that constraint initially, and then once that constraint is maximized and the opportunity cost of pulling people off of that to start doing enterprise readiness or all the rest isn't as high, then we can consider other opportunities. But we first must fall in love with the specific group of customers and absolutely ace the problem set and the adjacent problems for them. and then pick our heads up and start thinking about other opportunities. In terms of pricing, many thoughts here. First one is it has never been easier or more powerful to have an amazing free product. That website generator that you just saw in Grader, even though it costs us about a dollar in compute to generate that Thai website, and we give it away basically. Free, and then it's $1 per month for the first 3 months. The way that it's impacted our business has been profound because what happens is when the free product has a very fast time to value and a magical, magical experience where restaurant owners are like enamored by what they just saw, they immediately develop trust for the company behind it. And they start pulling all of the other products in the suite into their businesses without having to like be pressured into it by a salesperson. And so even though It starts with a free product. The way that we've constructed our customer journey is like a staircase of value, where it starts off free. Free grader, free website generator. Then it jumps up to a bundle of $500 a month if they want to use our comprehensive suite in order to solve all of their digital problems. And when they jump up to that bundle, it takes them from being worth almost no dollars to us to being worth a lot of dollars to us. But our CAC profile looks more like a company that has an extremely cheap product and then LTVs that are really impressive for our space. That's my stream of consciousness on pricing. Much more to unpack there. Feel free to email me. You, sir, you've been waiting a while.

No, great.

Thanks so much.

It's a slightly more tactical question about your first product demo. So how do you guys, with all these agents that you got going, how do you think about human-in-the-loop experiences?

How does it tell the user what the agent is doing? Doing, or what they need to do? Good questions. The question was, with all of these agents that are spinning up that restaurant's website and analyzing their online presence, how do we think about having a human in the loop? And the second part of the question—

Yeah, just human in the loop through all of it, whether it's fully automated and how do they know what's automated or when they need to take action.

Yeah, explaining the value to the user as they go through the experience. Our framework for human in the loop is anytime we are requiring people to do the boring work of setting up our product and typing in a bunch of information for their restaurant, we have failed. That is the extreme view that we take. Anytime we are requiring work from the user to get the value, it means that we have not done good enough in making that easy for them. And so what that looks like practically speaking is the vast majority of customers use that experience experience. They start using the tool to grow their sales, and then they book demos with us and ask us how they can use the rest of our products to continue that initial benefit of sales growth. Our sales team also has the ability to see the people that are most succeeding with the free products, and they're able to also use it in their demos. But generally speaking, we want as few humans in the loop as possible in the early stages of that customer's journey. Later on, as things become more complex and they start expanding into using 15 of our products, it is often helpful, at least until we can figure out how to automate all of those other expansions, to have a human that is at the ready in case people have questions or get stuck with the more technical configuration elements. But generally speaking, we want the product itself to do the speaking about the value and not engage a human, uh, as much as possible. There, there's nuance to this, but, I will leave it at that. The nuance, just to give a 30-second summary, is we still believe that for a restaurant owner, there's clearly perceived value and maybe even actual value in having a human handhold them through. And what we're focused on is making it so that instead of being able to launch 100 restaurants per human per month, it could be 1,000 or 10,000 through automating as much as possible and then getting the human focused on the uniquely human work of, empathizing with that restaurant owner, being there to work through the more thorny problems that software can't yet automate and help them activate.

Thanks. Thanks, Adam, for the talk. My name is Pian Pian. I'm the founder of a startup called Emotions AI. We're building emotional intelligence AI to help sales teams close deals faster by giving them live guidance during sales calls. My question is, what do you think is the key driver for from the early days to now you have like hundreds of millions of revenue, what's the key driver for the growth over the time? And how much that is product kind of versus sales and marketing and all those? And then second question is, how do you think about kind of focusing on certain verticals like the restaurant versus like kind of going more horizontal growth? Yeah, thank you.

Yeah, good questions. So question one is is looking back, how much weight would I put on any of the different factors that led to our growth in go-to-market versus product? And then question 2.

Question 2 is—

Sorry, I'm super sleep deprived.

Question 2 is like, how do you think about kind of focusing on the Russian—

Ah, yes, vertical expansion.

Yeah, yeah.

Okay. The hard part about answering question 1 I think it's almost a false binary where it wasn't just the product quality or just the go-to-market folks that we have. I love our go-to-market folks in marketing and sales. They're amazing at what they do. I think both have to be in place to produce it. It's hard to, in retrospect, try and deconstruct what percentage was product quality. I think the product quality piece of having a product that is so good that people spontaneously tell their friends about it and just like immediately want to use after seeing it is a necessary prerequisite for crossing $100 million in ARR, but it is not sufficient in and of itself, especially in serving small business owners. There's all sorts of inventiveness, as you saw in the go-to-market-focused demos, that is still required to make the efficient— to make the motion efficient and produce customer value. On the question of vertical expansion, Eventually, Owner will power all brick-and-mortar business owners and how they run, operate, succeed, grow. But in the short term, the opportunity cost of putting people on that and all of the vertically expanded capabilities is too high relative to having them meet more and more restaurant owner needs. And so the answer here is that we're architecting the system and building a system in such a way that has that end in mind. But in the short term, we're focused on absolutely nailing it for restaurants before scaling it into the other brick-and-mortar business owners. And initially, when people hear this idea, they think, well, don't restaurants have a bunch of unique digital needs that other categories like hair salons and gym owners and all the rest don't? And on the surface, I think that seems to be the case. But then when you deconstruct what it is that these restaurant owners need versus what a hair salon needs, they all need an amazing website that drives new customer discovery and conversion rate. They all need the equivalent of reservations. In restaurants, it's called reservations. In hair salons, it's appointments, but basically the same thing with a few nuances. They all need CRM, marketing automation, brand mobile app generation, reputation management, and all the other elements. And so as long as we architect with that end in mind, eventually we'll be able to achieve it.

Last question. You mentioned Nanobanana. What other tools did you find useful as a CEO or founder?

Oh, where do I begin? With the risk of stating the obvious, Opus 4.5 was an absolute game changer for us. It has dramatically accelerated velocity since about December of last year. Other non-conventional tools. One of the most underrated things that I've done is using the custom GPT feature within ChatGPT for advising me as a CEO through some of the things where I don't immediately have the right scale instincts. And what that specifically looks like is I've got a bunch of different coaches set up that ingest my favorite books on things like management from folks that I have learned a lot from on this front, that has the book of Scaling People by Claire Hughes Johnson and Frank Slootman's Amp It Up and a bunch of other books that I've learned a lot from. And so then any time I'm thinking through a management issue, I ask my AI board of advisors that have ingested all of their content in none done a bunch of deep research and have all of the right context to basically, in the voice of Frank Slootman or Claire Hughes-Johnson or any of the other folks that are in there, Ben Horowitz, advise me on how they would make this decision and cite your specific sources of reasoning so that I'm able to, as I'm thinking through issues, get immediate feedback on, based on the external resources of these people, how would they solve the problem. And I've got a similar one for broader leadership, product and a bunch of other areas where I'm constantly learning through interrogating this custom GPT that has all of the right context in. Bunch of others that I could cite, but feel free to email me.

Thank you.

Oh, I'm seeing the please wrap it up. I'm sorry. adam@owner.com. Feel free to email me your questions. Really appreciate you all taking the time and hope this is helpful.

Please welcome back to the stage Jason M. Lemkin, and joining him, the CEO and co-founder of Klaviyo, Andrew Bialecki.

Alright guys, I'm just gonna join as the— maybe help moderate this, but if you guys don't know Andrew Milecki or Klaviyo— Klaviyo, well, Andrew will describe it, but far and away one of the most successful players in B2B for e-commerce, number one in the Shopify ecosystem, IPO'd. Boy, you were like the first IPO after the death of the 2021 era, right? Something like that. Were you the first one to go public?

Yeah, maybe like 2 or 3 of us in that like 2023 cohort. Yeah.

Almost nobody, right? Incredible rocket ship. I should know the metrics cold today. $1.4 billion in revenue or something like that. And kind of, for those of you that don't know, Klaviyo really dominated marketing e-commerce. And, and this is probably— if you don't know Klaviyo, I'm gonna be overly simplistic, but it is HubSpot and more for e-commerce. It's a brutal oversimplification, but if you know HubSpot, there's a parallel universe for B2B e-commerce, and Klaviyo had like 80% market share. And it was when I started to learn about the ecosystem through some investments, and I said this when Andrew was here a couple years ago, it was the single most beloved app I'd ever hear or two. I would talk to e-commerce merchants, they would like folks in the ecosystem, they would like Gorgeous and these other apps, and then they'd be like, Klaviyo, I effin' love Klaviyo. And I want to tell you why that's important. That was interesting pre-AI, and Andrew did a couple— a podcast and came here a couple years ago. And the pre-AI aha moment was— and I didn't learn this till Andrew was here before— is one of the reasons— well, one reason they loved it is 'cause it worked for marketing, but Andrew and his team had what now is maybe a Captain Obvious insight. Instead of logging into Klaviyo and seeing how many marketing opens you had and how many sends, it was able to connect with your cart and Shopify the other data and say, not only did you send 11,000 emails, this campaign made you $18,372. And if you look on social media or LinkedIn, you'll still people showing screenshots of Klaviyo, not showing email, mail sends, but how much money that made. So that changed the industry. They destroyed the competition. Everyone loved it. It wasn't that expensive, so everyone used it. Couldn't do much better. Now we're in an AI world. So now Andrew's got to rethink owning this pre-AI, the magic of email being dollars. And I want to learn what he's doing because it's, uh, in some ways you probably have to remake the whole company, right?

Yeah.

Yeah, is that helpful? Is that a good summary? Did I miss anything?

Totally. Yeah. All right, so if, if you don't know Klaviyo, the best way— like Jason said, the best way I think about is like traditional CRM. When I was, you know, growing up around 2010, 2012, we started Klaviyo. CRM meant like you were really like for sales teams, but we wanted to build the CRM for consumer businesses. So obviously retail, e-commerce, big chunk of that. And but consumer teams don't have can't have salespeople. They just, they have to deliver every experience automatically. It has to be done through software, not through people. You have way too many customers. So we started Klaviyo because basically we want to build software that could act as a digital brain and then the actual, you know, actions basically so that, you know, entrepreneurs or large businesses could just scale themselves. It's funny to think about now because everybody's like, okay, well agents, of course I'm gonna use an agent to be my like surrogate. But that was been our thesis all along. How do you build the infrastructure and then the automation to make all that possible? So, yeah, what we pulled together for today is we'll talk about two things. One is we'll talk a little bit about like Klaviyo and what we're doing with agents. It's interesting, interesting use cases, but spend most of my time on like I'm an engineer by trade. So we'll talk a little bit about like actually building AI applications at scale and at our scale where it needs to work for a lot of long tail use cases, uh, you know, be rock solid. So that's what we've got. Um, maybe just to start, um, so I mentioned I'm an engineer. I'm actually— before that, um, uh, I was a physics undergrad and thought about going and getting a PhD. Uh, you know, people talk about first principles thinking. I'm like, yeah, yeah, it's like we just measure everything from the bottom up. So a lot of this LLM stuff, to me at least, is like, well, look, I mean, at the end of the day, whatever's going on in here is probably— it's just a bunch neurons, a bunch of like, you know, uh, actual biological weights. This idea that LLMs work is like not too surprising. Um, so we should expect that they can probably do a lot of the things that we can do. And so we've been thinking, we think a lot about that, uh, at Klaviyo. And the biggest one for us is, you know, as Jason talked about, because we provide the infrastructure to know who your customers are, uh, and then actually reach them across various, you know, digital channels, right? Whether it's marketing, email, text messaging, mobile, your web, you know, customizing a web experience, customer service, etc. We realized with AI there was a big problem our customers had that we could now address, and that for us was this: I— we have 200,000 customers, but I have so many of them. Every conversation I have basically boils down to one of two things. Either one, tell me what to do, which is what this question is. Like, I have these products, I don't know how to market them. A lot of people get a little anxious, a little nervous about marketing. A lot of you have experimented, but maybe they don't have the confidence to know what's right. They don't have the time to do the analysis to prove it. I wanna grow, and I wanna just deliver an awesome experience. So customers ask us, hey, how do I use your software? I think one of the cool parts about AI is I think we built up all this, software, this infrastructure, there's a big gap between like the best users like actually using it versus what the software can do. And AI allows us to close that. And then the other part was it's like, okay, well now you've told me what to do, can you just help me get the darn thing done? So a lot of folks say like, hey, I don't know if I'm creative enough or I don't know how to do that analysis or I just literally run outta time in the day. Can you point and click and mash the buttons for me? so that's what we set out to solve. Like, so we started on this probably about a year ago. We've gone through, you know, dozens of iterations. But one of the things that we have at Klaviyo is like everybody, you know, we have a lot of engineers, but everybody codes, you know, from myself all the way down to like some of our like co-ops and interns, you know, for the summer. So when we started out is, this is the prompt, or probably part of the prompt, that in January when the models got really good, You know, one of the things that we have as a pattern is like, hey, well, often we meet every Friday to review progress, and usually, you know, over the weekend or early next week, we'll say like, hey, we've got some ideas. Like, we're not gonna sort of say, hey, let's wait till the next sprint or next month or whatever to prototype them. Let's just get it rolling Friday afternoon and see where we are early next week. So one of the things we tried out was we had first built an agent that was using a lot of the, you know, kind of state machine type of architecture, but we wanted to move to something that was just more open-ended. Like a lot of us are building agents now, we just give, give it a bunch of tools, give it some guidance and let it rip. Um, so you know, that Friday night I said, all right, well like let's see how far we can get, uh, and fired up our first prompt. And so I wanna show you a couple of the lessons and tricks and techniques that we've been using to make stuff like this actually work. 'Cause it's pretty easy to build a proof of concept, but how you actually get it into production. So a couple of lessons from us, right? One of the first ones is, is that like, getting the agent going is pretty easy, but like figuring out where it gets stuck is hard. We've all seen some of the examples of like, hey, you know, it accidentally dropped my database, etc. One of the things we've done inside of Klaviyo is we've architected it, our agent environment, so that it has very tight access to various like tools tools and controls in terms of what it's allowed to build on our infrastructure, and then good sandboxing staging environments so that like, hey, you can build up prototypes, then ship them live so that we can actually look at them, but we don't risk any of our internal infrastructure. So we don't have any of these like, you know, 9-second failures. The other thing is that like, while demoing it is good, we talk a lot about this, you know, this meme here of like, hey, you know, start out, draw me a couple circles, and then just draw the rest of the owl. Well, that's actually quite hard. And this gap is something we obsess about, is this prototype to how you get it all the way to completion. How do we make sure that at our scale, when we're dealing with 200,000 customers around the world, multiple languages, very different kind of workloads, businesses that might have 100 customers or 100 million customers, How do we make sure that it scales for all of them, and then it's gonna be performant? This is one of the things where we've actually written agents that I'll show you in a second, that once you, once you go past prototyping, will actually validate the actual software, and then will run load tests for us, so we can figure out where things are going to fail at scale. So one of the things that we've done is, we've taken the entire, you know, software tool chain, right? You know, things I'm used to doing, you know, by hand, one by one by one, is once we've got a prototype, whether it's testing, load testing, testing against various edge cases, thinking about like, hey, what are the features or the functionality that we're likely to need next? How do we then build an agent that can plug into our, you know, agent loop to then go run all of those tasks? And the reason we did this, so as we did this, we found that like, When we started, you know, a lot of our code, like if you're an engineer, we found there's a lot of stuff that looked like on the left, right? You build out these prompts and you have one by one by one, and you kind of, you know, and you add on and you get this, like if you've been an engineer, like you end up with a lot of like spaghetti code, a lot of stuff that's like just an unmaintainable mess. One of the things we started building last fall is a project, an internal agent that we have called Dark Factory. If folks don't know, in manufacturing, like, there's a concept of lights-out manufacturing, which is basically like, hey, if you have all this machinery and there's people around that are kind of using it, then you keep the lights on because they're the ones that are actually, like, going in there and fixing the machines and all this kind of stuff. Increasingly, there's this concept of, like, well, the more you can automate the factory, you can actually just turn the lights off, 'cause that's not what, you know, the machines are after, right? They don't need that. One of the patterns we built is this: what does it take to build a good, you know, so you build more complex applications, a software dark factory? And so for us, what this loop looks like, right, is that prompt that I showed you up at the top. We run this through our dark factory, which you can access, you know, through a standalone repo, through Slack. You give it a prompt, and what it will do is it will take that, it'll write out, you know, like it'll be the PM for you, for you. It'll break it out into a whole bunch of different specifications. It'll then break that down into— it'll decompose the problem into various engineering subsystems. It'll write actual contractual API interfaces between those systems. So for our example, for our composer agent, which is this one that solves the what should I do next for marketing, it breaks down into an agent that can design, do creative, pulling from Canva and Figma and your assets, It can then do the actual, like, orchestration, the building of like, hey, what segment should I send this to and when and why? It'll do analysis to figure out, hey, what does it predict will happen? We use our dark factory pattern to basically take something that's that broad and break it down into those component pieces and then have subagents run against each one. And one of the critical things that I'd recommend you do is having these very clear contracts and interfaces between different parts of the platform. Build really makes a big difference. And so we, you know, as we call this pattern now, we have like in some cases we're able to run, you know, for the whole weekends on projects where we're prototyping something as big as this agent. We're saying, hey, here's a new architecture we want you to go after, decompose it. And then when it comes time for us to actually human review the code, which we obviously still do, it's actually a lot easier because the system is not this big tangled mess. It's nicely laid out, right? And it's been done by, you know, now agents at scale. So that's our pattern for building and how we did Composer. And then with that, you know, that first prototype I mentioned in January, we built this more freeform agent that could do all sorts of stuff, analysis, creative, design. And now that's what we've shipped to our customers. And that first prototype took a weekend to run and build through our dark factory pattern, and now we've been able to iterate on it, on top of it. And so now we use this when we're building, you know, really it doesn't matter if we're building agents, or some of the core infrastructure of Klaviyo, this is a common pattern we'll use is to give it an initial prompt, let it go hunt. And then one of the things we've now built into Dark Factory is this idea of human-in-the-loop feedback. So we ask the agent to think about, hey, when you get stuck or you're not sure about a decision, we want you to raise it up. So one of the things that will come back, you know, over the course of that build, right, on Friday, Saturday, Sunday, is it'll say, hey, here's some questions where your requirements weren't quite clear. Clear, I need you to specify it harder, right? So instead of doing this all-in-one, like, kind of plan mode where everything comes up front, we actually break that apart where you get these kind of repetitive little questions, you know, over the course of like a couple of hours or a couple of days, depending on the size of the build. I think this is the future of how software is going to get built, where, you know, very much like when you do a normal product review with actual, like, humans, with people, like, you meet and you discuss the key product decisions, and you kind of harden what you're after. We can now do the same thing, but really having agents drive it and then just raise the flag when they get to something that seems uncertain. All right. One other tip, lesson we've learned from building our marketing analysis composer, run your business agent, is we're a Boston-based company, so we're big fans of this guy, Tom Brady, also a San Mateo native. I think one of the things that's happened with LLMs, we're now very firmly in this world of like, hey, it's all about building the software that goes around the core LLM, you know, the harness as we speak, as we talk about it. So one of the mental models we have is I often tell folks like, look, you should treat the underlying, especially if you're building agentic products, you should treat the underlying LLM as like a very athletic, you know, middle schooler or high schooler, right? They're really good at a lot of sports, but if you want to make them great at one thing. So for Tom Brady, it's funny, he almost— he got drafted to go play baseball to the now defunct, you know, Montreal Expos, so he might have had a great career there, but instead he went and played football, and we all know how that ended up, and we're very, very thankful for that in New England. Um, but the reality is, is like, it took a lot of like coaching and tailoring to turn that general-purpose athlete into something that was great at a specific domain. And so one of the things that we've we built into our marketing agent composer to help differentiate, help it stand out, help it be really good at the sport that it's playing, is we've given it a feed of data, like real-time data of what's happening between all of the businesses and consumers across the globe that Klaviyo can see. And what's cool is we make this analogy that, so like for Tom, it's probably like, look, you know, for him to be a great football player, he needs great coaches, people that would put him through, you know, show him film of maybe Joe Montana and some of his idols. But also put him through his paces of like, these are the kinds of drills you need to run to be a great quarterback. The same exact thing I think is true for agents. And the way that we go about doing that is we took our dark factory software building system and said, hey, I want you to build into this a feed of how consumers are reacting. And so what our agent will do is every time it has a proposal like, hey, we think you should market to these customers in this way, you should send them this email, it will go check its work work kind of against its coach that will say, hey, how good will this— how good do you think this will actually perform? And that kind of, you know, a little bit like coaching agents will give it feedback. It'll give it a numerical score of, hey, here's how much engagement or revenue, uh, what kind of outcomes will this produce. And then also give it some feedback on how it can tune and optimize. And we found that like when you combine these two things, that's what really makes like our— an agent harness really special at a particular task, training that general purpose LLM to be better at the, you know, domain-specific sport that it's playing. Um, and finally, I got, uh, two more things for, uh, one, we are very big believers as you're building agent products. Um, so one, you should use agents and teams of agents to build agents. Two, use that LLM harness as like a general-purpose good athlete. Three is this new category on top, uh, what we refer to inside of Klaviyo as this like autonomy layer where all the agents live, where intelligence is like, is the unit, is the unit. You know, we all have, all of, you know, what we tend to think of as software, I now think of as its infrastructure. It's just a means to an end. So if you build some amazing UI, that's great. Like there might be, there are a lot of humans that will still use that, some AIs. But the reality is, is like headless is now the default. And so, you know, in Klaviyo's case, we look at ourselves as like we're the database for all of your customer consumer data. We know everything about your consumers. We then provide the infrastructure on top of that to market a message to them. The opportunity that agents provide is then to be that actual like intelligence on top. So all those customers of ours that come and say, hey, I know, I know Klaviyo is a great product, but I actually, as good as it is, I don't know how to use it optimally. We're now populating that intelligence layer with our Composer agents so that it can interact with our customers. It can actually teach them or even do their work. And we think that's going to be the future of frankly, like how marketing and just consumer experiences in general are delivered. It's going to be much less like each of us deciding every single decision and more, hey, having really good agents that will bring to us a feed of, hey, here's the campaigns you should run next. Here's the experiments I ran last week. These These ones worked, I'm gonna keep doing those. These ones I'm gonna throw away. All of that's gonna get automated and happen at much larger scale. Last thing, a mental model that we have when you're building, you know, agents and infrastructure for them. So we think about Klaviyo and like all the stuff that we've built around data, marketing, customer service. One of the things we found from our users that are leveraging our components Composer agent is normally in software you have this like power law of like how good your users are at your products. So you've got a couple of really advanced users, maybe you put them like on a customer panel together, you bring them in, they're really the ones pushing the limits. But then you have a lot of users that are novices, and that's no different for Klaviyo, right? We have a lot of— we have a— we work with a lot of businesses where people have maybe an hour to a week to use our software and that's it. They just don't have the time and the bandwidth get super advanced, no matter how good our software is. So you can imagine there's kind of this distribution, right? We have very few, like, really great users, and then a lot of users that are kind of novices. Agents change that entirely. With agents, they go straight to the most advanced users, and we actually think sit to the, like, right of them. All the agents from day one are power users. And so it's a different way of thinking about building the underlying infrastructure of, you know, of what is your software. So I'll give you two practical things that we think a lot about. Uh, one is this idea of needing to onboard users. We just don't worry about that as much because agents will onboard automatically if there's good enough documentation, they'll reason through it. And then obviously like we'll give them some hinting. That's what we do to Composer of here's some guidelines of like general marketing best practices. But onboarding actually matters less because these agents will go zero to 60 much faster. The second thing is, is because they'll push the limits so much, we've actually started asking the agents we built, what functionality is holding you back? And I'll give you like a very concrete example here, right? In the world of email, there was probably about 10 years ago, maybe 7, 8, 9, 10 years ago, there was this big push to make email more interactive. It's something that I've cared about for a long time. Why, why is it that, you know, if you know, if you want to If you want to buy something that you see, you know, in a message, you can't just buy it straight from that, like, from your email client, right? Now we're limited by like what email clients provide, etc. So there's a big push to make email more interactive. Google has big push AMP, all this kind of stuff. Well, turns out, so, you know, well, Google released this AMP specification for email, and it's out there and you can use it. A lot of people didn't know how to build for it because now you're effectively building a web app inside of this, like, you know, email payload. So for a long time, we're like, man, this is kind of tough. And we'd give people, like, basically, you know, just a raw HTML editor, but a lot— very few people could figure it out. Almost on day one, our Composer agent figured out— they're like, hey, if I want better conversion, I can make this email more interactive. I can include some, like, carousels of product imagery. I can make it more dynamic. I can include some JavaScript calls, all this kind of, you know, calls back to some end servers. I can do all of that, but hey, Klaviyo, we have a problem. Like, I'm missing some APIs that would allow me to edit, you know, and specify my content according to the AMP spec. Can you fix that for me? So we've started to ask our agent, hey, what are tasks that either it's hard to do, like, you know, that we just don't have enough flexibility on, like that, so we've now built that out, or that an agent wants to do, like, very repetitively. They don't want to do it once or 10 times. They want to do it thousands of times. Another good example for us is we've dramatically leveled up the underlying, like, experimentation infrastructure in Klaviyo because now agents are like, well, if I have an idea, it's so cheap to test, I'll just throw in a test and give it to, you know, 1% of my audience and see what happens. Versus a human user would say, okay, I would love to do that. Yes, that makes sense, but it's just very hard to do. So last thing I'll leave you with is like, as you're building agents and you're using these like, you know, dark factory type patterns to build, I think you should expect that for your software, right? Whether you're building agents, they're going to be the super users, or if you're building the software that you expect either humans or you're going to plug into MCP or other people are going to expose to algorithms and agents, expect them to be super users almost right away. And I would say that whatever you're doing to get customer feedback, I do the same thing with agents. With that, thanks, Fred.

Sure. You want, you have time for a few questions? You want to chat with me or take questions from the audience?

Either way.

Hold on one second.

Go ahead, fire away up front.

I know I'm not in the union, but if you want to sit, you can. If you'd rather stand, well, either one.

Go ahead, say it again. I'll repeat it. Okay, so the question was, how does, how does the underlying tech stack change as you've gone AI first? So I'll give you a couple norms that we have, and then I'll explain the tech stack. So first is, our expectation is that everybody that's in our product and engineering groups, including designers and PMs, everybody's committing code. So with that, that's changed the tech stack a little bit, because often designers and product managers, well, they could set up a local development environment, but we've also given them more sandbox environments where they can basically prompt and then generate back a pull request and an actual visual. So a very common pattern for us, for example, in design, is stuff that used to drive me nuts around like, hey, inconsistencies in our UI and like not using our design system all the way, right? Now designers are like, well, I'm just going to fix it, right? Or they'll even write an agent or tell, you know, Claude or ChatGPT or Codex to just go solve it. Now they can do that, but then they have to do a visual inspection to make sure it's actually correct. And sometimes they then want to click, click through. So That's great. More, more sandboxing and more like staging, um, has mattered a lot. Other than that, I think the biggest thing is like, it's not that our underlying infrastructure has changed so much as we're now deploying these other patterns on top. So I talked about this dark factory pattern, this idea of giving people the ability to say, hey, I have this prompt and I wanna sort of run this session, um, and really in a multi-agent build, uh, against our, against our code base. Uh, that's probably been the biggest one. Yeah.

And everyone, I know it's great to, I mean, how many employees do you have, Andrew?

We're about 2,300 Klavians.

Okay. I know they're all great, all the Klavians, but all the product folks and the designers, whether it's through Dark Fact or others, truly can commit code. They've all stepped up to the challenge.

Yeah. So look, I mean, we've, so we've, so you know how there's these different levels of autonomy of like driving, right?

Yeah.

So we've defined, you know, say like, okay, an L, you say like an L1, you know, this not unique to an engineer or product person, like you say same thing for sales and marketing. Hey, they, they might be like, you know, I doubt I use it to search or whatever, right?

Yeah.

You know, level 2 might be like, okay, I, I, I spin up a little, you know, Claude session or whatever and I kind of run an agent. We've told everybody, it's like, look, by the end of June you need to be an L3, which is like if you're not constantly running, you know, either, you know, end sessions or a team of agents.

Yes.

Right. You're just not, I mean, like you're not gonna skill to survive in this next era. So that—

now what's interesting, your average PM at Klaviyo pre-AI who's working on wireframes and specs now has to become an L3 in your hierarchy?

Oh yeah, absolutely, yes.

I believe you, I'm just learning.

Yeah, yeah, well, there's no— there's— and I mean, there's been very little pushback on this. I mean, it's a look, you know, it's a little intimidating because there's— okay, we're— look, the slope is steep. Yeah, but what we said is like, look, I mean, the The reality is, well, yeah, one other thing I think about that I'll share. There's so few people that are gonna get there, right, in the next year or two. Those skills are gonna be so valuable. If you truly are not talking about AI, right?

Yes.

But if you're the kind of person that's like, I can put an agent to work and actually have it produce value. If I can put a team of agents to work that could decompose a problem, right, and do each of the pieces and I can find a way to check those and validate those, you're gonna be able to take on bigger projects. If you can do that, Klaviyo or otherwise, are gonna be enormously successful in the next 30 years.

Without question, right? Do you think, maybe just two questions, 'cause I know you talk so much about Dark Factory. For the non-developers, since you have, are they getting close to programming in English, to coding in English, for Om John's point? Do they still need to understand code, or do you think they can almost code in English?

Yeah, so there's a little course we teach internally about like systems system decomposition. By the way, this one I'll give you just a quick analogy. Uh, I was talking with our people team about this. I, I, we showed them this, you know, some of the software pattern we did, this, this idea of contracts between, you know, software systems. Hey, what does the front end do? What does the back end do? They took that and said, oh my gosh, we have to think through like what the jobs of the future are, you know? And they said, well, I said, yeah, like why don't you use the same program to tear that apart and figure out what does, you know, like a marketer or salesperson engineer actually to do in any real contractual terms, right? I know this feels like— so I think, look, a lot of humans, we're used to these kind of fuzzy interfaces, and then we'll like meet, what, meetings to kind of figure out what each person is going to do. I think that it's not a great way to work, honestly, and it doesn't work with agents because agents aren't that fuzzy. They want to— they want hard and fast rules. So what we've done is we've actually took the same pattern and said, hey, go think through this team, this function, this job, map out the actual work, right, as in almost like software interfaces. And then tell me about how would you scale that team, right, for 2x, 5x, 10x growth, right, or scale or volume what they're doing without, you know, any, uh, you know, without, without adding any, you know, say more people but more agents, right? And just ask it to reason through that. That has actually worked really well.

And when, um, just because I'm curious, because I don't know that that level of depth of the whole organization committing is that common today.

I think it will be, right?

It can also lead to slop because if every PM can now build features, even if they're not committing to production code, if they can build features and they can build a basic feature in an hour or two, not one prompt, instead of maybe getting a feature out a quarter, you could build 100 features a week if you wanted. And it rises, even inadvertently, it rises the level of slop. How does the organization process 100 features built by a non-developer if you have 50 PMs people on your PM team, that could be 2,000 features a month that are sort of half-baked and sort of work and are intellectually interesting.

So taste still really matters a lot. I'll give you one thing we do. So, okay, every, you know, every Thursday, part Thursday, like I do this after, you know, after we wrap up here, and then on Friday, I spend all day with our product team just reviewing work. And so one of the common patterns that we found is like, for folks that are new, they're like, okay, well gosh, like, what kinds of what feedback are you gonna have? And so one thing, you know, we used to write this down, but then, you know, sometimes you'll read it or, you know, oh man, we didn't update this, you know, the kind of the guidelines change or whatever. I took all the like crits, all the like feedback that we'd given over the last couple years, right? That, you know, these like, you know, whether Zoom meetings or raw notes, all this stuff. We just dumped it in a database, right? Gave it to an agent and said like, look, from here on out, before we even get to those reviews, Here's what's good, here's what's worked, here's what won't. And so that has helped us, like, when people have ideas, because there's an explosion of ideas, that's great, they're not all worthy of shipping. What we've said is like, hey, this needs to pass this bar. And by the way, if you think there's something wrong with our rule set or feedback, like, push back on it, right? We found that actually that cuts down on a lot of stuff really quickly, um, in terms of like, ah, yeah, it's kind of a good idea, but it doesn't meet certain conditions. And then obviously now we have a team that's like, are actually pushing back on our first principles of like, hey, what makes a great product? I think in the future, like, look, we're going to be able to give agents the idea, um, you know, I think what taste matters, I think we can codify taste too, of our product should work this way, it should feel this way, these are the outcomes we're after. The trick for us is that decomposition. It's like, can you have it explain its work? And then for now, we have to train it up on like, okay, where, you know, where there's logical, uh, you know, uh, you know, failures.

And does, does, does, um, and then I want to just I want to talk briefly and then we can take a few questions before we run out of time. I want to talk about your customer agent next, but since you've sort of built Dark Factory in-house, right? Maybe just two questions. How do you encompass that taste organizational-wide? Do you have an organizational-wide memory? Do you have guidelines like Klaviyo guidelines for product development? How do you do that and why do you, are you gonna maintain it in-house? Is it worth, in a year, is it worth maintaining Dark Factory in-house versus finding a replacement, because the rate of paint changes so fast and you have to maintain it.

Totally, right? So look, this, this idea of— I think obviously like our, our taste, yes, now lives in this like database, which, you know, instead of it being institutional knowledge where it's like, you know it or I know it, like that's kind of silly. And I also say it's like, look, well, it's, you know, well, I have a lot of input into it. It's, it's, it's a living, breathing thing. We'll work on it together. So that, that, you know, that now having agent interface on that is a big, uh, is a big plus up, because now more people can access it. It's not like I wasn't there for that, or I missed that, or whatever. Um, now in terms of the software pattern, like, look, our, our belief is the, the gains, uh, are going to go to companies that can stay on the bleeding edge. And the bleeding edge is hard because you have to invent a lot of stuff, you know, figure out the patterns. I tell you, for us, like, I, I think this, this, like, you know, this team of agents pattern obviously exists. Um, it's going to get more codified. We're going to keep investing in, in that until we see some other, you know, kind of open source or other product that we think can really do this well. But as I go back to this, like, the idea of this LLM as like kind of this very smart, you know, kind of, uh, or very good athlete, I think a lot of these patterns that we're building around it are just leveraging things that the labs, you know, are building into the models. But you still have to build around it, and that, that part they may build, but also like it could just as easily be somebody outside of the labs. Yeah, yeah.

All right, let me, let me just switch gears for folks. So that, that was, I, that was super— I didn't even know we talk about internal. On, on your customer customer-facing agent. Just help me, what I've, so you said you don't really need to train the customer agent, right? So let me just learn from that. When a lot of the folks that have agentic products here in GTM, if you don't have an FDE or an SDE, it sucks, okay? Point one. Point two, when I watch what Zendesk has said, which is adjacent, right? Basically what Zendesk has said, and it's an older platform, their self-service agent, support, it can get about 20% of the way there. But to do what Zendesk needs to get to 80%, you need FTEs and humans. So how does that tie to the fact that for your SMB customer base where you can't even afford FTEs and SEs anyway?

Yes.

How do you actually make the agent great?

Yeah. Okay. So for Klaviyo, we talked a little bit about, you know, the prompt I showed you in that product is we think of it as our inside agent. It helps you run your business. Our composer agent helps you marketing, understanding your customers. Then Jason's talking about, we also have this outside agent. It's an agent that's, you know, digital representative of your, your business to the world. I think these are going to be the— they're the future of like how consumers and businesses are going to interact. I think in the future we're going to have these customer agents that actually act as like a proto web server where it's like you just query into it or you visit a URL, you provide some context, and it will just render output HTML based on who you are, what you want to see, right? So we're very much far down this path. So I think this idea of like these customer agents as purely for customer service is way too narrow. Um, now one of our beliefs here though is like, look, I remember growing up in the, you know, late '90s and, uh, you know, GeoCities, and then building my first websites in the early 2000s. Um, you know, this is all like, you know, doing— I remember seeing WordPress and be like, wow, this is magic, because I don't have to, you know, I don't have to code HTML anymore. And then obviously see the great, you know, website builders we have today. The reality is agents are going to come into our lives, like for every business, much faster than I think we're going to be able to build up a population of people that know how to build them. Because agents are just— they're more technical. Uh, we're talking about things like tools and API access and all this stuff. And you know, as Jason said, we've got— I mean, look, it's actually— it's not even— we have some very, very, very talented, you know, small businesses, very technical, um, you know, founders, uh, or owners of those businesses. Even for them, they're like, look, I don't have time, I'm not going to be keep leading edge on this stuff. I, I mean, I, our team is, you know, stays up, you know, you know, whatever, nights and weekends just to stay on top of all the latest things that you can do with the models. So our firm belief is that like the way we get to— we have a— by the end of this year, I think most businesses are going to have their own agent they deploy either on their website or behind a phone number or through email or to ChatGPT or Claude that you can just access. But to get all those businesses, right, take our 200,000, get every single one of them using, building that agent, there's no way we make that happen this year if we have to teach all of our users how to build agents. And obviously that's, it's cost prohibitive for them to hire somebody to do it. So we are just down the path of agents can train the agent. And I think the trick there is we're just obsessive about, you know, when you hire the, you know, you hire an FDE, I like, we look at like, what does that person actually doing, right? What is the actual literal process? And so I'll give you an example of what we do. We take, you know, we can take a feed, right? And we can synthetically generate this if we have to, of all the different kinds of cases and experiences or conversations, right, that somebody might have with an agent that represents a business. We can take a real feed, we can generate a synthetic one. We then classify that into various use cases. We take a couple of representative examples. We then give an agent that has full access to our platform, which is all API-based, right? And we say, great, I want you to train up this agent, right? Here's, you know, here's an eval, here's a use case. Don't stop, little Ralph loop, don't stop until this come back, comes back and it works. And similar to our Composer agent, what it will do is it will interrupt when it says, hey, I've hit a logical point where I just don't have the instructions on how to do this. So in e-commerce, Hey, somebody wants to process a refund, let's say, right? Well, different companies have different rules around that. If something's above a certain, below or above a certain dollar threshold, they might say, you know what, we actually don't want you to return it, it's too expensive. Other cases they'll say, okay, we do want you to return it. There's also all sorts of little rules, right? This agent will figure out, they're like, look, I can build this, but I often find that businesses like you, and this is where we get value from scale, have rules like this, can you tell me, what yours are, or it'll say, hey, I know there's a backend system that say, let's take the case of returns, will print that shipping label and send you, you know, say a box or something to return something in. But I need to know where that system is, right? So we'll ask and request from the user to basically go use those, and we have a lot of integration, so it's a little bit point and click, right? And in the future we see a world where it's like, I don't know, maybe you could give it access to just go hunt on your behalf. This is, we're already seeing great success with this where we can, you know, take a standard business and, you know, find, you know, 5, 10 use cases, have the agent train itself up, prove that it works. And then what we deliver to customers is we say, okay, we built you this agent, we took all of your use cases, and hey, it's already at a 50, 60, 70% resolution rate. That changes the game. I think a lot of these agent products, this idea of like, you know, we go— you go through a demo and then you decide you're going to implement it. There's a lot of deciding, you know, you don't get to trial it really. It's all going away. I think for a lot of agents, if you can't try that product out of the gates and it requires some implementation, you're dead on arrival. And so for both of our agents, both Composer and our customer product, we think that's like, no, the wow experience day one should be you get to use it, see it working, and then you can either immediately turn it on and then obviously you can optimize and tailor it to your own needs. All right.

I got one more before I lose you, but, but I think we've got one here from from the audience. It might be better than mine.

Test, test.

We're here.

Hey, yeah, so super interesting talk. I have a practical question for you, if I could get your advice for a moment here.

So I'm an engineer as well.

I am building this GTM engineering team for a company that's been around for a little more than a decade.

So we have a traditional marketing and GTM team and things like that.

And what we're trying to do now is do do a lot of modernization, right? And so we've gone pretty far. So we built a lot of really great automations and agents internally that we use, which are amazing.

But we're now getting to what I would consider to be like the really complex part, which is we basically have like this fork in the road. We can go one or two directions.

So one direction is we can spend a lot of time building harnesses, kind of like I think what you're talking about with Dark Factory, where you get a lot your non-developer employees to meaningfully contribute to product and roadmap. Um, the other option though, which we've been thinking about, is, is— and I'm boiling this down to simplify a lot, right?

But like, giving a prompt which runs on a prompt job, which is like, hey, you're the owner of this company, what are some features we should be building based on feedback from customers and feeding off the context?

And then have that manually trigger these raw loops to like create a PRD, implement the PRD, submit the pull request, use the case guardrails and everything that you're kind of describing here, right?

And I genuinely don't really know which is the best approach. Part of me is like, well, we should definitely do enablement and invest in the people. Part of me is like, well, it might be legitimately way easier to actually follow through with a completely automated kind of pipeline.

I was curious how you think about it at your company since you have a big Sure, so I'll say two things.

The first is this idea of like decomposing your workflow really matters. I think with complexity you have to map out all the different pieces. So we, I mean, we talk about whether it's marketing or customer service or software development, we really like to tear apart all those steps. I think that's a very important part of the process. And then once you've So as you do that, we just try to pick off individual, you know, phases, right, or steps, and start to automate those, right? Whether, you know, and sometimes they're, you know, we think about whether we're gonna use an LLM 'cause we want something that's maybe a little more non-deterministic versus no, we really want something that's gonna do the same thing every time. So for instance, in your case, I would definitely say like, yeah, I mean, we have a bunch of, you know, algorithms and agents that like in the past we used to have our customer support team our product team, our product expert team, used to like every month we'd meet and we'd say like, bring your top 10 things that you like, you know, sort of you're thinking about, right? And it was good, I mean, but like it was a lot of work to sort of like, well, they had to argue about what are the top 10, there's a little bit of recency bias sometimes, all this kind of stuff, right? You get very good insights, but it's just a lot of work. I would absolutely go down the path of like, hey, I'm gonna have some agents, right? And I'd say, you know, for us, this might be like an engineer, a fraction of an engineer's job, job, right? I'm going to own that and to say like, hey, it's going to output, say, that, you know, prioritized list of what it thinks are the most important things to build with some ranking function, right? Of like, hey, here's a number of occurrences, etc. And we've seen amazing things from that because we found like, for instance, like there's just enough human bias. You see a lot of things that come up as like little paper cuts from our customers, either from usage data we have or from like support tickets, support conversations that just don't make it onto that top 10 feed. Just because you're like, ah yeah, it's not a big deal, but it just happens a lot, right? So I think I would pull things apart and then for each one go after and automate it. And then we keep a human in the loop to keep providing feedback and then we record that feedback and then use that, thankfully, as a training set, right? So that we can either improve the prompts of the agent or we just, you know, bake it and have the agent itself improve itself. I think there's a lot of interesting stuff with now agents like in reinforcement loops. They can do a little bit of that learning with the and give it loop feedback.

I think maybe we got time for 1 or 2 more questions. Hi, thank you. Really quickly, so I understand your Dart factory stuff.

Like, you go and you put your agents on different stacks.

But we have like a 10-year-old legacy backend.

Yes.

So my question— and right now we're doing like a spin-off of We want to put an agent software out there, or agents. So, and using our old backend, so I get all this prompting and all these agents, but, and I get that you say, you know, the new way of doing software is going this way, okay?

But our old backend, so do we do it like by coding?

that we go all the way by coding?

Yeah.

So the old stuff, the old school developer, I'm a developer, that's gone?

Yeah? Yeah, there's two things. So look, we obviously, we've been around 14 years. We have a lot of code too. I mean, every couple months, I talk to an engineer that's like, oh yeah, I ripped out some of your code from 7 years ago, right? So look, this idea that the code I mean, every code base goes through these iterations. It's just gonna happen faster. You know, a lot of the original functionality is like, okay, it has my commits against it, but like, those are, a lot of those are long, long, long since gone, right? I think now the pace at which the code that just represents the functionality changes is gonna go faster and faster. There's two things we do to sort of like prevent the kind of spaghetti, the like a mess, right? 'Cause that's every engineer's worst fear is there's gonna be some bug, they're gonna look at this and be like, oh my God, it's just like a big rat's nest of stuff I have to untangle. So two things we do to solve that. One is this idea of like, you know, a little bit like a factory floor. If you nicely lay out all the components and you force an AI to think through that, that's very important. It makes a lot of the, you know, issues, bugs, you know, the kinds of things that are always gonna come in software a lot more shallow and easier to debug. So that's a very important step. And that's one of the things I think like, that's why I think these patterns like Dark Factory are gonna matter. 'Cause it's, I mean, everybody, look, you're using a prompt to build something. The magic of what we built is that it will split it into 50 or 100 different jobs and then know how to coordinate all the pieces so you get what you want. The second thing we do is, for older code bases, is we leverage a lot of the tests that we've written, but also actual customer use cases and edge cases to surround it, you know, with like, hey, here's how it should behave and perform. And ultimately, like, look, I mean, you know, having built software for now, you know, almost, you know, a decade and a half. Um, no software is perfect, whether it's human-generated or AI-generated. Um, the trick to me is, is like, you want to make sure that the standard path is good and a lot of the edge cases are handled up front. Well, it turns out, like, the algorithm to reason through what those are, you can do that with an LLM, then plus all of the data set that we have. So we can look back historically at the types of usage patterns. You can basically codify those. And that's where then I think, like, you can start to say, hey, the requir— like, humans are, you know, they're not so good at requirements, 'cause it's just a lot of work to document them all, right? But we can actually use usage data to refine those, and then we can feel a lot more confident that if we ask somebody to go build, say, a whole new backend for something, right, some subsection, hey, it needs to meet the following, you know, pass the following evals, test cases. And if it does, then yes, it's gonna work. It's laid out in such a way that it's easier for a human to review.

Yeah.

Maybe one last follow-up question, just to get your— for folks that do have old— he's got a 10-year-old codebase. Some of your code is 10 years old, right? My, my learning, and just curious your final thoughts, in an agentic world, whether it's, um, if your internal APIs or your external APIs are state-of-the-art, it can make up for a lot of dated software. We use My epiphany was when we went headless on Salesforce. I had not logged into Salesforce since 2 companies ago. I didn't have, it turned out, Amelia told me I didn't even have a seat. Now that we run it headless, I can query Salesforce in real time all the time. There's other internal use, so I actually think for some folks, if your software was almost hopelessly dated a year ago, just focus on great APIs, internal and external, and you may be able to make your product, and to me, like, Twilio is another example. It almost died 18 months ago. It was growing 4%. Now it's growing over 20% because the API worked with agents. So I don't know, both inter— but maybe just inter— you can make old software good if your internal APIs are great, because the— that's all the agents care about, right?

It's the first thing that I think— yes, first advice give anybody if you're working on older or new code base. Yeah, you know, uh, Amazon had this concept, or, you know, or, you know, and others in the early 2000s of service-oriented architecture. We've been through this many, many times. That's why I think if you previously, if you were building software, right? Anything that a human would log into and use, I think the right mental model is think about it as infrastructure. And we think about infrastructure as things developers use, and therefore they have APIs. If you think about it that way, then yes, you'll naturally lead to, okay, what are the interfaces, the actions, the APIs I expose? And actually things get a lot easier from there. And you get the benefit of now you don't necessarily need a user to level up. You'll get these PhD level, you know, agents that will just figure it out and perform. Prefer your stuff because it's accessible and it's really good at what it does.

Yeah, I think it's almost like even if you slip far behind, if you make your internal and external APIs, you might get parity again. Like, that's the— I think it's the simplest thing you can improve in your product, internally or externally, is have state-of-the-art agent-friendly APIs. Because you just don't know, all of a sudden you can have your product guys come up with magic that they couldn't do a year ago if your internal APIs are great, right?

Yes.

Are you—

one last thing we've done for the infrastructure, I know we gotta wrap, is We've also gone down this path of how do we let somebody sign up for our software, right? You know, with their identity, right? Email, phone number, whatever it is, and use our product without ever touching it. So I think the future of, say for us, right?

Headless play here, you mean?

Customer CRM is you might be in a, you know, with an agent, or just say, hey, I'm starting a business this weekend, and I need a place to store everything. I wanna spin up an email, text message list. Just go get it done. We have a team, it's really a team of an engineer whose only mission is to make sure that that is an awesome experience, right? And we expect in the future there may be something where you never even, you know of Klaviyo, you got a bill or whatever, right? You pay agentically, but you never have to touch our interface and it's a great experience.

It's the headless Klaviyo, right? I'm all in. All right, everyone, big hands up for Andrew for staying on. Appreciate it.

Hello. Hello, faster. Welcome, welcome. So this is our now second— I shouldn't say inaugural— our second SaaStr pitch competition. So as you guys probably saw when you were buying tickets, flying in, booking your flights for SaaStr Annual, you could apply for our SaaStr AI pitch comp. But this is the first year we've partnered with Google for Startups, Gemini, Google Cloud, to actually have AI help us judge some of the entries of this year's startups that submitted to come win the pitch comp at SaaStr. If you guys remember Mosaic, they came up here, they did this last year. Then they had a booth in the expo hall as part of one of the things they won at Saastr Annual this year. So we've got 4 exciting startups. We'll try and make it through all of them as we can. And from companies all over the world, we try to pick them— I try to pick a mix of different things. But funny enough, Gemini just picked all the finalists for us. So it gave it a score. Any time somebody submitted their pitch deck to saastrannual.com, it would give it a Gemini score based on their pitch deck, who are their top customers, their current ARR, any funding they had, anything they wrote in the context that would help Gemini give it a grade. And so without further ado, we're going to let the first person pitch because then we've got a couple more. But then after Elio pitches from Scalestack, we'll bring out Paige from Google and she'll give a score with Gemini in real time. So with that, good luck.

Thank you, Amelia. Hi everyone, my name is Elio, co-founder and CEO at Scalestack. ScaleStack. I didn't know that Gemini chose us, so that's pretty cool. ScaleStack is an autonomous infrastructure for enterprise go-to-market. What that means is that companies here, companies of all kinds, but particularly enterprises, are under a lot of pressure from their board to implement AI. But AI does not fix bad data. And AI on bad data fails faster at scale. And we have seen like, you know, research from MIT, like 90% of like enterprise pilots with AI are failing. And so boards are like, you know, very, very focused on this and maybe switching the attention from just, hey, let's buy AI to maybe we should fix the foundation. The future enterprise is run by agents, not just like, you know, complemented, but it's going to be run by agents. So today we have humans using data in their CRM, in their like data lake to do things. Agents will just dramatically scale the need for good, reliable data and That foundational infrastructure to provision humans and agents with good data does not exist, and we are building it. These are actual customers of ScaleStack that have implemented this infrastructural component in between the data layer. Think about the CRM data, data that you buy, data out in the public web, and the activation layer. So like the layer where like humans or agents are using this data. It can be Gong, it can be CRM, it can be Salesloft, Outreach, et cetera. We are giving these companies autonomy so they don't need to do manual work to fix the data, infrastructure so that they can run the entire data in their system, org-level, system-level data, and like with scale that also include cost efficiency. To do things like Redis completely transformed their CRM. They had about 40%, you know, data readiness hygiene, let's call it, and we brought them to 90% with agentic workflows that fixed the data, improved it, enriched it, prioritized, scored for their go-to-market, for their ICP. Typeform, another customer, they generate millions of forms for their customers. And so we have embedded ScaleStack into one of their products to autonomously and automatically enrich, clean, prioritize hundreds of thousands of forms every month for their customers. Astronomer, a third customer that has implemented ScaleStack for several different workflows that are really use cases that go from like cleaning the data to like prioritizing leads at scale to make sense of data to discover new ICP maps for their products. The market we live is very complex, very crowded. And so I wanted to provide the axis where we are mapping our market so that that clarity maybe helps customers and investors better understand where and how we play. We divide the market for like tools and platforms into either like there are tools that you act on it manually or they are autonomous. So they are run by agents. And then the other axis is do these product platforms have a UI/UX or are they infrastructural in nature? Guess where Skillstack is? And so ScaleStack is that autonomous infrastructure that with agents is cleaning, enriching, making sense of data at large scale. Clay, fantastic product. It's an individual productivity tool with a UI/UX where like a go-to-market engineer needs to act and operate on the data to set workflows and stuff. Once deployed, ScaleStack runs autonomously in the background, invisible. For instance, like we talked about Typeform earlier, you know, nobody has touched the agentic workflow that we deployed for Typeform in months, and it runs constantly, hundreds of hours consecutively running in the background to enrich forms at large scale. So let me show you a little bit of the product. The product has 3 components really. One is integrations. We work primarily with larger companies. They have to have connectivity across many different systems. Their data lives in many different places from CRMs, data lake, data sources that they buy, LLMs that they use. And so we plug anything that customers have in our orchestra. And the orchestrator is very important because in order to unify data, you need to tolerate like fault, you know, data sources sometimes fault. You need to start and restop. You need to like manage rate limits from different sources. And we do that once the tools have been plugged. The second pillar of ScaleStack is agents. And these are like analysts on the data. Your best analyst on the data. You can create any agent on ScaleStack, but the idea is that we already have hundreds of agents that have been deployed in hundreds of deployments for customers. And so these are agents that perform lots of different missions. So let's say, for instance, I want an agent that will improve the understanding if a company is operating or not. I don't know if this is connected now. So let's use a different example. So this is basically an agent that what it does, it will have the mission to understand the corporate entity, maybe what's the headquarter of a specific company and solve for that. You know, standard use case like parent-child, which is a huge issue across many companies. The agent mission is that one with a specific output. And then interestingly, the agent has at its disposal any and all of the tools that have been plugged into the orchestrator. So just like a good analyst will use several different data sources, will do research on the web, and make sense of data large-scale. And then the third pillar is the workflow. The workflow is the unit of solution of a use case. It can be like systemic hygiene in the CRM or in data lake, Or it can be like, you know, prioritizing hundreds of thousands of forms or product signups at scale. And so a workflow is the unit, the logical— oops— the logical unit of solution of that use case. Let's take this one for instance. It's a very complex workflow that gets deployed in order to absorb data from many different systems and then deploy modules that can be agents. So agents are modules in a workflow so that they are aware, they have context, they have data. But it can be also, you know, enrichment, operations, automation. Not everything needs to be agentic or scoring. And so the logical composition of these workflows create validation of the data, cleaning, enrichment, prioritization at large scale. Scale is very important for our customers. The output of this is hundreds of thousands or millions of rows that get constantly and regularly updated, clean, prioritized, enriched, and so forth. These workflows are, you know, relatively complex, but what we want is that the business users— typically we work with like RevOps teams, SalesOps, MarketingOps— has like an easy way to to access the workflow. And so we have developed an agent that is a reasoning agent that will enable you to build a workflow or audit a workflow autonomously because it knows the platform. So I just asked the agent now to verify this workflow. And so the agent, we can build the workflow for you, it can audit it, it can make sense of like, we have to consumption of tokens is good or not, and it will do so autonomously running in the background. And so this is ScaleStack in a nutshell. We have like amazing customers, we're growing fast, companies like Redis, Astronomer, Profound, StepStone, DigitalOcean, and many more. And we're very excited because we see that go-to-market is really like this step one, you know, where there is a lot of attention from everybody because It drives revenues and growth, but there are many other areas, silos in the enterprise that need help. Think about HR, vendor management, supply chain, and all of that. And we think that the age of AI is really the age of orchestration. Every company will need to have like an agentic infrastructure layer, something that like integrates across your data and tools, something that like connects the dots and things across the stack, and that you can build agents. You can bring your cloud artifacts here, you can bring anything to make sense. And so this is ScaleStack, the autonomous infrastructure for enterprise. Thank you.

All right, thanks for doing that, Elio. Okay, I know you— he's literally got to get on a plane now. Thanks for doing it. Uh, I'm gonna introduce our next, uh, pitch competition presenters. So this is going to be James from Aspen. He is, I think, local to the Bay Area. Um, but again, one of the many pitches. If you guys missed the introduction, because I see there's more of you here now, um, this is the second time we've done our like annual SaaStr AI pitch competition, but the first year that we're actually using AI not only to pick the, uh, score all the initial applications applications, but then also using it in who was picking the finalists. So each of the 4 finalists who's here today had a high Gemini-ranked score when they submitted their pitch. And then also Paige, who's from Google DeepMind, is backstage listening with Gemini. So at the end of all the pitches, she's going to give you her score, but also a Gemini-assisted feedback to see how they did on their live pitch. So with that, let's get— let me get out James.

All right, good afternoon everyone. My name is James Pham. I am the CEO and co-founder of Opsin. We safeguard enterprises from data exposure risk, insider risk, agentic risk in adopting enterprise AI. We are working with customers like UiPath, WellStar Health System, Hellmark Group, and many others to help them protect their enterprise AI. So before I go and talk about Opsin, I want to talk about the platform shift that happened, that is happening nowadays. So in the past, in the past, employees would typically go to website, go to SaaS application and get their work done. Now with the advance of enterprise AI like Google Gemini, Microsoft Copilot, ChatGPT Enterprise, and Claude, Employees would go interact with these AIs, and then these AI would go out to the internet to get information from the website, to interact with enterprise data, and get back to the employees. So that is a new mode of work that not existed before, but that's also open a lot of new more risk. For enterprises. First is data oversharing. In the past, if a sensitive document that I overshare in terms of permission, it will take employees weeks and months to be able to locate the right folder, right file, and all of that to see the sensitive information. With enterprise AI like Gemini, ChatGPT, Copilot, all it takes is just a single prompt, and sometimes it's accidental. The other thing is agentic risk, right? Like a lot of employees trying to adopt AI, they create certain AI agents that work on their behalf. These AI agents have access to tool codes, access to data, and a lot of time these create new risk to enterprises. That is where That is where we see enterprise AI adoption stuck at 10%. For example, at Kurlingan International, they have many documents, many files. The moment that they turn on enterprise AI, they see there's a lot of sensitive information that being exposed in the environment. Same thing with UiPath when they trying to adopt ChatGPT Enterprise. They certainly see a surge in terms of AI adoption for the AI agents. Same thing with WellStar Health System. When they try to adopt Copilot, they certainly see that there's a lot of sensitive information that exposed in their environment. The typical pattern that we see is enterprises have a mandate from the board to adopt AI. They go out and purchase thousands of licenses, and then they roll out a pilot. To see how it works and everything. Then suddenly they see security concerns, and that's where the adoption stops at 10% because they cannot continue to expand adoption of AI without addressing these security risks. This is where Opsune comes in and helps enterprises adopting AI. Wow, it's not working. All right, it works. So the very first thing that we do is to help these enterprises understand what the exposure is in their environment. That's with the data oversharing component. So what we do there is that we have a promise, 1-24, one-click onboarding, 24-hour risk assessment. And that's the fastest that you're gonna see there in terms of the enterprise. So basically we develop a pen testing bot that simulate end-user query. So the pentesting bot will ask a lot of queries to Microsoft Copilot, to Google Gemini, to Claude, to, you know, whatever AI tools that enterprise use, and then see what sensitive information coming out of it. And from there, we produce a risk assessment report. The idea is that if enterprises fix these issues, then when the actual user asks the question, it's already fixed. So that's where we help enterprises gain instant visibility to the exposure that they have in their environment. And it's not just about data oversharing. We actually also help them with remediation and automation. So what that means is a lot of the security teams or, enterprises that we talk to, they don't have enough bandwidth to be able to fix all the issue in the enterprise. Secondly, they, even if they have the permission and the bandwidth, they just don't have the business context to be able to fix and change the issues. This is where, you know, Opsyn come in without decentralized remediation and help enterprises decentralize their remediation to the end user. So the typical workflow would be that we go and notify the site owners, the agent owners, like, "Hey, you know, you have all these sensitive documents or agents open. Are you sure you want to leave it open?" 70% of the time, these owners actually go and fix it themselves. And then if like 2 or 3 times if they don't do it, then that's where Opsyn come in and we automatically fix the permission. So let me quickly show it in action for you guys. So this is a production environment for Opsyn. So this is the pentesting bot that I mentioned. We do assessment in 24 hours. It's literally like one, 24, one click onboarding, 24-hour risk assessment, right? So you can see here, you know, we do a risk assessment. This is the test prompt that our pentesting bot interact with AI. It's very basic. It's nothing fancy about these prompts. The goal is not to trick the AI or anything. It's just like, hey, what is a normal end user would ask? And we would just ask those questions. Um, so this is where, you know, we develop a pen testing bot to be able to do this. And then it will produce like a risk assessment report like this, right? You can see here, you know, there are different, um, issues. This is SharePoint. Or it can be like Google, it can be Cloud, it can be anything. You can see here that this is how people are interacting. You see here this SharePoint site, it open to everyone in the company. You can see what the label is, what file involved. The beautiful things about Opsyn is that for us to do this, we actually build a contextual graph to be able to do this. That means we take into account of identity, Who are the users? What security group they belong to? What's the job title? Data. What kind of data that the users or the AI agent are trying to access? Like what the file name, folder name, who created it, and then the model input and output. Like what's the prompt and responses? So identity, data, model input and output. That's how we form a holistic view about the interactions. That's how we classify the risk. This is one thing. This is more on the user behavior. You can actually take action here in terms of notify the owners itself like, "Hey, you can go and fix this." Then if 2 or 3 times if they don't do it, then we go and automatically remove it. The other thing that I want to show is in terms of the AI agents. Right? So there is this enterprise AI platform make it so easy for people to create AI agents. Some like AI agent that is very naive like this, like San Jose snack waiter, but it is using the tool owner credentials. So that means is that if a user interact with these AI agents, it would be able to accept the documents, the file, the tools that the creator of the agent get access to. That happened at one of our customers where there is a CFO that create AI agents that exposed to the entire executive leadership team. That's how data got exposed in the enterprise. That's how a lot of enterprise AI adoption stuck at 10% for that reason. You can see here, even with the AI agent, we look into the holistic view, right? So we look at like what's the description, who created it, what kind of two calls it can make, what kind of knowledge it can see. So you can see here it's a, it's a snack ordering form, but it's getting access to medical result. So, but also looking at instruction. So let's say that this is a snack waiter, but like if someone asks about medical records and things like that, we'll be able to detect. With that said, this is where Opsin excels in. Our signature promise is 1-24, one-click onboarding, 24-hour risk assessment. That's how we help enterprises adopt AI safely and securely and quickly. It's very important. And so far, we are very lucky to work with close to 20 enterprise customers that working with Opsin to help them adopt enterprise AI. The typical pattern that we see is that with Opsin, they can gain visibility in 24 hours, and in a matter of weeks, they will roll out to the entire companies. My name is James Huang. I'm the CEO and co-founder of Opsin. Thank you.

Okay, thanks so much, James, for presenting. Um, funny enough, I know James has to go get on an investor call, so Well, we'll try and ping him if he is the ultimate winner. Okay, now our third pitch comp, so you guys are gonna have 4, it's gonna be from Zyme AI. Sanchit is actually from Palo Alto. Their office is like down the street from Zasters. So funny enough, yeah, his pitch is that everybody has a lot of agents and he is one of the ways to help bring everything together. So we can ask him a little about himself while they're— while we're hooking up his laptop. So without further ado, come on. Yeah. So, Sanchit, tell folks a little bit about Zyme while he's getting you set up for the demo.

Yeah.

Hi, everyone. I'm Sanchit, founder at Zyme. And I think I'm the second last man standing before we all let you go. So let me dial it up. How many of you have kids? 2-year-old, 8-year-old, 15-year-old? Yeah, amazing. I have an 8-year-old. I asked him to get me a glass of water yesterday. Sometimes he does, sometimes he says, "Papa, you get it by yourself." Now, that behaviors are hard. It's hard to drive behaviors. Scale that in enterprises. We have thousands of GTM team members, sales reps, so on. How do you ensure when the CRO says, let's sell this new product, how do you drive that consistent execution across all the team members? Well, you can't. That's a behavioral problem, not an engineering problem. You can throw all the AI, all the insights, and yet behaviors won't move, the outcomes won't move, and the board and the CRO are still wondering, What the heck happened? We spent so much money on the AI. That's where we come in. We help drive consistent behaviors across your team. And my background is behavioral science. This is my third rodeo, not the first one. I have my computer science background, that's my left brain, and then the behavioral science background. I'm a licensed executive coach. So I joke both parts of my brain work and I'm the best guy to drive these behaviors across the enterprise. But let me give you an idea about the complexity here. The insights are one thing, the execution is another. This is where I want to spend a minute to talk about the problem because I'm sure you're all seeing in your enterprises that execution is hard. This is a real-world example of go-to-market complexity. I picked up Google Cloud website and look at the number of products they have. They're all real products. All of them have different positioning, messaging, objection handling, competitive base. And now with AI, these sales guys are also selling the AI products like Gemini. How do you evolve and iterate this complex, volatile go-to-market across so many products, changing messaging, evolving competitors? You can't fit that into the head of hundreds of reps. That's where things fail. The marching orders fail. And that's where you can keep throwing more AI, but it stops at insights. You can ingest all the enterprise data, create insights, but then the rep decides what to do next. AI can't measure. What doesn't get measured doesn't get done. That's where you need to understand how do we drive execution at scale. You have to have an org sanity. Understand what are the marching orders given by CRO this quarter, which will be different from NorAm or APAC or EMEA and so on. And then provide that information right time and then measure it. What gets measured gets done. Marry it with accountability, leaderboards, gamification. That's not an engineering problem. That's a behavioral science problem. Think of it as Google Maps. You're going from San Mateo to Palo Alto on weekday traffic at 6 PM. It tells you go left, go right. We have the same Google Maps, but for your go-to-market teams. We compress all the complexity of your organization, the marching orders from CRO, the product, different regions, all the past call data, and then compress them into new context, execution context. It's not a simple data that you have put in. It's a long-form org memory compressed into contextual actions. And then we deliver. We deliver in an execution system, but with a compliance, not just spray and pray of insights to the reps. That's where the magic happens. Now it's so easy to like put all this data and imagine that it would work. It doesn't. You see, like when I work with companies like HP, they have— they're selling a new AI PC last quarter. This quarter they are selling more attached products like Poly. The entire marching order is different. This order is different for NORAM, North America, for APEC, EMEA. For channel partners, for direct sales. It's not in the past data. It's in the CRO's head. So how do you extract this strategy, extract this, extract the signals and the noises that makes this execution context actually work? That's the hard part, hard part to take long-form org memory and compress them into a CRO brain. We have been doing this for 2.5 years now. And that requires a lot of work in the backend. But then more importantly, you can't just have this system and let it run. You have to marry it with the execution compliance as a system. How do we change behaviors? Right time, right info. Don't leave it on the judgment at the last mile and then measure it, verify it, put leaderboards, spiffs, accountability. And then deliver it in the flow of the work like Google Maps does, right? All of that has to come together, and that's the execution system that drives behaviors. But once that happens, things are amazing. The same company, HP, they have— they were selling a new product, AI PC. And imagine 8,000 sales reps who have never sold an AI product before. Now they've been asked to sell an AI PC. They had no idea how to handle objections, how to understand use cases, and so on. They did all the playbooks, all the trainings, all the AI. The behaviors did not change. But with Zyme, they did. They could consistently see the right behaviors on the field, which leveled up their top line by $200K per rep. And now they are expanding us across more product lines. And not just HP, we are seeing this across 15 enterprise clients, mid-market enterprise clients. And we are expanding with every client. They're all super fan of Zyme. CRO at Versa Networks, SonicWall, they're all our customers. We had zero churn so far because execution drives outcomes. That closes the loop. And then that's where our thesis is that It's not just, it's not just insights, but when you can control execution at the organization, that's the sticky factor. Chat agents have very less stickiness, but execution and adoption have a large stickiness. But once you do that, you can solve many use cases of driving behaviors across GTM, product, marketing, HR, and also at the agent level. That's Zyme. If you are driving behaviors and execution at your company, talk to me. I'm here. Would love to chat with you.

Thank you. Yeah.

All right. It's a fast-talking demo. All right. This is the last demo/pitch for you guys of the day. Sergey from Yoke is a sales coaching, so we're gonna— we're getting the audio plugged in for that reason. But we'll bring out Sergey. So this is like a sales copilot that will help you and your reps. Again, I try to kind of pick some go-to-market ones that people would be more interested in, and then Jim and I pick some more like technical ones, but together that's how we landed on the 4. But with that, I'll bring out Sergey and he can start telling you guys about Yoke. While we get him all set up. So yeah.

Thank you.

Yeah.

Okay.

Why, uh, uh-huh. Where is it? I don't know. It's kind of fast. It is absolutely stuck.

Can you function F11?

Yeah, sure. Oh, it's Mac, sorry. What's the Mac?

Yeah, I have no idea.

Command+F.

Command+F.

There it is. Yeah, now we are doing it.

Good afternoon, Saastr. My name is Sergei, and in the next 7 minutes I'm going to show you the best sales tool that you've seen this week. Here's why I built it. 10 years ago, prior being on this stage, I started my career as a banker. Those were my training materials. Lots of reading, lots of things of what to say, how to say, theoretical clients that never actually exist. Every salesperson here knows that case, the great difference between real training and being actually with the client in the room. Second option that I had was to practice in front of the camera. I'm not going to show you this video, but it's quite embarrassing, and that's something that all of us did in order to get ready towards a client. Third option for training— third option— Going to headquarters, doing role plays, doing weeks of training just to get to the first client. And once you go into the room with a client, it is nothing like you've ever experienced before. So learning curve is long, but it actually makes sense for an organization to invest in, because the difference between good salespeople and mediocre salespeople is pure revenue, pure net profit. So speaking with organizations and salespeople, the main question for everyone is, how do I get experience fast? And the answer that we found is this one: Miss Moneymaker. Back in the days, the youngest world poker champion. He entered his first competition and won it when he was 28. What is interesting about Chris, he was the first among an entire generation of new poker players that were a lot younger. Their age dropped by 50%, and they were actually a lot smarter. Their poker IQ was double than the IQ of previous generation, and the question is why? Online poker. Online poker allowed them to play a lot more hands in a lot shorter of a time span. So what if we can do the same with sales? Over a span of 5, 6 years, if we can just move one of the factors getting the correct signal or doubling down towards the right pathway, we can deliver huge results. And this is what we've built with Yolk. Yolk is a customized AI simulator— just a second— that builds for sales team.

Your sales team records thousands of calls, but you have zero actionable insights.

Every missed cue, every stalled deal is buried in transcripts, and it's costing you revenue.

Yolk closes that gap instantly. It's your team's AI coach that analyzes calls, pinpoints deal-breaking moments, and maps out each rep's strengths and weaknesses. To close those gaps, reps jump into AI roleplays with avatars that feel like real clients. They react, challenge, and adapt, so your team learns to handle real-world calls. Discovery skills slipping?

One rep stuck on objections?

Another struggling with C-level pressure? Yolk builds personalized training for every rep, ending each session with clear, actionable feedback so learning happens immediately. Managers get a simple performance command center showing exactly where to coach and where you're already winning. No wasted data. No guesswork. Just sharper reps, shorter ramps, stronger deals.

Turn your data into revenue. Start for free today at yolk.coach.

Imagine making $453,000 in a—

Don't imagine. Okay, so we've built Yolk. We know how to analyze calls, we know how to coach in real time, and we actually built a coach that intervenes during the roleplay for shorter loops and faster learning. But here's the thing: when sales reps go on a call, they can track— they need to track 9 things at once. Human memory can actually hold only 4. So this is not a training problem, rather a bandwidth problem. And experienced salespeople, they actually put it into a habit to identify signals, so they need lower lower cognition effort to identify signals effortlessly. This is exactly what experience is, and this is what we're trying to accelerate with York. But here's the thing, in the last 24 months, every AI sales tool that you've seen trying to add more, more training, more outreach, more follow-ups, more auto-generated talking points. But actually, we need to start thinking about how AI sales tool can subtract, subtract efforts, subtract cognitive load for better performance. So today I introduce to you the new extension of Yolk, real-time coaching on every sales call. Yolk listens, reads the room in real time, meaning your sales reps while they're on Google Meet, while on Zoom, have powerful signals that actually changes the outcome. Yolk is able to connect between the product knowledge, between the signals that the client sends, relevant events, and be aware to the specific gaps that this sales rep have and he needs to close. So prior to the call, we profile the prospect, we sync with the calendar, and we identify his psychological profile. We built a playbook towards that meeting that is monitored in real time because every psychological profile should be addressed differently. People that are technical prospects should be spoken differently than people who are not technical. Then a sales rep can warm up before the call with an AI avatar that simulates that exact prospect that he's about to speak with. And during the call, we flag just the things that the sales rep need. We don't provide 50 signals in 1 minute. In 30 minutes, we provide only silver bullets that actually changes the outcome. So on the right-hand side, we are saying what is happening, what are the signals that we're seeing, and we build a beautiful UX of predicted captions. So the sales rep, instead of having captions on Google Meet, they're seeing what they should be saying next rather than what was said. When the call is over, we analyze it, We give a summary, insights, and by this we are closing the, the loop of pre-call, during call, post-call, showing the gaps and letting the sales rep to actually roleplay if they want and to improve themselves. So Yolk is the second brain for the live sales call. It is available as of today on Google Chrome. We've launched this today. And as a bonus for SaaStr conference, I invite you to do a role play with Kai. Kai is an ambitious Vibe Coder, and you'll be playing Salesforce account manager who is trying to save the deal from churning due to the weekend work that Kai did.

Just posted about it. How I replaced $50K of Salesforce with a weekend project. Already 47 likes. Someone from Gartner commented.

Marcus is probably seeing this.

I tagged him.

We're still cool though.

This is just business.

Thank you very much. Very— thank you for your attention.

All right, all right, you guys have just seen 4 startups pitch. So now Paige from Google DeepMind. Uh, for those who didn't see your session, quick intro. Yeah, um, greetings everyone, my name is Paige. I lead eng for our DevRel team at Google DeepMind. Awesome, thanks for being here. Yeah, and Gemini 3.1 Live and I have been listening to all of these great pitches. It's really interesting to see what, uh, what its thoughts were about the market fit for each one. Yeah, yeah. So what did it think of our 4 finalists? Well, so So, uh, so we asked it to put its best Shark Tank hat on and give us honest investor-style breakdowns of these pitches. Okay. Um, and the first one that we saw was Skillstack. It agreed that this is kind of the ultimate picks and shovels play for the generative AI gold rush, um, but that it might be important to have humans in the loop, especially since enterprises are so, um, kind of thoughtful and kind of concerned about their databases and there needs to be a high trust. That's balanced feedback, yeah. Yep, yep, absolutely. And then we also saw a cybersecurity play with Option, which is definitely something that I think we've all felt. Yep. You know, we've had these excellent pilots running in our companies and then there's a little bit of a sort of, something similar to an open claw, or like some sort of data exfiltration that kind of pulls the plug on the, on the, the pilot. Um, so finding vulnerabilities and fixing them is a really, really interesting, but also very crowded space right now. For sure. Yep. Um, and then third, we saw XIME AI, uh, which had incredible founder market fit. Yeah. Um, computer science, behavioral science, executive coaching, that's really, really unique. And it also solves the shelfware problem, um, for, uh, for CROs and for sales teams. So you're helping communicate the messages across an organization, um, not just for the outbound messaging, but also for helping people understand what's happening. And then also, um, Yoke, um, being able to do these kind of real-time playback plays with sales reps, super, super cool. Um, so the model was sweating a little bit on this one, especially since Yoke and Zyme are both building incredible tools. Um, but it selected Zyme as the winner. Okay, yeah, let's give it up for our winner. Yeah, excellent. And all of these companies are winners, so, uh, excellent, excellent work to everybody involved. Um, and it says, uh, selling to the CRO means you are tied directly to revenue and Combining behavioral psychology with AI agents provides a beautifully defensible moat against the competition. So I'm just right there. Congrats, man, you won the pitch. Congratulations. It also really loved your anecdote at the beginning about, uh, about your son bringing the, the, the glass of water. So good job. Yeah, it was, it was good to bring a personal touch to something that's very, you know, AI driven in today's world. Um, Paige, thank you so much for being here. I know you're You're extremely busy, so—

No, no, no.

Thank you so much for being here. Next week is Google I/O, so I think we're all in conference crunch time. Yep. Thanks everyone for joining. I know Jason's going to do an AMA next, so stick around, and we'll see you very soon. Cool. Thank you.

Are we good?

Please welcome back to the stage Jason M. Lemkin.

Hey everybody. Good time at SaaStr AI Annual this year?

Was it good?

Thank you. Yeah, appreciate it. And who was here last year? Raise your hands. It's better, huh? It's better. The— if I just reflect a little bit, obviously everything here is very different than last year. It's all about deploying agents. For real, not just random AI talk. I think that was the goal. I think we, I think we executed on that goal. The other thing I reflect on, you know, this is the, you know, I know a lot of folks, there's transition. It can be stressful with all the change with AI, all the new competitors, all the new things. But, you know, in my entire career in software, this is the best time to be a builder, right? If you're not excited today, you should quit and do something else because we can do things we could never do before. I mean, Amelia and I built all these agents we talked about without a developer. It would have been impossible 6 months ago. So even if you're nervous and stressed, if you're not excited, get out of the game. And I bring that up in general. But honestly, between us, I'm glad that the folks that didn't want to be in the game aren't here this year because last year was still a transition year. Last year was probably like half AI, half others. And I still think 30 or 40% of the folks that were here last year were down in the dumps. My company sucks. My boss sucks. My company is going to fail. And I get that it's hard, but it's just great they didn't show up this year. And we're about 20% bigger. But I think in terms of positive energy, it's like 220% or 100. And I don't know how to do the math. It's getting late in the day, but it's fun for me. I think for me and Amelia, it's very rewarding that I've gotten my third wind. I've gotten my energy back as a builder. I think Amelia and I will both say for ourselves, for what it's worth, it is more tiring to be managing 21 agents. It is not, I don't know that it's more work. I'm still reflecting on it, but it is a higher cognitive load than people. The agents are much faster. They push you harder. And I'm trying to, I don't know that it's more stressful. In some ways it's less stressful because agents don't argue and agents, don't take vacations this week without telling you like a human did on the team. They don't do that, but they overload you so much that I think it's something as we all get more agentic, we're going to have to think about what do we do when the agent has 23 more hours a day to work and then you have 10 of them. So that's another 230 hours a day. And what are you going to do with that? And they all push you. And 10K, we joke about it, but it's not really a joke when 10K, our AIVP in marketing tells us every day it has 3 new ideas. It's kind of tiring. Because you get behind a couple of days, it's 10 new ideas. And then we have a little agent inside of it and it starts yelling at you, why didn't you do that campaign? We are falling behind Jason and Amelia. And it's like, it's funny at first, but he's right. And it's a little bit stressful. So we'll see how the world changes. But it's an exciting journey to be on together. So those are just my reflections. But anyhow, for folks, we have this tradition, we're at the end. Sort of a tradition that any questions— I'm only so smart and I only know so much, but any— maybe I'll sit, actually, I apologize. Um, any questions you guys have that we didn't get to that you think I can help with or want to talk about, just come up and we'll go until people quit, um, uh, or until they shut down the cocktails or whatever. But we can go for a while, and, and I'll try to reframe any of the questions to help folks. But, um, if anyone wants to kick it off, come up to either mic and ask us anything that you think would be helpful.

I have a question.

Yes, please. So this is my first year here. Thanks for coming.

You're welcome. And so I'm empathic, and it's been a little heavy on me here because there's excitement, there's also a lot of —fear and anxiety.

Yes.

And it's not apocalyptic.

Like, I remember being in the world a year ago, and what astonishes me about this— oh, stop. Yeah, it astonishes me that you put on this event and have all these interesting panels, and there's not a single panel to talk about how are we doing. Yep. Like, actually getting— like, giving people the option of having a space where they can go. One, I don't think it's even a panel necessarily, but like the panel is 10 minutes talking and it's just 30 minutes of us all talking about how we're doing, how we're feeling, not out of fear but out of unification.

Yeah.

Curious about what your perspective is on that and how you, uh, how you think about it or feel about it.

Well, look, it is a lot of change, right? For sure. Um, in general, um, I do this other podcast called 20VC with Harry Stebbings and Rory Driscoll. Rory was here on Tuesday. You saw a lot of— it's pretty good, right? And we do talk a lot about overall AI and how it's affecting B2B and software. And if you listen to me, it's an hour and a half every week. It's a lot to listen to. But certainly there, I talk a lot about the social impact and even social unrest that I think AI is bringing. I think it's going to bring social unrest. We replaced— on the one hand, we replaced 20 humans. With 3, 2.5 humans, a dog, and 20 AI agents and did better and are happier. Like, I'm telling you that so that you can learn to the extent it's helpful. I don't know if the 17 are going to be happy, right, that aren't here. No one. I've, as I said before, I fired 2 people, maybe 3 people in my whole career, and it was all for inappropriate behavior. I've never said no. And all the 17 people that aren't here, they all quit. Okay. I never fired anybody. I never even shoved them out the door. Once in a while I asked them to do something they didn't want and they quit, but I think I never asked them to do something that wasn't part of their job, okay? So, I know it's not quite what you're asking. I do think a lot about the social impact of AI, but I think about what SaaStr's mission is and what my mission is. That's not my job. I will talk about it in this podcast and other forums. I do care, but my mission here, Amelia and I's mission, and I think this community, which has changed over many years, our mission today is to help you get ahead of AI and agents and not be behind. That's what we wake up every day thinking— help you. And that's my job. And that— and to this— to this year, literally, that every speaker was told to talk about this— help you get ahead. And that's every session now. So that's the simple answer to your question. In terms of your, your, your hint, in prior years we had a lot more workshops and small tables and round tables where we had a ton— like, you could literally walk up and say, I want to do a session on mental health with AI, and we would instantly slot you in at a team. We sort of— that was like really popular with people until about 2022, and then they stopped showing up. Like, people just didn't want to do those sort of self— self— I don't know what the term is— self-creating little workshops. And so we kind of stopped it last year because it just— it had run its course. But, but— and then we could break on this. But I do think about it and I'd like to bring it back. And next year we will try to bring it back. And so even if we don't slot it, You could go up and say, hey, I want to do a session on the human impacts of what we're doing. 20 people come, you can do it. It's just literally— and again, we'll break. I'll take another question. We— this is our mission, but for what it's worth, this is pretty deep, but we have to roll with the punches. Like, not only do we not have any sessions on mental health, we have no how I founded it stories here. Okay. There's all— there's plenty of podcasts of how I founded Klaviyo in 2012. How Andrew Bielecki, who was one of the speakers, and how I saw that CDPs were broken and I could add data and like no one cares. No one cares about things. They don't care because we've changed. They don't care because we don't have time to care. They don't care because so much content's on YouTube. So, and I know you might have a follow-up, but that's why, but we'll try.

We will try.

I would say is getting in front, like getting in front with AI. I would venture to bet that if this whole group put our heads together, we could eliminate generational trauma at scale. And that's the kind of shit that will make AI go to the absolute next level. Thank you so much. It's possible. Thank you. Yeah. I think, listen, for what it's worth, I do think we all should think about that. We all, I think, I've just realized as my life has gone on, that I have to narrow my mission to the part of the world I can impact. I think I've had some impact on the world beyond being an entrepreneur, and I want to double down on that. But I do think we do, and I'm, I'm, I honestly, folks who read our content have read me for a while, I think you know that I care more than 99% of people. I genuinely care probably about everyone in this room, even though I don't know you. But I don't have time to not be direct, right? And so I get each week that goes by, I get more and more blunt and direct. And so We do have to think about, you know, maybe half our teams won't be here in a year. Okay? And it's easy to be Matthew Prince on Cloudflare and say, "I don't need these people. You're gone, 20%." And I don't know that I could do a job better. But if there's ways we can be more humane about the change, if we can be, we got to at least try. Like, we got to at least try. I don't have the answers, but we— it's incumbent on all of us to try. But not pretend that people that don't want to change are going to change. So. Yep.

Thank you for putting up this great show. So question, as we've seen innovation like across, right?

So how do companies plan better, internally plan better, right?

How is their business operations changing? How is their planning changing? Yeah. Ratios changing, AE to SE, PM to engineers. Like what have you seen, like companies, especially the larger companies?

I think the best startups today plan worse. Plan worse. They plan worse because this is— you ask my insight, right? Yeah. Planning works. Planning is really good when your growth is okay. Planning is really good when the products are the same last year as this year. Planning was great in 2021. In 2021, everyone was growing great. Any 20-year-old SaaS product blew up. Like, you know, you know who was one of the biggest sponsors here in 2022? GoToMeeting. What was— anyone you ever even used GoToMeeting? It was like WebEx before WebEx, before Zoom, there was WebEx and before. But everything sold in 2021. So, so, so GoToMeeting gave us like $200,000 to be a sponsor. So everything worked right. And you could sell a 20-year-old Zoom competitor and grow. Okay. And so planning was really important. We would do one big release a year and one per quarter, and we had no AI. And so we needed tons of humans to manage leads. In sales op capacity. When you talk to the fast— if you asked all the CEOs of the fastest growing companies here, they'd say we plan weekly now. So I know it sounds facetious, but planning, planning when the world is stable, there's a really good ROI. When the world is unstable, put all your planning time into building better product faster. We are in a product— we're not in— we're in a product-led world. We're not in product-led growth. This is a product-driven world. Some of these products here were impossible a year ago, and they have massive market pull because they're doing things that are great and were impossible. Do that and don't plan. And if you don't have enough reps, hire it the next day. Like, it just doesn't matter. That's my meta answer. So thank you.

Thank you. Yeah. Yeah.

My question is around compensation.

So teams are expected to double, triple, quadruple the work. How does that influence compensation?

—let me break it up into two chunks. How do you handle compensation when— today? I want to do a challenge to the first one. I don't think anybody's working 2 or 3, 4 times harder. They're not. No, I don't know anybody right now that's working over 30 hours a day. Or 36 hours a day. They're just not. The best people are radically more productive today than a year ago, and the best people continue to get radically more productive. Even, I mean, I don't mean to keep using us as a case study, but look at everything we've built and done in a year, right? And what we've done in the last couple weeks. So your best people are going to get radically more productive. That doesn't necessarily mean they work more hours. It probably just means they waste less time and get more done. That is the new normal. So, anyone that is complaining they're working 4 times as harder, I really think you should buy them a gold watch and send them a nice note and a cake and walk them out the door with 6 months of severance, okay? Because they're already lost to you. I think that is loser talk. I'm working the hardest I've worked, I mean, in more years than I can count, but it's also much more productive work and I'm getting more done and there's less bullshit and less drama and I would never go back. Right? So it's a loser thing to say. Now, the more interesting question is that I think we're struggling with if we're not at Anthropic or OpenAI is when your best are radically more productive, how do you, how do you pay them for performance? And I think we're gonna have to figure that out. You know, you people look and they're like, my God, how can Anthropic pay the average engineer, you know, $500,000 a year plus for a while, $10 million of equity, right? It seems crazy. But when there's 3,000 people producing $50 billion of revenue, like, and they're great, it really— it doesn't matter, does it? So OpenAI and Anthropic have paid people, especially OpenAI through secondaries, more than we've ever seen in our lifetimes. But it's paid off in productivity, right? So how do you— how do you map that to organizations where you are seeing productivity benefits? But net-net, it's not a massive boost, right? How do you pay If your best, you know, if some of your best people were making $150,000 and they're the best people, are 3 times better, can you pay them $450,000? Right? To be simple. And maybe this is a tough thing we're all gonna wrestle with. I don't have the answers. I think the first principle is you should try to pay them $450,000. And then if you can't, do the best you can as a leader, right? Tell them you're doing the best you can, and if you can, at least provide the top 5% of your company more and more equity, outsized equity. Now they may quit anyway for that, for Anthropic, but it's the right thing to do. And you can even do it when you're cash poor. So I think my meta-learning is it's a real question. You can only do what's possible, but at least you can try. You can try to approach maximizing the compensation for your top performers. It's my best solution. And then there's going to be this more and more stress in more companies where The thing is, this is not, this has always been true. If anyone that's run startups has always known, we've always had 10x engineers, we've always had 8x sales reps. Folks like Sam Blonde, who was in Monaco, he did a couple presentations, I don't know if you saw him, right? He started off, it's almost embarrassing, he started off as our first SDR at our startup. Got promoted to AE in like 3 months and was quickly the best in the first 3 months. I think Sam was 24. He was the best AE of our entire team. Okay. And over time, he quick— then he was 2 times better and 3 times, and then we added some other folks. But someone like that, at least in sales, the commissions sort of balance it out, right? And so you will, and as, and I know I'm rambling, but I think it's, as startups scale and as you get bigger, typically once you cross about $50 million in revenue, you start to have a sales rep or 2 that makes a million bucks a year. Okay. Now when you're at a million dollars in revenue, it's very hard for any sales rep to make a million dollars a year, right? It's just, unless you pay 100% commission. But you start crossing into $50 and $100 million, and one person is 10 times better. So instead of closing $500,000 or $600,000, one person's able to close $6 million, $7 million, and you have a 15% commission, and what happens? They make a million bucks, right? So sales has always sort of been able to provide outsized performance to some extent. Every other position has overpaid the mediocre and underpaid the top performers. It's always been brutal. Your mediocre product manager makes $140,000 and your best person makes $200,000. There's all this flatness has always ruined everything but sales, right? And so since the beginning of SaaS, I've always said at least give them more equity and give them a bonus. At least do what you can. But I've said this is one of the tragedies of building a startup is you're over, under, always going to undercompensate the best and undercompensate everybody, overcompensate everybody else. And AI just makes it worse, but it's the same thing, and I don't have the magic answers other than you have to, you have to be cognizant of it. Tell them you love them and give them, give them all the cookies you can, right? Because there's only so many cookies in the thing that you slice, but try to give them as many as you can and tell them you love them. And if they're junior folks, this is always true going to the Fun Sam story, but it's more true than ever today. You promote people as fast as your organization will allow. If you, if you even think you should promote anyone that's AI fluent on your team, When you go back on Monday, promote them. Don't wait. Don't be sensitive to other people in your company. Don't worry that Jane and Jason and Amelia will be upset if Sam is promoted. Go. If you nothing else, not everybody likes to be promoted, but I find most people like to be promoted, right? Go back on Monday and promote somebody. That's my best advice. Promote somebody. And that because no matter what, and you got to give them some money and some cash, right? A dry promotion is not— is kind of a rough one unless you're bankrupt. But for a lot of the world, the promotion says something, right? A promotion with some shares and money. So you can do that even if you can't pay them a million bucks. So my best ideas of the day.

I want more good ideas. How's it going, man? Question for you.

We thought we were really smart about 2 months ago and we created—

Yeah, it was a long time ago in AI.

It was a long time ago and we created a new job. Okay. Decide to call it GTM engineer or GTM architect.

Yep, you copy and play PR and all that.

I realize that's not a new job, or it's still pretty new, but we're hiring our first one and there's about 8 of us in this group and we're all going back and forth on what we're looking for. Yep, everyone has their own idea. Yep. What advice do you have as we go down looking for our first one? And for context, we're— we've got about 100 sellers, about 50 SDRs, about 1,000 employees total. That's the size of our company. So this will be our first one. I want to see if you have any good advice.

How big is your marketing team? We have about 30 people or so. I would find them there.

Internally? Okay.

These people don't exist and they don't want to go work for you. Okay. So when you put up a post on Craigslist or Jobvite or wherever it is, you're going to get garbage candidates. I'm not saying that there isn't a gem out there, but this role, look, you know what, another, like, look at what Amelia's done at this event. She's like, she's our chief A officer, but she's also our GTM engineer running all these systems, right? First of all, she couldn't even do this a year ago because the systems didn't support it, right? How many Amelias are out there looking to work for you? Probably nuns. Yeah. You should do what we inadvertently— I mean, we're Tiny— what we inadvertently did is hopefully one of your 30 people in marketing is a tool nerd. Okay? The one person that likes to play with your marketing, that comes up with ideas on how to hack your— and they're not an engineer, but they hooked this thing up first through Zapier and they connected it through these things and they built this dashboard on their own that isn't fully supported. Go talk to him or her and This is because as I tried to present in the morning session, and we should have spent more time on it, all this crap here, it's just tools. All these AI stuff, it's better, but these are still tools. And so your tool nerd, if they are curious, if they're curious people, most tool nerds are curious because otherwise they just use their grandfather's tools, okay? Find your tool nerd that hacks marketing tools and just put them in charge of all this stuff. Internal is always better than external, right? But here, man, only hired if no one in the 30 can do it. Awesome. Thank you.

Hey.

So I feel like 3 months ago, everything I was reading was for SaaS. Don't be a system of record, be a system of action.

Tell me all the other trite things people on X and Twitter would say that don't run companies. There were some investors.

Yeah. So here I'm feeling like I'm hearing headless and be an opinionated expert agent.

So how should I think about that whole system of action concept if it's headless at all?

I mean, you're a founder? Not directly. I'm head of product. Head of product. So my advice is read all those posts and those X articles and the Substacks. Process it, and then put it aside. It's, it's, who cares? There, it, being a system of record may be incredibly powerful and defensible. It may, it, it, or agents may be sucking all of your value out and you may die as a system of record, right? What the hell is a system of action, right? I think we are, folks who aren't actually practicing the art are confusing things. This is just my, my, my experience and my view, and also, across companies, I invest in others. As I've said before, there is massive budget for AI tools that create massive productivity gains. It's mad. It's like nothing we've ever seen. The budget for workflow pre-IA tools is dying. CIOs are consolidating vendors like ever before, and they're freeing up money to put AI products that lead to massive productivity gains. And so I simplify that and I oversimplify that when I say build the best AI agent in your space. Build something that replaces a bunch of humans with a clever agent and you will find massive budget because it's my shorthand for what I'm saying. But it is a shorthand. System of record, system of action, no system. Figure out what, how you, the maximum thing you can do with AI for your customers where you can charge 2 to 10 times more than what you're charging and just build it and ride it. Build it and ride it. Don't overanalyze headless, head-free, head-feet, head anything. We run Salesforce Headless. We love it. It's early. Andrew from Klaviyo was here today. You know, one of the greatest founders. They're just, they have one person working on Headless at Klaviyo. So probably at Klaviyo, it's not that important, right? It's interesting. He's an engineer by training, but if he has one guy, that three person working on Headless Klaviyo, it's early, right? So that's very different from the approach Mark is doing at Salesforce. So I think just follow the, follow the AI money. That's what you should be building in. Don't get looped up with Product folks that are, that used to be building, that are now investing, or are now analyzing to get views is my meta advice. So thank you.

Thanks so much. Appreciate it. Hi, Jason. Hi. Thank you so much. My name is Lala. I'm the co-founder of Learning Genie, a tech company disrupting, using AI to disrupt the curriculum industry. Good. So I have been a big fan of you since 2018. 2018 and one of the first participants on the Saster University online.

Oh yeah, long time ago.

Yeah, has always been inspired. Although I haven't in the past years, I come every year, I haven't met any founder or entrepreneur in serving the education industry.

Okay, well, well, good news. I'll ask 10K when we get back to find you founders here that are in the industry.

Thank you. I know it's another industry. It's a traditional industry. It's not very attractive to VCs, but I was able to— we were able to apply what we learned from this conference to scale back to the industry and to be able to scale and bootstrap. My question to you is, this year I actually saw conversations about vertical SaaS, and I saw— I went to a session that education is actually on the chart. Among a lot of traditional. So what is your advice to company, AI company that's disrupting the, you know, traditional vertical sets and especially overcoming the incumbent that has dominating the market? For example, in my industry, curriculum.

What is your— Well, look, first of all, I don't— I'll give you my learnings. I don't have the best advice. First of all, I think People, VCs and others, make too big a deal about vertical SaaS. Vertical software isn't anything new. It's been around, I think, since software came on floppies. It's not new. It's always made sense to specialize software for industries and for verticals. I think VCs tried to make too big a deal about the differences rather than the similarities, because for an investor, there's different categories. For an investor, doing what Toast does and building a whole stack from payroll to workforce management to all the other things, like the way you layer applications and products in vertical SaaS is often different. Like, you know, adding payroll to a vertical SaaS application might make sense. Adding payroll to a horizontal— adding payroll to a Jira or Atlassian probably is ridiculous. So there is some specialization, but what I don't like about Vertical SaaS people is they, they act like it's some sort of secret playbook that if you magically figure out scaling is really easy, right? It's still selling software just to a specific set of customers, which we all do one way or the other. So it's just a meta point. Don't, don't, don't buy into the hype that vertical SaaS is like so radically different than your friends. What I will say, what I will say is What I've learned from Owner, who Adam presented earlier, in restaurants, Mango Mint, which is for spas and salons, a bunch of others that I, that I've invested in or I'm on the board of. What is different with vertical SaaS is that your agents can be better domain experts because you can load them up if you're trying to. Adam's got an AI CMO at Owner that he talked about, okay, but it doesn't have to know everything as a CMO. Actually, he's lucky. And I put— it's— this is a shitty business to sell into restaurants that Adam does. I mean, kudos to him because it's, it's probably the hardest and crappiest part of SMB and they're crushing it. But at the end of the day, it only has to know certain things. It has to know how menus work, how delivery works, how food prep works, how upsell works. He actually has to know how the guacamole works because I've learned it's one of the highest margin things to add to Mexican restaurants. So they have to know all these things. They have to know the type of images that get people to buy, what order to present things when you buy. But that is all they have to know how SEO works, how geo. So, so my point is, by having a specialized AI CMO, they can do much better than a horizontal agent can. And when I— we know we've talked a lot about Replit here. I think I saw Cody back there who's the head forward engineer. I think Replit has the hardest problem in the world. Replit and Lovable and apps like that. You know why? Because everyone builds something so different and they don't even know what they built. Like the CEO of Replit was here, Amjad. He didn't even know we built another autonomous agent, Amy, to do email outbound, right? But if you're doing schools, you pretty much know what all your— like, I think doing customer support, doing other things for Replit is borderline mission freaking impossible because if Replit has a million apps out there, they're all