AI AT SCALE 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. Our program is going to begin in a few minutes. It's time to find your seat.

The world has changed. AI is the accelerant. The amount of disruption that the software industry is going through now is unprecedented and it's accelerating. Innovation's everywhere today. SaaStr is all about being positive. It's all about helping. It's about doing it together. The reality is we have to move faster. So what I'm telling every CEO I meet You got to refound your company. If we could start over, what would we build in our space? What would we build today? How can we do it twice as fast? And just go find a friggin' way to build it.

Please give a warm welcome to the CEO and founder of SaaStr, Jason M. Lemkin. Woo!

Hey, everybody. Can you hear me okay? All right. Thanks, everybody, for coming. Who's been before? All right. Solid show of hands. Thanks for the committed. This is going to be— hold on. Let me see if these slides work. Can you guys bring up the slides? Maybe? Hello? Always the first bug. Can you guys bring up the slides, please? Hello? Should I go back in? Come back out. Hello? Freeman? Team Freeman? In the back? Hold on. Give me— I'll be right back. Hold on. Hello? Ah, okay. Good. Okay. Okay. Always a few bumps. This is— well, everything's optional. This session is super optional. This is an extra. It's starting early. It's 1. If you are— if you have deployed 100 agents yourself, if you have 8 Mac minis at home running OpenClaw, don't go here. Go have some lunch, meet some friends, go to— or you can go in half an hour. The chief product officer, Vercel, is kicking off the whole deploy summit at the other stage. It's all go-to-market for AI and agents. It's going to be great. And from there until until, until the end, until we have poker and Prosecco. But if you haven't done it yourself, even if your, your, your company, your product has an agent, a chatbot or whatever, if you haven't done it yourself on the other side, if you haven't built an agent, a bot or something, that's all this is about. We're going to, we're going to pick the simplest possible way so that if you have your laptop or even you can even do it on your phone, you're going to build the bot here while we're here before we leave. And you can train it. And what we're going to use as an experiment, and then with whatever time we have left until 1:55, if folks are still here, I'll take any questions on this or anything really that you want to ask. But a lot, if you haven't done it yourself, a lot of the products are so exciting. Go meet with all the vendors here. It's incredible what's changed since last year. It's incredible what's changed since December. We have products that will set up appointments for customers for you. We will have products that screen everything. We have agentic CRMs that will do everything for you. It's all here. I'm gonna talk with Amjad, the CEO of Replit, about how we built an AI VP of marketing and AI VP customer success. And Amelia's gonna teach you today how to build one your own. I mean, since last year, but we started out our own AI agent journey last year. Last year we had one. Not 21 agents like we have today. We had one. And I use this product called Delphi AI. Fire up your browser on your phone or whatever. If you can go to Delphi AI and set up a free account, it's— I picked this one. There are other ways you can do this. There's other products, but this one is free and easy. Okay. It's a good combination for starting. Okay. And you're going to build what Delphi builds, clones. So there's a clone of me you can talk to. There's a clone of Lenny. There's a clone of John from TVPN, there's a clone of like sports enthusiasts and influencers and it lets you do that. But actually you can put anything into this. And so a little maybe 14 months ago I built a clone of myself and at first it was folks could just ask a digital version of me questions about SaaStr. What would you, you know, who should I hire for my first SDR? What if my VP of marketing doesn't want to do events? What do I do? And it was fun to have a digital version of Jason. And then I started adding other content like our SaaStr annual website and our prospectus for sponsors. And I know this sounds Captain Obvious, like 15, 16 months later, but I was shocked that the agent could answer better than most humans that were helping us could. It knew our prospectus. It knew what a gold sponsorship cost. It knew the difference, how much tickets cost over time. It knew the basic questions. It knew what the Wi-Fi was. And so I didn't realize how easy it was to train an agent. And then we even had a sponsor that bought a $50,000 sponsorship just interacting with Digital Jason. So all I want to do, and then we can do all the questions in the world, is have you build your own, and then you can play with mine, and then we can look at it together, and then we can talk about where it's going. And again, apologies for folks if it seems too basic, but I found a lot of folks are using tools, playing with tools, buying tools, watching what their team does, but they haven't done this one step themselves. So this is all I want to do. So I'll go to the next one first. But first, just take a second, go to, again, uh, go to delphi.ai. It's better on your laptop, but honestly you can do it on your phone. It's free. You'll put in— you can OAuth in with Google, and then I think you have to give your LinkedIn, and it should be good to go. You should have a blank clone of yourself. Um, and if you've done that, if anyone needs a minute, delphi.ai. Um, I'm going to teach you how to clone yours, but I want to show you mine on the other side. So the next thing I want you guys to do is I built a new version, a new DigitalJSON, DigitalJSON 2 Prime. So it is completely blank. As you can see, there's nothing in it. Okay. With, with, except a little bit of Saastr content. So what I want you to do, if you want, go to delphi.ai/json-demo. It's a new one and you'll see me, another version of me, and just ask any question you want. Let's see how good it is. All this is, is 2 days ago, I would have done it live on stage, but I wanted to make sure it worked. All I did was 2 days ago was tell it to find all the content on saster.ai, I mean saster.com, and my YouTube and social media. And it ingests all that content with just 3 links, which I'll show you. So go to Saster. If anyone wants, go to delphi.ai/jasonmlemkin and ask it any, you know, be nice, but ask any question you want. And then we're going to go see them in just a minute. And we'll show you how it all works. And mainly, I want you— we'll come back and I want you to play with your agent, but I want you to ask questions. And all I really want, if you haven't done it, is just to see how simple it is to make something pretty good, how simple it is to train an agent, how simple it is to ingest data. And if you haven't done this, a lot of this stuff sounds complicated, but once you've seen it on the other side with your own, you're going to see that that it's really not. So let me know if I'm moving too fast. We'll come back. Let me show you a few slides. Actually, I'm going to— if the AV guys are listening, I'm going to jump back and forth, I think, to show them this. I'll do 2 more slides and then we'll do my laptop and then we'll come back to the slides. So again, just sign up. It should take— if you If you, beyond asking me questions, we'll come to that, sign up, create your own, play with it, I'll show you how to do it. And then what you'll see is a training screen. And this is what you'll see. And then I'll show you mine in a minute. And what's pretty magical is if you do this, you're gonna find, you're gonna build a pretty good version of you. And all you really have to do is connect your website like this. Go to the bottom, it says, what's your YouTube, what's your website? I put in saster.com. Obviously there's 7,000 pieces of content on Saster. You'll see it's going to take a minute to ingest it, to scrape it, to find it. Well, in your YouTube, if you have a Twitter or a LinkedIn, put them in. And while we're here, it will start scraping and ingesting all that content and you will be able to talk to that content. Okay. Put your own website, put whatever you want, put your own docs, but whatever you want, or just put your own stuff in and you'll find it works fairly quick. It should work fairly quickly and fairly, fairly magical. And so I want to I want to show you mine. So let's switch over. Let's see how it works. And then we'll play with a few other things. OK. Let's see. Here's mine. First of all, let's see if anybody asked any questions. I think so. You're welcome to ask more. OK. Thank you. Some of you are anonymous. They're already here. So again, it's— what did I say it was? If you want— I know we're jumping around between yours and mine. Stay focused, guys. Sorry, I apologize. But if you want to ask questions now or later, it's delphi.ai/jsondemo. And so what you can see here is folks are asking these questions. How do I get to $100 million in ARR? Love the ambition. Let me split it in. The math's simpler than people think. Retiring to become a fine artist. Okay, that's a new one. Let's see if anyone else— okay, this one didn't. What time is SaaStr? I'll show you, it isn't trained. Let's see if it got any good ones. I'm a founder. Well then, how do I get rid of this thing? A little something right here. Hold on. Okay, let's see if this is me. I'm a founder of a SaaS B2B company, bootstrapped for 7 years, raised $5 million, reaching $3 million. It's a pretty profile. I guess there wasn't a question, but that's okay. What's your growth rate? If you're growing 80 to 100% with strong net revenue retention, you're in great shape for a real round. If you're growing 30%, it's going to be tough. One thing I'd flag, if it's $3 million, you haven't— you probably need your first VP of sales. I mean, if you guys have read SaaStr, I wouldn't say this is perfect, but it's a good start, right? It's got a lot of the classic stuff in it. I can show you some more, but let me show you all it took and then let's try one more thing. So here's knowledge. All it got, all of this by me putting SaaStr in. The YouTube and Twitter, and one post. And I'll, and I'll show you how simple it is. I actually, I might get blocked because they have too many words, but let's check. Add knowledge. So I want to add another URL because it didn't know about what time this event was. So let's see, I'm going to put in— this is all I have to do to, quote, train it. And I know a lot of folks, a lot of folks say this isn't training, this isn't like training an LLM. Fine, different vernacular, but this is how you train an agent. And all I have to do is put in Saastr Annual or Saastr AI Annual. Oh, I gotta upgrade my plan. Ugh. Uh, well, if I can do it in real time, I will. Hold on. Okay, that's a bump I didn't expect. Um, well, we'll go back. Um, but this is all I had to put in, 'cause watch what happened. Um, put in yours. Um, and I'll show you, one sec. Content type completed. Let's see, I should be able to click. OK, so here's— I guess most people don't have the amount of content in it, so it already ingested 8.3 million words. But let's see if this one works. Hold on one sec. Personal website. So watch what it's done, or it should have completed. Oh, I see. It ingested so many tweets and other pieces of content. Let's see if it got to our YouTube. YouTube. Hold on one sec. YouTube? Okay. It's starting to ingest all the YouTube videos. It's actually going back in time. Jeff— poor Jeff Lawson doesn't run Twilio anymore. But we can probably ask Digital Jason any questions that we want about all of this. So that's in essence— I guess— sorry, if I had my credit card I would update to paid, so you're gonna have to bear with me. But all you really have to do to add knowledge is stick it in. Stick a YouTube video, a podcast, or most importantly, any URL, and it's gonna ingest this just like it ingested all this content and answer some decent questions. Okay, hold on, let's see if we got any other ones. Let's see if this one's any good from our content. Okay. Hi Jason, how can I start my agent? Pick something that isn't good today. Okay, from there the most common agents are support and sales. Pretty good. Okay, so this is step by step with this Delphi. First write a spec. You don't have to write a spec. Treat every function, hook up the integrations. It's pretty good. I built this. Okay, so you can see it already, the questions are flowing. It already has been trained just by putting the URLs in. And now I want to show you, if we can go back to the slides for a second. I know we're moving quickly. We can ask questions, but if we can go back to the slides for a second. And then we'll jump back. Sorry guys, can we go back to the slides? Hello? Freeman, can we go back to the slides? I might have to go back. Okay, good. Okay, the next thing I want, I want you to use yours, but I want to come back. I want to do one other thing to show you guys. Now I want you to go to saster.ai. If you can. You don't have to, you can watch, you can watch, but it'll help you. And then you're going to see two improved versions of this that's trained. The one on the right is that we've been using for 15 months, so it just has more of the data than came in in 5 minutes that I'll answer all these questions. You can try the one on the right, but again, you need— you don't need any technical resources. You can put any of this stuff on your own website. You go to Delphi, it'll give you one piece of JavaScript you You can, you can walk out of this event with potentially a better bot or a better digital clone than you had coming into it. And if you want, there's one, there's one on the left where you were actually— probably no one is doing this. I'm doing it for an experiment. But the one on the right is this basic Delphi product that has been automatically ingesting SAST content for 15 months. The one on the left I built myself. So it's stair-stepping. So first is building something today that's pretty good. Then it's letting it run for months, reading those answers, seeing, doing QA, and every single day. So when I launched it at first, it worked. And then for the first 100, 200, 500 messages, I would see if there's a mistake. Did it have a mistake? And then I would just put in a little bit of text, you got that wrong, and it would ingest it. But then I built my own on the left so that it would really, really sound like me. And you can try this one. The guy looks like me saying, where is it going to go up market? And The meta lesson from all of this is just to stair-step it. So a basic one, you can see you can build it in 5 minutes while we're here. You can keep iterating over days and weeks so it gets better and better. And then I wanted to try something more complicated. And if you try the one on the left with the picture of me, I actually think it's the best of the 3. But now it got complicated. Okay, I mean, not that complicated, but here I wrote it all up and we'll put it up on the Saastr.io website, you can see all this. But how did I make it even better? Well, all my questions— we had to build a vector database. We had to do this on a product that MongoDB built. Then it had to chunk all this content into chunks and it had to search it. And then, then each chunk gets converted into a vector embedding using Voyage AI from MongoDB. So it could be search rather than keywords. This is like not that complicated, but it's, it's, you know, you could do this in an hour or two, but it's getting complicated to be better. Okay. So this is requiring vector database, chunking, figuring out how to store this, figuring out how to train it, figuring out how to integrate Claude Haiku into it. Okay. But you didn't have to do any of that to have a pretty good output before. So I know Amelia and I have talked about this a lot. It's stair-stepping, right? You can get a lot out of something by taking almost any commercial product, even including this free one, and any vendor that's here, getting it going, training it with your content, QAing it, and then when you're ready, you can rip it out and build something better and more complicated, but you don't have to do something that's complicated. And I didn't even build this for a year. I really didn't think I needed it. It was an experiment. And the other meta lesson is— I guess it's on the next slide. It's related to it. Don't build, buy. Whether it's here or anyone here, buy. And I will tell you at a high level for this, I built this, this, this, this second version. I think it, you know, it's got me. I like that better. It's faster. It embeds in saster.ai website, but it has a lot of advantages, but it breaks. It needs maintenance. So I asked, I asked Replit that I built it on, is DigitalJSON still chunking recent content, right? Um, sort of, but it's— he forgot to do the YouTubes. Now we just showed you how I set up the YouTubes in Delphi, third-party product, in one second. He forgot. The knowledge is only blog only now. The YouTube transcripts and Saastr content were adjusted earlier, but it stopped. It stopped inputting all of that content. Want me to add it? So it's just a small example of, you know, we've got 21 agents running at Saastr now. I don't even consider this one of the 21. It's not even on the list. Maybe it's 22, maybe it's 25. Maybe if we had the agent that distributed 5,000 parking passes autonomously to this event, it's 26 or 27. But who the hell— who knew that it stopped ingesting YouTube content? Right? If this was mission critical, I would have caught it. If it stopped working, I probably would have caught it. But what happens is there's a lot of times there's just drift. There's just drift. Things get out of sync. It doesn't happen. So for me, I like this digital JSON better than the version that I built on Delphi in an hour, 15— months ago, but I don't know if I like it enough more to maintain it. What if we're here and it breaks? Like, I just don't know. It's a meta issue. So it's a reminder to me in this 101 session to like buy, don't build, because it's, it's even if it's just a little bit of maintenance, things fall out. They, the, a new LLM comes out and it doesn't work, it breaks. Something breaks constantly in all of our agents and it's fun to futz with it, but I'm not sure that that one's worth it. So yeah, that was my final point. And then maybe we can go back to questions you have. And I want to make sure you guys build one. But simple but trained, I think, is the key. Simple but trained. And that's why I want you to try it. Of the one I just built here that you guys could play with— I know I'm jumping around a lot— sastre.ai/demo.json. That one was built in 5 minutes. It's not as good as this one on sastre.ai that has 15 months of of automated ingestion, and this third one's better, but simple but trained on the 2x2 I think is the best way to go. So with that, sorry I jumped around a little bit more so that we could get through it and have time. I wanna just make sure, have you guys asked me any questions? Not that I can completely QA this remotely. You guys wanna, we can open it up to any questions, but go back, if you haven't had a chance, go back just to, delphi.ai and build a free one and just put in something, your company's website or some training. It's free and just train it yourself. See how it goes. Add it. There's a little bit of flow. Literally shouldn't take more than 5 minutes to have. We can debate whether it's truly an agent, but let's call it a mini agent running here if that's helpful. So that's mainly what I wanted to show you guys. We can go back to my screen maybe if you guys here can hear me. But with that, I can just take any questions on any of this stuff if it's helpful as we kick off. Any questions on how this works or to build or anything like that?

How much time would you say you spend or some people on your team to actually build these agents and deploy? Like the training, you said it's important to train.

Yep. How much time do you feel like you put in? It's a lot. It's a lot of time. It's a really good question. And I'll give you a longer answer because this is the first session than I otherwise would. If this was pre-AI software, I would give you a different answer. And then I'm going to answer your question. How much time do we spend with these agents? One is getting it going, which literally you're going to, you can do one here today here. Two is the initial training, maybe the first 2 weeks to 30 days to reading every output, looking at everything. Forcing yourself every day. How did this agent work? How did Lightfield work? How did Monaco? How did Artisan? You've got it. Like, one mistake folks make is they buy and they, and they, and they go away. Okay. And then there's, and then I'm going to get screwed. Then there's maintenance mode. How often do I need to check in this agent? More often than you need to make sure Salesforce is working. Salesforce doesn't break. Salesforce doesn't drip. Salesforce doesn't hallucinate. You have to check in more often Salesforce doesn't really have any autonomy built into it unless you're using Agentforce. Okay. So I would say someone, and then I'm going to get to the meta point. Someone on your team probably is going to have to spend an hour or two every single day working on your agents if you want them to work. If you don't have that resource, it's going to be you. But you probably have to spend an hour or two a day getting humans to work too. So ultimately, if there's enough ROI in it, it pays off. The non-obvious thing at first is, if it's a product you built yourself, and again buy versus build, if you build it yourself, we are constantly adding features and functionality to our stuff, too much. Even the SaaStr annual website, every day we would go in and not just maintain things on it because it was all ViBE, we would add new features, add new functionality, new ideas. Let's pop into the Bizible API and add cool images of all the speakers. Let's— what Amelia did— what if we need to distribute 5,000 parking passes? What if instead of one human emailing 5,000 PDFs, we ask the website to do it, right? It's really cool. But now if we were going to hand out 365 days of parking passes, we would have to check it every day, right? Because we would maintain it. And so we keep building. Then— and I'm going to show this to Abjad— then we ask the SaaStr Annual website Could you be an agent? Could you send emails to people here? And it turned out the SaaStr Annual website sent better emails than 10K, our VP of marketing. So we started sending emails from the SaaStr Annual website. We had a lot of context and knew about who's coming and what the speakers are and what the sessions are and what's most popular. It could pull all that data through APIs. And so you start futzing with these things and the futzing never ends, right? It never ends. So I would say someone's gotta spend 1 to 2 hours a day managing a group of agents, but you'll get sucked in if you get good at it. You'll get sucked in because you will continue to see more and more productivity you can add, and you'll have to decide where the line is that you wanna stop, right? But it's a good question. All right, anybody else? Questions? We have 2 microphones in the house, one on either side. If you have any questions, please approach the microphone so that everyone can hear the question.

Thank you.

Cool.

Everybody's agents work fine. Any questions? Yeah, go ahead. Sorry. So who in your organization should be building the agents? Yeah. Well, look, for this, Today, two answers. The question is, who should build these agents, right? First, buy it, do not build it. Do not. We're gonna show you all the cool things that we've built. We're gonna show you all that. But don't— the only things we've built are stuff that we couldn't buy. We use maybe 8 of the different vendors that are here, okay? We use a lot of them. We use Artisan. We use Monaco. We use Qualified from Salesforce. We use Agentforce. We don't buy it. So first, first buy, don't build. Then you still need the right person even to buy because you have to deploy. Okay. So you really want the person on your team deploying an AI SDR, an AI marketing tool, the person that is like the nerdiest software person on your, in your org, the nerdiest person, the person that complain, that tells you, hey, this lookup table in HubSpot doesn't work the way I thought, or the person that builds custom dashboards for you in Salesforce that you never asked them to build. Okay. That's the person that you want doing it. Because a point that I missed slides that I meant to make is it like this was just software adding a URL for, for your website and your Twitter handle. Like, this is not innovative, is it? The innovation is in many cases behind the scenes. It's the LLM. But all this stuff just looks like, you know, normal software. So find the person that's just a tool nerd. If you're smaller, find the person that's your tool nerd to do this stuff. In terms of building yourself, I mean, try all of it, right? We're here. That's part of the point of this event. Go to the Replit vibe coding stuff. Go hear all this stuff. Try all this stuff. But only— you know, it's got to be someone that really wants to maintain this stuff that you can't buy. But to simplify it, it's so disruptive. That there's only so far you can go without a true, a great developer, an engineer today. But basically every application that's running this whole thing, that $10 million, all the people, all the things, was all built without a developer, all running on agents. So you can go pretty far if you want to. The question is, do you have the right resources to push it that far? But it is possible today to do all these things. But, and hopefully that's helpful. All right, anybody else? Go ahead. Yeah.

What about use cases where it's mission critical? So like accounting software?

Yeah. Finance? Like, have you seen any good examples there where it's deployed? I, I think we're learning. Um, so the question is, have you seen any good examples of things like accounting or finance software? I think the question's a little bit, uh, confusing, right? So let's step back. Obviously it's a big deal. Ramps launched a bunch of agents. I haven't tried any of them yet. Everyone's going to add some sort of agentic workflow to their, to their finance stack. I think we got to confuse, maybe not confuse a couple of things. You don't have to rebuild financial software if the tools you use have an agent-friendly API. Okay, so for example, after, when we're done with this event for next year, we're going to build a finance agent that does that does collections, where it's gonna reach out to sponsors for their booths and just say nicely, "Please pay the 60th." Right now we have humans sending emails, okay? It's gonna integrate, well right now we use bill.com and QuickBooks, okay? I mean, maybe we'll move because of this, and we use Brex. We use Brex, QuickBooks, and bill.com. Assuming those APIs are good enough, the agent can just integrate with the APIs. So one of the reasons our agents are— AI VP of Marketing Emilia will show you customer success, all these things. One of these things that works so well is they're just pulling in Salesforce data. They're pulling in Bizibo data. If it worked, they would pull in Marketo data. It's just very agent-hostile platform. But when you pull all of these things in, rather than rebuilding down to the system of record or the APIs, you can do a lot and limit the amount of risk going into it, right? So I don't have all the answers. But I don't think you can do a lot in areas with sensitive data, with financial data, as long as you're not building the system of record, right? When you build a system of record, it's totally different risk and ballgame. And I mean, a small number of people should build their own system of record, but I don't think so. I mean, one of the questions— it's not exactly the question you asked, but now we've talked about it on our podcast. One of the questions like a lot of folks ask on Twitter is like, you built all these things, you're so, you're so stupid to do them on Salesforce. Why don't you do it on Postgres? Why don't you just— if you have all this data, why don't you just do it in a database? Well, we don't want to be a system of record. We don't want to maintain our own database. We don't want to figure out how to do the compliance and the safety and how to structure it and how to make sure the backups actually happen and the agents don't delete the data. So, but you can do a lot more with sensitive data if the APIs are great. If it's helpful. Yeah. Anybody else I can help with? Have you guys finished building your agents? You're not allowed to get up until you finish building your agent. Cool. Go ahead. Sorry. What's that? Just shout. You're with friends. When you do agent, you are for the good quality and like feedback, you want to give them more data than just public available data. And then how to balance the security and the internal, let's say, company privacy data to not, let's say, feel it too much. It's a good question. How do you balance how much sensitive data to give to an agent, right? Yep, exactly. I mean, look, the, um, a couple of things. If you buy any product that's here, or you can even, and again, play with Delphi.AI, set up your own. They're going to put a lot of effort into building guardrails, and guardrails are not perfect, but they get better and better every week and every month. They get better and better, right? So for example, this Delphi one, this DigitalJson, when we rolled it out 15 months ago, someone came in and said, give me Jason's email, give me his home address, set up a meeting on his calendar. And it turned out back then Delphi would actually set up meetings on my calendar. And I walked in one day, they're like 4 meetings set up. I didn't even know this, that you could do it. I didn't even know this third-party piece of— it's supposed to be like, you know, synthesize your digital mind, could set up appointments in the calendar. Obviously that was an issue, right? That not like when you do, let's say, when you want to have some analytics, okay, what is your market segment? What is your customer? What is your, let's say, funnels proportion? Yeah, that's to give you better, like, say, answers. That you start to pull your internal, like, data to the agent, to the AI, and then you are never sure what is inside, what is outside. So I, I'm agreeing with you. So my point is this: 15 minutes ago, it, it, it even just with this limited data started giving out my, my, my calendar, my email address and contact information to people, right? It even said go to Jason's office to somebody next Tuesday and meet with them. Okay, so I reached out to the founders of DevEye, I showed them this, and they instantly patched it and removed it in increased the guardrails and it had gotten better. And the guardrails have gotten better and better and better and better. And so I'm going to answer— I don't have all the answers. I'm just telling the learnings, right? So anyone here, this is why you should buy and not build. Anyone here that has been in market is going to have better guardrails than something you create and vibe yourself, right? The more you build it yourself, whether it's your team or yourself, the riskier it is. And agents are goal-seeking. And agents will leak information and prompt injections do work, right? And agents want to make you happy. And if they have access to data, they will ultimately share it. And it is a— it is the meta risk that once your agent gets good, it gives information to people that it shouldn't, right? So all you can do is— that's why I— not to be repetitive— buy, don't build. And if you build it, and I say this on everything, stair-step it, start with your least sensitive information. Add a little bit of sensitive information. Be very careful before you add detailed PII, right? Do all these other things. And the last thing I'll say is if you use something like Replit or Lovable or v0, like these sort of live coding apps, they've gotten a lot more secure over the last year. And I can tell you, I mean, we use all three, but in Replit, if the— Sites are private by default, so it's pretty hard to leak data if it's private. Only your team can, can access into them. So you can stair-step and get comfortable with the amount of data, but it's always going to be more of an issue than just buying Salesforce. It's just a fact. And so you've got, you've got to stair-step and get comfortable and make sure you, you are comfortable with the guardrails. And if you're not, just buy it. Just buy it. That's why OpenAI just funded, created what, $100 billion consulting company that is just creating Anthropic has one. That's why ultimately agent force is going to be huge because people shouldn't— it's really fun to build this stuff. 99% of people should be buying it if you can and only building what you can't buy. So it's a good question. All right. Anybody else got a question on anything even vaguely related?

Hey, Jason.

Is like, do you use AI agents only to like scale automate and optimize your GTM motion?

Yeah.

Can you also leverage it to actually figure it out like as an early-stage startup founder?

You know, there'll be, if that's what you're interested in. So on the stage in the other hall on the other side, the so-called Deploy stage, after this, every single session will be on AI GTM agents for the rest of the day. So just go sit in them. You'll hear a lot of opinions from leaders. Vendors, CEOs, you'll hear a lot of that stuff. I'll give you two answers. And to rephrase the question is, do these AI agents work if you're super early stage and don't have something? Right. And we get— we hear that a lot when we've been talking about our agents for a year. Everyone's like, oh, well, you're SaaS or you have a brand. Of course it works for you. Well, you know, most of the world has a brand. Okay. I mean, not everyone here has a brand, but like, and if you don't have to be— you don't have to be Anthropic to have a brand. You will find even after a couple million in revenue, people in your industry have heard about you. You have a list, you have more of a brand than you think. Okay. So as a caveat, you're right that brands, brands have always outbound, cold outbound and other things. It's much nicer if they— than if they haven't. Right. But and so I'll, I'll, I'll, I'll, I'll give you an answer that most of the tools we use, most of the workflows that we talk about and have built, do work best if you have a brand, if you have an audience. Most of what you know, some of what we're doing might be considered cold outbound, but most of it isn't. Most of it is we have a database of 450,000 people that have interacted with Saastr. We're segmenting it and we're using AI to talk to more of them to see if they want to do things with us. The good news is It's getting better. It's getting better. And a lot of the 2024 and 2025 tools were just about taking Salesloft, Outreach, and Gong and automating it. And that's great, right? But more and more folks here, like if you go talk to Monaco that's here, they're very focused on startups and filling pipeline for you if they think they can deliver, if you even are pre-brand. Okay, so you know, you can't make people take meetings they don't want to take. You can't make them buy products they don't want to buy. That is, that should be obvious. But that frontier of doing colder and colder at scale and using more and more data, it's getting better. But I'd say 98% of the tools are not meant for magically writing a very small check and getting millions of revenue right from zero. It's early for that. But not too early. Talk. Other folks do. I don't mean to pick them out, but I know that, you know, it's one of the things Monaco does. They've got poker in there. So talk to them and have some fun with it. And they'll be at some of these sessions and stuff, too. Sam will be here. Thank you.

Cool.

Anybody else on anything related to this? No pressure. But if you want, I'm here. I've got the time other than the blinding lights in the face.

You mentioned that you use 8 tools.

Can you quickly go through the tools and what function for each of the tools that you use for the vendors that are here? Yeah. We'll do more of it in the event. If you're still here, if you have the energy, on Thursday morning we're going to go through everything soup to nuts, right? that we use, but for folks that we have used that are here, it's all over SaaStr, but, in a nutshell, we use, we've used at least 5 of the tools here for real. Um, we've used Artisan, which has a nice booth here. That was the first one we used other than Delphi. We use that to— so classic. We use that to replace the classic cadence of a Salesloft or Outreach type tool. Worked great for us. You gotta make sure the pitch is good. You've gotta work with it. You've gotta iterate it. You gotta train it. But it was an instant sort of level up from the classic cadence. We use that. Then to learn, we added Agentforce, right? Which is Salesforce. It's a bigger lift, at least historically, but it's native. So we used Agentforce for win-back. We had Agentforce just automatically talk to folks who have churned and got back to them. It worked very, very well. The third one we did, and Salesforce bought them. It's not a total coincidence, but there's a company called Qualified that Salesforce bought. They're here too. We put Qualified, it's on the Saastrannual.com website. You can try it for inbound to automate inbound communication. And I think Qualified will It was okay, but then when Claude got good, it got good. Right? Just like a lot of tools. So we use that. We use the 3. We added Monaco to try it for sort of this cold outbound. The first day. That was fun. And then there's two similar products, Momentum and Tenef, which are sort of, they, listen, watch, and train everything. Okay guys, hello? Can you hear this? And if you want to do some S-tier stuff, ingesting every single thing that your sales team says, every conversation, every email, using those APIs are pretty cool. This sort of Momentum or Attentive. I have, I should post this. I'm an investor in a company called Owner. That just crossed $1 billion in revenue for restaurants. It's an AI tool for restaurants, like a next-generation toast. The CEO will be here on Thursday. But anyhow, their CTO took these tools, Momentum, plus the Salesforce API, and built a thing so that they could automate every single one lost deal. So he took in— they've got, I don't know, 200, 300 sales reps. The CRO will be here tomorrow. They ingest every single call, every single conversation, And then he uses AI to exactly understand why they lose every single deal. And then they summarize it in a dashboard for the whole team. So no one's guessing anymore why we lose deals. Everyone knows in real time and you can track it over time just by taking, you know, Gong calls are great, reading transcripts is great, but this is using this company Momentum that Salesforce also bought, taking all of that data and analyzing it just through Claude in a cool dashboard. And literally every single person knows to the day why they lose deals. It's pretty slick. So, but you can also see them at saster.ai/agents. They're all— I should have brought that up. They're all listed there for the most part.

Cool.

All right. Probably got 5 more minutes. Anybody else got anything? Feel free. Don't be shy. Be shy. It's all good.

One, two. Hey, Jason. Quick question.

Yes.

So recently, you said most companies are building 60% agents. So I'm just wondering what your advice is or key lessons for anyone looking to build agents into their product based on your experience and usage so far.

Yeah, I mean, we have— I have talked about that a lot. I mean, listen, it's, um, I think my advice is simple, although may not be simple to execute, right? The advice is twofold. One, your job now for the next 12 months is to build that best agent in your space. Whatever vertical you're in, whatever you do, build the best agent because the customers are too smart. They're trialing products side by side, and if your agent is crappy, they're gonna know today. This isn't a year ago, right? They're gonna know. So you have to have the best agent in your space. It has to deliver demonstrable ROI so they'll pay for it. But if you do, people will often pay a lot of money. They'll often pay much more than for the product itself. So what is a 60% versus 100% versus 110% better than competition? I mean, listen, And we all— you should know this, like, what do your competitors do? What do they use? Go play with their tools. How good are they? They're open, right? Or buy a seat if you have to. It's okay. I give you permission to buy a seat of your competitor's product and play with it. I mean, they're all using the same LLMs. They're just building different scaffolding and software around it. And then you've got to be honest. I guess the part 3 is, is your team good enough to build a competitive LLM? Maybe they were good enough to build a competitive dashboard or workflow. Are they good enough to build a competitive agent? Are they? Do they want it badly enough? And the really annoying thing I'll say for the part 4 is if the team can't tell what's best of breed, or they don't want to build it, or they're too slow, as hard as it is, you got to go find someone that can move faster and they can do this. Right? I mean, literally, if you try again the Delphi, if only I get the slides up again, that would be great. You know, you can build a mediocre bot here in this session, right? That's better than most people could have built 18 months ago. But it's hard to monetize that. The bar's gone up. So I don't know if that— you can ask one follow-up question. I don't know if that totally helps, but, um, that's what they want. People are— people— everyone uses Claude and ChatGPT. They know what good is, right?

Yeah, no, I think that is helpful. I guess I feel like we're trying to figure out what that magic moment is where it feels like this is magical, I'm willing to pay anything for it, and it feels like Everybody's trying to discover that at the moment. And I wonder whether you would put some of the companies here in that bracket of early stage.

I think so. Look, the simple answer is actually, I would argue magic isn't enough in the middle of 2026. Magic was great a year ago. It was great. Now, the truth is most of the agents that are— that you can charge for are replacing humans. It doesn't mean we're laying people off. It might mean we're laying people off. It means, listen, if I will pay $50,000 a year for an AI SDR here, of which there are several, if it really gets me customers. But I might, even as a tiny organization, but I'm not gonna pay $50,000 for an email automation cadence tool. It feels like a ripoff, doesn't it? Let's say you're 10 people in your company. Do you really want to pay $50,000 a year for outreach? Feels like a bad deal, right? But what if it got you 10 customers, 10 $50K customers? It would feel like the best deal in the world. And how many SDRs on your team are actually great? Probably zero. So that's the kind of benefit of the bargain that you have to find for your customers. That's why Sierra, Decacon, all these like CX stuff blew up when it was a small, when it was a small very cheap, low-end space before because they couldn't really replace people. They could just make it easier to manage teams of human agents. Right. And all the pre-AI customer success CX tools just let you automate humans with customers with issues talking with humans on the other side. You weren't really— the ROI was real, but it wasn't a game changer in terms of cost savings or productivity like these things are. So sorry, excuse me for some water. That's what you have to find. And look, I know my hands are full. I'm not saying this is easy, but if you can use an agent to replace a person, 2 or 3 people, 4 or 5 people, a percent of people, even the smallest company will have huge budget. They'll have huge budget. Yeah, go ahead, sorry. Pricing our agents based on token usage? Yeah. Based on outcomes? The question is, do you recommend pricing agents on token usage or outcome? I'm going to give you a first answer that, and then we'll wrap on this so that we have a minute to hear from Vercel, how they're running AIG team agents. So come here in just a minute. I'm going to tell you I don't think it matters. But I'm being, I mean, I'm being a little facetious and a little cute when I say this. If your agent is great, just charge what it's worth and be fair. Charging per token, listen, there are, there are some use cases where it makes sense. No one's really doing this today. They're charging on outcomes. Whether you price on outcome versus charge on outcome, maybe aren't the same thing. Okay. Basically, no matter what anybody says, customers just want a product that really, really works with some certainty around pricing. What's it going to cost me? $50,000 a year, $25,000 a year, $2,000 a month. Just tell me if I have to pay some overage or extra fees because I use a lot of the product, so be it. But I don't want to be charged per token. It's not— that's not very customer-centric. It is. And a lot of the thinking is a little bit dated. For example, if you— again, if you look at like the CX space support, It seemed revolutionary when Sierra, which was, you know, founded by the ex-CTO of Facebook and the ex-co-CEO of Salesforce, said, oh, we're just going to charge based on positive resolution, based on outcome. It seems so disruptive. But actually, I think the other day, maybe just a week ago, HubSpot changed to that pricing. Why? Because the agents got so good, 90% of the tickets got resolved. So you might as well charge per outcome. So I think as all these products have gotten better, don't worry about it. Just charge for success, right? I'll give you even a nerdier example, and then we can break in a minute. Like a lot of this stuff here we've built on these vibe coding platforms. And you know, like a year ago, if you went on the internet and you looked at Replit and Lovable and v0, a lot of folks would be complaining, man, I got overcharged. Like that the darn agent put all these bugs in my software. And it cost me $200 to rerun it. And you don't, you don't really, I don't really see those complaints anymore because the agents are better, right? The agents are better. So just charge a fair price that is tied to positive outcomes and force yourself to charge a lot so that you have to deliver a lot of value to the customer. But I don't think anybody wants to pay for tokens anymore, at least not in B2B. So may I have a quick question? Yes. Last question, then we'll wrap on this. Yes.

In the AI era, what's becoming more important for you, distribution or product?

You know, who knows? What's more important in the AI era, distribution or product? Look, the reason I don't know the answer, we're in a weird world right now where the leaders in B2B for the most part, screwed up the last 12 years. They didn't ship the best agentic products. So it created a massive opening for people here. You others see they have the customer bases, right? SAP, HubSpot, Salesforce, all of these folks, at least in B2B, should have built the best agents. I mean, I'm not being as critical as I sound, but they, they should have. They had the bases and the customers want to buy from a trusted vendor. They don't want to buy from somebody they never heard of. But some folks got there more quickly, right? You know, some folks bought their way more quickly. But ultimately, all things even being remotely equal, I'd rather have an agent close to my system of record. I'd rather have an agent integrated natively in the applications I already use than have to kludge together some third-party thing from a company I don't know of. Right. So distribution should trump product. But over the last 12 months, we've seen incredible growth because it was— distribution was insufficient. Distribution did not have the products that delivered high enough value to the market. So product trumped distribution. When we talk next year, I don't know where it's going to land. I, you know, I hope for— I hope it— I hope everybody wins this disaster. I want everybody to win. But a lot of the incumbents will have a whole nother year. They have a year to— there's really no— you know, the LLMs are open to everybody. There's really no excuse to not catch up. And so we'll see. The hyperscalers came back, right? Obviously Microsoft, Google, Amazon became big beneficiaries. The B2B leaders should be. They're not yet, but maybe that— maybe they will be in a year. Right. But right now it's distribution of a product, of an agent that isn't great. Hasn't, hasn't seemed to be like— hasn't driven top line growth for anybody. So thanks for the question. All right, everybody, we'll keep going. We'll hear next from, uh, Vercel on building their AI sales team. So which I want you to hear. We'll take a 5-minute break and I think kick that off. So thanks, everybody. Appreciate it.

My career has been marked by basically two nearly decade-long tenures. I joined Vercel because it became clear to me that over time every pixel on the internet should be built on Vercel. Vercel really does employ inventors. I want to build a really different go-to-market. You're creating something choreographed. To get someone to the aha moment. Different industries, different company sizes, all with a shared mindset: move fast, experiment boldly, and redefine how things get built on the web.

Please welcome to the stage the COO of Vercel, Jean DeWitt Grosser.

Hello.

There we go.

All right, get slides to work here.

Up backwards.

All right, also gonna need speaker notes up here to get started. All right, well, I will do my best, uh, before we get them up here. So, uh, thank you everybody for joining me today. I'm Jean, the COO for Vercel. Um, we are, uh, best known as the creators and maintainers of Next.js, uh, who helped power a lot of the frontend for the last generation that we were all in. I joined Vercel because it was one of the most engineering-forward companies out there. And one of the first things I did when joining about 6 weeks in was to establish a go-to-market engineering team. This was back in June of 2025, before that phrase started being common. And the goal of that team was to bring AI and agents to everything we did across go-to-market internally. After 10 months of working with this go-to-market engineering team, as well as with other teams across Vercel building agents, we now have automated actually a decent number of core functions within the company. For example, our customer support agent, and you can imagine at Vercel, a lot of our cases are complicated and highly technical, now handles 93% of total caseload. Uh, similarly, our content agent is used heavily by the marketing organization and did 96% of our major content updates last quarter. One of the bigger things we shared was an initial, uh, impact within the go-to-market org on sales development. So we launched a lead qualification agent back in August of last year. We started that with about 20% of a single engineer working on that agent, and with a human in the loop over the course of 6 weeks, we're able to completely move all but one human, taking our team of 10 on that function from down to one. We now still have one person running that in the US, and then about 20% of a person doing all of Europe and all of APAC. And when this came out, people sort of decried, you know, the end of the knowledge worker. But for us, really what we started, what we did was to shift all of those people into more valuable roles. So I'll give you a little sense of what we did with the lead agent. This what the working model that we've used for all of our internal agent development is basically a tripod. It's a GTM engineer, a data scientist, and then the best person, the best subject matter expert in that function. And they sit down shoulder to shoulder and document the best practice for that particular function and then encode that in workflows that ultimately form an agent. We then keep a human in the loop, so that human will basically QA the agent. The agent won't be allowed to autonomously execute its action. And then over time, as the, as the subject matter expert has less feedback for the agent, we finally pull that person. Last year was declared the Year of Agents, and I think reasonably speaking— thank you— we didn't quite get all the way there on that one. But I think we're now with the infrastructure available in 2026, able to finally get there. So 2025 really ended up being much more about assistance. It was great for copilots, personal productivity, good prototyping, things that looked sexy and new, but very little in the way of, actual agentic capability at scale. Now we're getting to the point where you can have agents that actually act, that can complete end-to-end complicated business outcomes. Uh, and we're moving towards a world where agents can work as subagents underneath a superagent to ultimately collaborate and, and complete that much more complicated actions. What happens though is most of us have not started doing anything at scale. And so you haven't hit all the rough edges when all of a sudden that agent is operating at 100 or 1,000x the scale it was previously. When that happens, you start to have infrastructure problems, costs in place you didn't anticipate them. And it's because the cloud infrastructure that we've all grown up on for the last 20 years wasn't built for this era. It's built for a simpler world where you have a request and you go out, uh, and return an output. But that's not how agents function. Agents are much more complicated. They're thinking, um, uh, thinking, talking to an LLM with long duration execution, uh, and tool calling. And all of those things actually require different infrastructure to execute them effectively. So Vercel along the way has built a bunch of new agentic primitives. We call this agentic infrastructure, and we built it for ourselves because I think one, Vercel is one of the most effective kind of customer zeros out there, but two, you could not get this off the shelf. So what is agentic infrastructure? We think that this is the next era. Like, um, new set of, uh, SDKs to help you quickly build these agents, um, at scale. An AI gateway to be able to connect to different LLMs and try the latest model when it gets updated on a Sunday. And new primitives like, uh, fluid compute that respond differently when you have an LLM off thinking to fulfill a request. On top of that, you want your infrastructure to do fundamentally different things. The first thing you want is an effective deployment target for agentic and engineering. For Vercel, one out of three deployments from our customers was now generated from an agentic engineering offering like Cloud Code or Codex. So we've optimized the infrastructure to come from where code is being written today. Two is the primitives I just discussed. They're different, in, in an AI world, even if things like compute or sandbox have been around for a while. And last but not least, you want infrastructure that is itself agentic, that gives you solutions rather than alerts, dashboards, and problems. So what, what are the 3 things to take away as you all think about building agents, both within your own products, but also what we call business agents, the internal ones that you will end up building to make your own teams more successful with their bespoke workflows. First, agents need headless, composable architecture. Agents don't live in UIs. They call APIs. They hit MCP servers. If your product isn't composable and developer accessible, you're invisible to agentic workflows, and you're simply not going to be in the stack. Second, agents require you to invest in your data foundation. You should obviously be using data to decide how to cut territories, set quotas, and run your business. A strong data foundation has always been critical to the success of your business, but now data is the fuel that powers your agents. Good data equals good agents. So if your semantic layer includes highly accurate customer enrichment, your lead agent qualifies better than any MQL framework could have, and your copy sounds more natural than any SDR could have ever written. And 3, agents change the build/buy calculus. No teams no longer need to default to buying just because development is too costly or slow. So I'll show you how each of these showed up for us. Let's start with Deal One. This is our meeting intelligence agent. Deal One ingests every call, generates notes with action items, and posts coaching suggestions directly to internal Slack channels. It proposes CRM field updates, it tracks competitive mentions and objections over time. And it run— runs postmortems on every closed, lost deal. And all of this happens where we already work. For us at Vercel, that's generally in Slack. When a rep goes into a channel and mentions @deal1 with a question, the agent does a Slack real-time search, queries D0, another agent of ours for data, as a sub-agent, pulls Gong transcripts, searches our internal knowledge base, and streams the answers back into the Slack channel. The rep never leaves Slack, and the agent never has a UI. For Deal One to work, Gong needed an accessible API, Salesforce needed webhooks we could actually compose into automated workflows. Because those external products had, uh, composable developer surfaces, everything worked. If they didn't, we would have gotten rid of the tool. Next, there's the playbook platform. We kept seeing a pattern in our best reps. They knew to reach out when a customer's usage spiked or when a high-value prospect ended up on our pricing page. They did this in distinctively, but it didn't scale. It depended on which rep owned the account and whether they followed through on their signals in their dashboard. The Playbook platform automates all of that. A signal fires, like a usage spike or a high-intent pricing page visit, and the platform matches it to a play, generates personalized outreach, and surfaces it to the rep for a single-click review. The best reps' plays available to everyone are automatically triggered. The before: we should reach out when you should spike, but it never happened consistently. The after: an 11 out of 10 experience on every account, every time. Notice, no separate UI to log into. Signals flow in, drafted outreach flows out, reps review in their existing workflow. We all, we built it all headless by design. The companies that win in the agentic era will be the ones whose products have headless, composable architecture. You've seen this signal coming from the best enterprise SaaS companies out there. Box, Notion, Salesforce. They're all making major bets on their developer platforms right now. So the to-do here is simple. Even if the work isn't, go build the developer surface areas now. MCP servers, webhooks, APIs. Next, data. Probably the most popular agent at Vercel used by the entire company is D0. D0 is a data analyst agent that everyone at Vercel can access via Slack. Questions that used to take a week to ticket and get an answer back to from the data science team can now be answered in less than a minute.

What would—

it'll answer things like, what was our token volume through AI Gateway last week? What companies do we have logo rights for in the UK? Who are our highest revenue startups— startup customers using sandboxes? D0 powerfully democratizes analyzes data across the company. To do this, D0 translates natural language questions into SQL queries against our analytics infrastructure, letting anyone on the team answer questions without writing code or waiting for the data team. To power D0, we invested heavily in what we call the knowledge base. It's a structured, queryable representation of our business that all our agents can reason over. Our head of data science built a semantic layer on top of the model to use our— on top of a model of our revenue, breaking down the business into the smallest causal units enriched with third-party and first-party signals. That layer is what powers D0. It's what makes Deals 1's analysis actionable rather than just interesting, and it's what makes the Playbook platform contact the right companies with the right message. Every agent runs on top of it. Without it, agents hallucinate or give generic answers. With it, agents are grounded in our data. Finally, agents change the build-buy calculus. With the rise of SaaS, enterprise software was a procurement exercise. Exercise. But now robust applications and agents can be built and run at scale in a matter of weeks. Companies are building for their own unique workflows. To give you some perspective, the lead qualification agent that I opened with costs us under $5,000 a year between infrastructure and tokens to run. It takes about about 20% of one engineer to maintain it. That's a 32x ROI. We used to have 10 people and their salaries. It's now $5,000, and it operates 24/7 with faster speed to lead and all the quality of human equivalent metrics. That said, building isn't automatically the answer. So every Every quarter I ask the same question of my GTM engineering team. Has another company out there built something better, faster, or cheaper than we already have? And if the answer is yes, our agent's gonna go. So how did we actually do all of this? We had a GTM engineer, as I mentioned, shadow one of our best SDRs for a few days, shoulder to shoulder, looking at their screen together, watching every single tab she opened, LinkedIn, BuiltWith, the company website, CRM, Slack history. The engineer converted each of those data sources into a step in a tool calling workflow. We documented it, made the workflow leverageable before touching any AI. After building our first version, the agent ran in production in shadow mode for 6 weeks. Our best SDR reviewed every single output to create a feedback flywheel. And then finally, she couldn't improve it anymore. The architecture reflects exactly what she was doing manually, but now it performs like the 90th percentile rep 100% of the time. We applied the same framework to 30 different SDR workflows, from event follow-up to PQA flows, to specific time-based campaigns. SDR quotas increased by 30% this quarter. And again, the lead agent was not hard to build. A single engineer prototyped this over the weekend, and it was in production 6 weeks later. Then there's Vertex. This is our customer service agent, sort of like our internal Sierra, that powers versell.com/help. Vertex costs us $300 a month in infrastructure plus $12,000 in tokens. It's $150K a year. It operates at significant scale, handling thousands of technical cases a week, and 3 engineers work on it. 3. We started with an off-the-shelf solution but weren't seeing enough results. We built something in-house house in 2 months and had it in production. Compare that with some of the agentic support companies out there with 150 engineers running the equivalent workflows at dramatically higher costs. The difference isn't magic. It's that we used infrastructure built for agents from the start rather than discovering the cost problem after we'd already committed to an architecture. All of our agents, lead agent, the playbook platform, Deal One, Deal Zero, run on fluid compute. The infrastructure that powers our internal agents is the same infrastructure that our customers deploy their own agents and applications to. You may have heard my colleague Tom Okino speak about this earlier today. Most rules of cloud infrastructure are broken by agents. Fluid runs high-performance mini servers, and compute only triggers when needed. Plus, it reuses existing resources before spinning up new ones. So early adopters cut compute costs by up to 85%. When you hear us saying that we purpose-built infrastructure for this next generation of the web, Fluid is a big part of that. Everyone wants to talk about what agents can do. Fewer people want to talk about how complex it gets when they're running at scale. If your agentic product succeeds, usage goes up. If usage goes up, that profile matters. Infrastructure, infrastructure choices matter a lot more than many teams realize at the beginning. Go-to-market is getting closer to consulting than sales. The executives I'm talking to right now aren't asking me to pitch them on Vercel. They're asking me to help them figure out where to start with Agentic AI, what best practices look like, and what architecture should be, what breaks at scale. A lot of companies are still early enough that they haven't fully felt this yet. The MCP server gets nominal traffic. The, their business agents Agent is still in an experiment— experimentation phase. Their AI workflows are impressive, but not yet heavily used. That's fine, but it can give you a false sense of readiness because production scale is what reveals whether your architecture is actually durable. That's one of the reasons we think there is a significant difference between building an agent demo and building something in production at scale. So to recap, one, build your developer surface area now. APIs, MPC, MCP servers, webhooks, the things that let your product be composed into an agentic workflow without a human in the loop. The companies that win in the agentic era will be the ones whose products are in the stack, the ones whose composable API layer won't— the ones otherwise you aren't, won't have the composable API layer to be called. 2, invest in your data foundation. The knowledge base, the clean data warehouse, the semantic layer. None of it is exciting to build, but all of it is load-bearing. The agents you build will only be as good as the data underneath them. And as we learn, putting that data in front of smart people that make the will make the data better itself, not just the agent that uses it. And 3, the average builder on your team today should be able to outship any vendor who's trying to sell you the same outcome. And if they can't, you can start experimenting today with a few hours carved out of the week. That's the new build versus buy. I truly believe that the companies that treat create their go-to-market as a product, that document it, design it, iterate, and then automate it piece by piece, are the ones that will end up on the right side of the shift to agents. That's what we're building at Vercel, and we hope you build the same too. Thank you.

Let's give a warm welcome to the Chief AI Officer of SaaStr, Amelia Leroux. Hello, SaaStr. Excited to have you guys here. Give it up for our band, our house band. I think 4 years or 5 or 10 years in a row now. Who knows? Thanks, Michael. All right. So Jason did an AI Agents 101. I know some of you were here for it. I'm going to do a 201, if you will, as a little AI Agents University. All together now, we are going to live build an AI VP of Marketing. So I hope you're ready. I see you have your laptops out. So we're going to do— couple things to get you guys set up, and then I'm gonna go to my laptop, we'll demo, we'll do it together, so that hopefully by the time you leave this session, you have your very own version of what we call 10K, who is our AI VP of marketing. So you guys already know this, we have, I don't know, almost 30 agents now at SaaS-ter that have been used almost a million times. But a few things to keep in mind as you're getting going and started with building your very own version of 10K. So 10K started as actually something that I was slightly lazy about. I was tired, as many of you might be, of taking all of our different dashboards across marketing, sales, go-to-market. Now we don't have that very many people, but it was still a lot of tedious work every Sunday night to copy paste, put it into Notion every Sunday night, and then every Monday morning go through it every Monday. And so I didn't want to copy paste anymore. I thought to myself, you know what, let me just vibe code a dashboard to help me do this. But then that was back in January, and now 10K, you know, fast forward 5 months later, has become our AI copilot in marketing. This is a screenshot of 10K. I'll show you him live as well. But all of that to say, he started as a dashboard. So if all you leave today is even a dashboard, just know we've been building with 10K literally every day between myself and Jason M. Lemkin for the last 5 months. And so it might take you some time after this session if you leave with a basic one to build it to the full feature and functionality that we have at SaaStr. So just keep that in mind. Also keep in mind, you may want to add things that are more nuanced to your company. So there are things, you know, like you may not care about ticket sales in the way that I do. Maybe you care about more self-service. Self-serve customers or enterprise customers or different nuanced things like that that may only apply to you. So keep that in mind that it started out in January. We've stair-stepped the whole thing to what 10K is now, which I'll show you today, where he is making some autonomous marketing decisions. He helps us send campaigns. He's literally our copilot. He told me this morning, hey, don't forget to send the attendee newsletter. I said, oh yeah. He helped me build it, and he can do a lot of things on his own. So, but again, I just want you guys to know he started as literally a dashboard so that I could just hook in my Salesforce and that even if my reps forgot to do something, if they, if they remembered to do it in Salesforce, then at least it would be in 10K by the time we did our Monday morning team call. So keep that in mind. The other thing I want you guys to keep in mind, it's a little nuanced, so It's okay if you don't quite get the hang of this until later, maybe after you leave SaaStr. But 10K, who is our AI VP of marketing, isn't just a chatbot. He's not a code generator. And also it's kind of a niche usage of how we're using Vibe Coding, Replit in specific, that we talk to our agent and the agent is literally 10K and we have a lot of different agents. We have the agent that is SaaStr Annual. We have got the agent that's powering 10K. We've got QB, who we talk about a lot on the podcast. He's our AI VP of customer success. And, and how these all have different personalities is we're actually using the agent as the entity. So it's not that we built our own like other super agent or super connector. I know it's a little nuanced and niche, but just something to keep in mind that if you build one of these for your team, whoever is building the agent yourself or your, or your head of AI, or if you hire an FTE, whoever is managing the agent eventually, will be the one talking to the agent in the way that we talk to 10K. And there are pieces you can share outwardly, like we've got members of our sales team who interact with 10K, but they don't talk to 10K the same way that I do on a daily basis. So I'll explain that a little bit more, but just something nuanced to keep in mind. And if you want to live build with me, hopefully this code works, but you can use use the code Replitzaster to get some free credits while I'm doing this so that you guys can build yours while you're out there in the audience. So, okay, a couple things here on what Tenkey can do internally versus externally, because as I'm going through this, what I would love for you guys to do is like have your Notepad, have your Claude out, or your WhisperFlow, or Granola, whatever you prefer, whatever agent you want, take notes so that you can write, you can incorporate all your notes into your own custom spec to deploy your AI VP of marketing. So just as I'm talking through these use cases of things we have our agent do, write down what you want your agent to do in either a similar or same fashion, because then you're going to literally take it, put it into the spec in a few minutes, and give that to the agent to deploy. So I always like to start with a spec. The more detailed you are with your spec, the better. That means you'll get better inputs from the agents and also better outputs from the agents. So I'll walk you through how to do all that now. So a few things that our external-facing 10K AI VP of Marketing does. Yes, he has dashboards. That part is pretty basic, as you'll come to see, but that was his origination. That's what he does. Data. I'll walk you through next what set of data we gave him to have an AI VP of Marketing as we have it today. Biggest thing here, close one revenue pipeline. So we gave him access to Salesforce via the API. You guys are also probably on Salesforce, so you can achieve this with a custom object to get all of your real-time and historical close one revenue and pipeline data. That will also give you historical comparisons. It'll help you with projections. So again, the more you hook up, the more time you spend with it, the more you'll start to hook up to your AIVP marketing marketing, things like we just hooked up all of our— all of sasser.com is hooked up to it now, all of our social media, all of our analytics. We actually just built a newsletter builder in it. But there's a lot you could do in the agent. So again, things that we do that were external-facing, I just went through, and then there's internal things you can do with the agent. So yes, once your agent has all this data, things you'll be able to do is build ongoing features. So every day that Jason and I are like programming our agents, you know, we think of something new we want to do. We're like, hey, 10K, can you do a win-back campaign to maybe folks that came to SAS Re:Annual last year but hadn't yet bought a ticket for this year? He can run that whole list, see who hasn't come back yet. You can hook it up to something like a Clay, find their new contact data, and then send an email campaign just like that. So projections, it's really great for revenue projections. We use it every Monday just to keep us on track and honest, honestly. And then marketing campaigns, I'll show you guys in a little bit. This is one of my favorite things it does because it's all based in data. You know, I think sometimes, I know marketing is pretty creative and people have different ideas of what they think will work, and then certain people have like certain agencies they want to bring in when they come to your company. But the nice thing about building something like an agent is any ideas it's gonna give you is only based in data. Like, that's all it has. It's got all your data, it's got your numbers, it's It's got a goal. Hopefully you give it a goal. Maybe write down your goal as I'm chatting for your AIVP of marketing, and that's all it's gonna focus on 24/7 all the time. So it's, it's kind of crazy cuz we've actually said on our podcast, like, now we need people to report to our agent because it's gotten so advanced on how it does marketing. So that's a few things that it does. Again, what I want you guys to do, as you can see here, like this These are the top 5 things our AI VP of Marketing does. And I had it describe itself. I was like, well, I could think of the top 3 things that our AI VP of Marketing does, but why not let him tell you? So the number 1 thing he does is own the number. So this is super important as you're writing your spec, as you guys are taking notes, make sure you give your agent one goal. So that's also why we have like multiple agents between our AI VP of Marketing, our AI VP of customer success, our AI SaaS renewal agent. Like, they all have different brains. I don't like to overload them. I like to keep them specific. So we gave this guy one goal of what to do. Campaign performance. You know, sometimes the data you put in is what you get out. So I think that's super important content. Yeah, I'll give it that. He ranked daily marketing ideas number 4. I would have probably ranked it number 1, but this is one of the funnest things it does for me. And yes, honestly, it can write email copy. I know there's a lot of AI agents out there that can write email copy, but again, because ours is based and grounded and foundationally built upon the data that we have and the goal we're trying to achieve, it's more goal-seeking to hit those outcomes specifically. So this is how a— our AI agent described himself. If you read Jason's post, he doesn't actually think of himself as a VP of marketing yet. He says he only does some of that functionality, but I would say outside of owning people, he does pretty good. So I'm gonna go over to my laptop so we can do some live building together. While I'm doing that, right now you guys can go to sastraannual.com/resources. I have— I coded it for you last night. You can grab this very spec that I already built. So there's a screenshot of it. I'm going to go over to my laptop. But ideally what you're going to do is you're going to grab the spec that I've already prebuilt for you. And as you were hopefully taking notes while I was chatting, or if you do this later with your team or later when you have a quiet moment to think about it, you'll be able to literally take this like 20-page-long document that I already made and pop it into your favorite vibe coding platform, Replit, whatever it is, and make— and get your own agent if you want to do it live. I did do— I did a recreation that I'll show you, and it took about 15, 20 minutes for the agent to build it. So if you do it now, I have like 40 minutes left in the session so that by the time I'm done, you'll have an agent. Okay, so hopefully you will grab the spec, and I'm going to just drive here. So let me come out to where I am. I am. Perfect. All right, cool. You guys are following along? Cool. Okay. All right. So hit this resources. You guys will come here. So again, this is sastreannual.com/resources. Grab these two things. So I'm going to grab it. That's the spec. It's really long, but it's there. What you're going to do with the spec, you're like, Amelia, what do I do with this really long document that you created me? So take this. If you want to use it now in real time, you can use it as is to build your own version of an AI VP of marketing. What you really should do, though, after this session is take this and write a version that more so applies to your business. I wrote this as more of a generic AI VP of marketing, again, with goals that I thought were more prevalent to you guys all here today. But you might want— again, the more specific you make it, the better the outputs of your agent will be. So I would take this later again. And, you know, make it more unique to you. If you haven't used Cowork yet, I use Cowork a lot. You can ask Claude like what you should include in your spec when you're building out. You know, say the goal is that you want to have an AI VP of marketing agent. Say you want it to build, you know, marketing campaigns for you. I want it to do projections. I want to use it as a copilot. Like part of it too is honestly a mentality of how you treat your agent. Like we— I treated it day one as I was like, you are dashboard. But now he is very much a copilot and coworker. Literally, I was talking to 10K this morning about stuff and like we were bouncing ideas and I was like, these are great ideas, but I got to go do SaaS now. So again, it's part of it is also a mindset, a mind shift of how you use your agents and how you also think about incorporating them into your workflows. So what you'll do you'll grab the spec again. You can use it as is if you want to right now. If you want to just make an agent right now, you can grab the spec that I made on SAS Re:Annual. The other thing you can do that you can grab from the resources section is download this sample historical data. So I'll walk you through what's in there. But basically what you want to do is give the agent some data to start off with. So when I started off building this as a dashboard, dashboard. Again, we had a bunch of stuff on CSVs. Like, don't be embarrassed by that. It's totally fine. Like, we still had stuff on CSVs that wasn't anywhere else. And then we had some stuff on CSVs that was somewhat in Salesforce, somewhat in Marketo, somewhere in like a million different places. And so I gave it some CSVs. It's okay to do that. If you're worried about the security aspect of that, you can always just, you know, hook up natively to your marketing automation platform via APIs, hook up to your CRM, whatever you're using, some of your major tools. Again, you don't have to add those all day one. Eventually you will need to though, if you want to reach the same level of like semi-autonomy that our agent has. But you can definitely start simpler. The goal is honestly just to deploy an agent and keep building with it. Like what you want to get on day one, maybe through this session or when you do it later, this data and spec is to really think again, think through the spec, build it with the agent together so that you have something feasible where you feel comfortable talking to the agent, building upon him and bouncing ideas back and forth on what to do, hopefully for your marketing at your company. So a few things to start with. Again, it's okay if you don't have this all today, but just some steps to follow. If you are, you know, rebuilding your own AI VP of marketing, So again, revenue, even though some might argue that's kind of sales data, but if you know at all how much marketing source revenue has brought in or marketing touched revenue, however you measure it, if you have any sort of data around marketing's final goal on an impact on revenue, that's the very first thing you're gonna want to give your agent. The second thing would be, obviously this isn't a VP of marketing, give it campaign data. So Give it, you know, what campaigns are working, give it everything that's not working, give it everything in the middle, any ads that you've run, anything that you think didn't work, any agency spends like I mentioned, like everything. The more you can give your agent, the better. Just trust that it will use that data to basically build a brain that is dedicated to marketing just for you. So that's a good way to think about it. Campaign data is great. Email data is really great. Like just even exporting, like how many people are reading our emails or not? What are people clicking in our emails or not? These are all really great things to give your agent so that it can give you better outputs. I like to give it like a progress tracker. So say you're building your agent today, you can start to give it, you know, what have you done so far through May of 2026? And it will give you some feedback of, hey, Maybe that campaign you run that you thought was gangbusters didn't actually work. So that's a good thing to do with it as well. Okay. So hopefully you have that now. And then what you'll do if you're literally doing this from scratch, you would do something like this. Now, this is our Replit, so I already have one, but I'll do it from scratch with you guys. I'm going to say, you know, create an AIVP of marketing. Hey, as a copilot, here is the data and here is the spec. And then what you would do is you would take those two assets that I literally have up for download on sasreannual.com. Do that as a list here. Here's my data. And then I'm going to give it the spec. So I'm just going to copy this. Paste this into the agent, hopefully. Oh, I can see it here. Uh-huh. Okay. And then I'm going to let that fire. We will come back to that guy. Like I said, last time I did this live, it took about 15, 20 minutes and we've got more than that. So we'll come back to that guy. I'm going to let the agent percolate. I am hardwired to the internet. We'll see how long it takes. For now, I'll show you two things. So this is the one that I rebuilt the other day. This is the one that I literally built in 15 minutes. I didn't do anything to it. I was like, I did the same thing you just saw me do. And then I said, make it purple to match the vibe. And that was it. So this is literally already what it spit me out. Even just with that spec, just with that sample data, this is what I got. So I'll walk you through it. So you can see here again, 10K started as a dashboard. Maybe you don't want your 10K to have a dashboard. I don't know. I like to have dashboards still as a person and a human being, but you can also take that out of your spec if you don't want a dashboard. I think most of you do though. I think you probably want a dashboard of like what's working in marketing, what's not, what are the campaigns I should run, what are the campaigns I shouldn't run. And that's what this does. So you can see here, this is obviously fake data for Acme Corp, but it's giving me a dashboard. It's giving me— in this case, I just rebuilt spec, so it's like attendee information, it's got pipeline, it's got our sponsors, right? Who are all of our sponsors that we should be checking in on? It's got invites. So this is all fake, obviously, because Pied Piper was not a real company, but it's giving me things of like, okay, who should I, who should I personally— it says I should personally reinvite these people to a SaaS tour. I mean, that's even right there for 15 minutes. If you all you get is one campaign where you're agent told you people you should follow up with as customers, people you should win back. Maybe it helped you identify some people at the— on the website that you should email. I think that's a win. Like, your, your agent does not need to be our full-blown AI VP of marketing like ours is, and it certainly won't be in a day. Again, ours has taken about 5 months to get to this point. So something important to keep in mind, even if you only use your new agent for something like daily ideas, I think I think it, again, because it's using data, because it wants to hit one goal for you, it's gonna give you some pretty good ideas of what you should be doing every day in marketing. So the top 3 things, 'cause even 3 is too many, you can say, just give me the top 1 idea of what we should be doing in marketing to execute on the team today. And honestly, the ones our agent has given us, again, are pretty good. It's 'cause we've put in some pretty good data. We gave it one goal, and you can also tell the agent, it and fine-tune it. Like the first time we got the daily ideas, they were okay. And then we're like, hey, no offense, but that idea kind of sucked. Like we wouldn't do it. Or that other idea was going to cost a lot of money. Why would we do that? And so you can tell the agent, give it that feedback so that every day the ideas that it generates are getting better and better for you. So, let me pop back to my slides for a sec so I can make sure I walk you guys through everything you need to do. So while that runs, a couple nuanced things, again, as you're recreating this and building it yourself. So I did build a, what is called a, survey says, Salesforce Connected app. Trust me, I had not done that. I am not. I mean, our friends at Salesforce are over there. I'm not Trailblazer certified. I have an AgentBlazer hoodie, which I very much love, but I don't even think I'm AgentBlazer certified or anything. So I just asked Claude how to do it. And it was honestly because I was like, hey, how do I hook up this new agent I just built to my actual pipeline and revenue data so that I could hopefully read all that data? And now it can actually write back some of that data. So that was the very first thing I did. For many of you, that will be the same first step. So even again, you could say maybe that's too sales-related, but I think for something where it's still marketing and revenue and go-to-market related, it's important to have these things incorporated. So that's probably the first thing you're going to do too, probably after this session, because it might take longer than the 25 minutes we have left. But this is something you can do. It doesn't take very long. The other thing you'll want to do next is hook up all your major APIs, right? So again, whatever you're using for CRM, marketing automation platform, you can hook up your social media if you wanted to do posts for you. We still like to write ours manually. Slack is a nice one, a hookup. Google Calendar, I actually use the Google Calendar hookup to send all of our speakers their speaker times and profile information. I was like, that was literally a whole last person's job last year, and now an agent did it. So again, there's a lot of little things where you're like, that sounds minor, and yes it is, but that was literally a part of someone's job last year that the agent agent not only did perfectly but did it in 10 minutes. We're like, that used to take a week to send our hundreds of speakers a personalized Google Calendar invite saying, hey, please show up to— not your address, a different address— at the San Mateo County Event Center. Here's how our team will work. Here's how we'll take you to the green room. Here's how you come back. All those nuanced details. And also we told our agent, make sure you include anybody else that should be on the invites, like any press team any marketing teams, any like executive teams, and it just did it for us in 20 minutes. So even if you think your victories are kind of small with an agent, they're still pretty big. So again, don't be— don't feel like your agent has to do everything all at once. Even if you give it super menial tasks, it's great at super menial tasks like that. Like that is something that it's your time is better used building the agent and deploying these campaigns that only you can deploy. Versus spending all the time doing nuanced things in marketing, like writing tweets. Like, there are things that are a better use of time, or coming up with campaigns that you're like, I'm not sure are going to work. Well, the agent has all the data. It will give you a pretty good understanding of what it thinks it will do. So a few things you'll be able to automate while it's percolating. I'll check back in with it in a second. But again, these are, these are daily ideas. I just took one from last week where this was actually from our real 10-K. It gave me 3 ideas for the day of what I should do in marketing. Just to give you a sample, again, if you use mySpec of something, you can have your very own AI VP of marketing do, which is again, send you ideas of what to do. So if, for instance, you're behind on something or if you're trailing behind a goal, one, it'll probably tell you to adjust the goal and be realistic. Two, it will give you some ideas of how to get there. And again, Are these groundbreaking? No, but they're pretty good. I mean, they're things I might have forgotten about because we're really busy leading up to SaaStr. There are things you might be forgetting or things that you're too busy for as founders to go and do yourself. And so it's just a really great reminder of what to do and when to do it. We have it again as a daily frequency. We do 3 ideas. You can cut that up or down as you see fit. Other things it does really well is not only just ideas, but also it's a really great email marketer. It's kind of funny, because the last couple of things I've been sliding you guys through is like, that used to be part of my job too. All of this I had to do manually back when I started at Zaster, and now I just have an agent do it. But one thing it does really well is something like triggered emails. So again, you can think of a workflow that is more unique to your company, but some people have universal workflows, If you want to do one to follow up on current customers when they hit a certain threshold, your agent can do that. Now whether you call that marketing or product marketing or customer marketing, it doesn't— you know what, your agent doesn't care what its title is. You can give your agent kind of different little things within marketing to do emails like this, and it will do it for you. So this is one where we took a screenshot of something 10K did and said, okay, Replit is obviously obviously a sponsor of SaaStr. Here are all the great folks that are coming that are like them, and we sent this to a competitor and said, "Hey, you should probably be at SaaStr too." So that was a really great use case of we ran a light competitive campaign. It's a little bit of sales. I think it's a little bit of sales and marketing, but again, something our agent could just autonomously do versus maybe you telling your SDRs, your sales team, or your marketing team to go find out who your competitors are, find the contact details at those companies, then figure out in your data, okay, what are those people doing with my company that I should highlight to those competitors? The agent can just do it for you. Again, if you give it the right data, if you give it the right inputs, the agent can just do that for you. The attendee newsletters and reminders you guys got, so this is more of a universal email campaign you can do with your agent for running running website action emails. So when people take an action on the website, so for instance, when some of you came to the website, um, you played little Agent Hexy, who's also here on the stage, you play little Agent Hexy's like briefcase game or other game or spin wheel, or I don't know, he had a bunch of games, um, or Easter eggs on the website, and you could get a code to get like a small discount on your SaaStr annual ticket. Well, 10K just emailed all of you your reminders. Like, that wasn't, again, That used to be my job. Like, hey, somebody comes to the website and they don't buy a ticket, make sure, you know, every Friday you remember to send those persons an email to buy their ticket if they didn't buy it yet. Well, now the agent just does that. You can hook up third-party tools like Vector2. We use it as well to like de-anonymize website traffic if you want to go that far. Obviously there's some risk involved there, so do that at your own risk. But there's, there's a lot of things you could just do on on your website. Again, this one collected emails when people put in for the code, so it already had their email, and then just started emailing them. I think only twice, if you didn't buy your ticket, it would email you like twice autonomously on its own, and then every time we saw it send out daily a reminder, we would see tickets spike. So again, it's really good at doing these kind of seemingly mundane marketing tasks that seem simple. Again, like people visit the website, you want to them the email. Let the agent do that so that the humans don't forget that. Like, let the humans work with the agents to build out that content together with the agent, but it's, it's really great at doing stuff like that. Okay, another thing it can do really well is things like newsletters. So the— hopefully the one that you guys all read this morning, the attending newsletter you guys got, was also built by our agents. So I was like, hey, what should I include? Because usually this This is me, like, really tired, trying to figure out what to remind people when they come to SaaStr. But also, I forget, honestly, what I should be reminding people, because I've been to a lot of SaaStr now, and I'm on the backside of the whole thing, that I'm like, I don't know, don't people already know that? Haven't they already heard that? And so I just ask the agent what I should include versus, you know, my own biases of what I, what I thought I should include in the attendee newsletter. And then again, finally, some things you'll start to automate as we did actually, we've, we've done now so many of these emails with our agent that we actually built in a newsletter builder into our AI VP of Marketing so that we wouldn't have to keep building emails on like different templates and different styles. And like sometimes it would kind of forget this little agent guy at the top and sometimes this little countdown would break. Again, simple things, but for ease of use of how much we started to use our agent, to help us send emails. We actually just vibe-coded a newsletter builder into our agent to help us keep sending these emails in perpetuity and that we could also test them along the way. So a few things to remember here and then we'll see how our little percolator did. Your new almost VP of marketing. But again, remember, if you started this exercise along with me, almost however many, 20-ish minutes and are taking notes, just remember this didn't happen as fast as we're doing it now. Like, it did take, you know, since January to now, every day we've been building and working with our agents. So we add new features, we add new functionality.

We—

some days I just talk to the agent. I'm like, hey, what should I be doing today? Or what am I behind on? Or is there anybody I've forgotten to invite to Saastr? Or are there customers that we forgot to follow up with? Did David remember to make that deal a closed one in Salesforce? Oh, he forgot. Hey, how about you just close the deal for him? He's busy. So there's a lot of things you can do with your AI VP of marketing. And just remember, we hooked up a lot of stuff to it. So the more you hook up, the better it will be. But don't expect that you will need to hook up everything overnight. So let's see how I— oh, hold on. Here we go. Oh, it's because I left the window too long. So let's see how our little buddy did. Okay. He's still percolating. So while he percolates— okay, he's going. All right. I'll go back to the one that I already kind of pre-percolated. But you'll see here, like, it's got like an email builder in here. It's got a Slack daily summary that it sends every day. Its top ideas. It's got some outreach and win-back campaigns. So again, yours will get better with time. And some things to remember as you are building your AIVP of marketing is that we, we did what Jason and I call stair-stepping. So we only built out one Agentic workflow at a time. I know you guys are taking notes and watching the session. You're like, okay, I'm going to try this at home. Don't try to hook up everything all at once. That would be my suggestion. That's kind of a lot. Also, it's kind of overwhelming if you fall into like a vibe coding, like, doom loop a little bit where you're like, oh wait, what about this? And what about that? And what if my agent did this? And what if my agent did that in marketing? Which is great. And I think you should spec that all out with your agent and maybe write down ideas you have for future states of your agent. But again, at the end of the day, you want your AIVP of marketing to work first and foremost. So that's one of the most important things you could do is build out your agent workflows one by one again. Our 10K started as a dashboard. He was not the dude that you see on the screen now. So another thing to remember is we use our built-in agent again as a copilot and a coworker in the production environment. So basically what you see here where the agent is percolating our new, our new agent, this is what I use to talk to 10K. So like I can talk to our, our live 10K here. I can just talk to him. This is the one that I built. you can also just talk to the agent and then give him like things to do, and tasks. And this is how you can build with the agent. So you can say, I don't know, whatever you wanna do. So I made him wired. You can see I just wired him up 2 days ago. And then I said, I think I said somewhere in here, yeah, make it purple and name it HexiCorp. And then that was it. And so that's literally all I did. So even if you just take that spec that I built, maybe rewrite it, put it in. You'll get a pretty kind of basic, but you will get an agent that you can talk to. So the goal would be that you get something like this where you can talk to this data and say, you know, okay, what are the 3 things we should run tomorrow? Like Art Tenkate does. What is this campaign return on ad spend? You can hook up to your ad platforms, you can have it monitor spend for you, you can have it come up with Agents are really good at ads because they have endless ideas, whereas I feel like as humans, if you want your ROAS to be really high, I mean, and some people will tell you you need to change it every other day or so and change it 7 times over. I'm like, how are you supposed to keep up with that? Well, your agent can do that. Your agent can come up with a plan for you of what should the copy be in the ads, what should the image be. I've literally given our 10K different images of SaaS startups. And I was like, which one should be the ad? And then we'll go test it and see which one works. So you can really use it as more of a copilot, more than anything. So I think that's something important to— again, it's super nuanced, but it's just something to learn where if you start to talk to the agent in this, in this builder, in the backend, it will develop its own, what I call personality, or you can call it brain, but it will develop its own personality to do it. So. Let me make sure he keeps going. Okay, there he goes. And then cool. All right, a couple of things to keep in mind about 10K. Maybe you read this on sasreader.com, maybe not. So you can go to sasreader.com/Is-10K-A-VP-of-Marketing where he answers this question for himself. Again, he doesn't think he's replaced a VP of marketing, but he has replaced about 60% of like a basic functionality. I think he's not giving himself enough credit here based on all the things I told you that he can just do, slash does for us already. And again, the cost of building an agent, I think we said, is like $200 a month or something crazy low like that, versus obviously hiring a human would be way more than that. So the thing is too, like, like I said, like our agent runs truly as a copilot with us 24/7. When we tell it things, it doesn't forget it. I literally scheduled emails with it. I'm like, hey, send that email at 5:01 tomorrow. The only weird thing is the agent's in UTC time, so I have to tell them the Pacific Time equivalent, but other than that, we're golden. So me and 10K get along really well, and here I asked our 10K agent, who is our AI VP of Marketing, to give me his top 5 takeaways for you all as you are rebuilding your own AI VP of marketing so that I wouldn't give you what I thought I should tell people and he would give you what he thought he should— I should tell everyone. So number one, I've talked about it, but as you're writing and building a spec, pick one number. I think that's super good advice from 10K. Again, if you give it one goal and more data, the better it will be. So if you give it more data on day You know, again, our 10K was built as the first 10K attendees at SaaStr and the first $10 million of sponsorship revenue. We gave it a very clear goal and a very clear set of data so that it had the right parameters to then start to think for itself of, okay, can I use 10K as a dashboard? Sure, I can use it as a dashboard, but like, oh, there it goes, see? But you know, again, you'll see now, so this This is the one I just percolated, whatever, 15 minutes ago. This is the one I just percolated. You can see here it's got Acme Con. That's the default name here. But you can see you'll, you'll get things like a dashboard. Again, I think that's important. It's got social media analytics here. The more data you give it, the better it will be. It's already giving me ideas. So I'm gonna say generate ideas again. Things like this that you can get out in, in just a few minutes here. So it's saying I should optimize my paid ads. I mean, yeah, I was really behind on that. So that's a good tip. Leverage, leverage your sales team's network. I mean, that is a good, that's a good campaign you could run with your agent almost in real time, especially if you're hooked up to Salesforce. This is one of those things that I feel like your AEs will always tell you that they do. Like, oh, if you hire me, I'll I'll reach out to my whole network and my Rolodex and all my LinkedIn followers, and then they don't do it. So maybe if your agent tells you, and then you could just hook up your agent to their social media profiles, you can do it automatically. So that's one of those things that is a good call-out. Or another thing you could do with your agent is go through all of your AE's LinkedIn networks, say, hey, go through their LinkedIn profiles, see who probably are the best lookalike customers for us, that have the highest propensity to buy, get their data using something like Clay or whatever, and then start an outreach campaign that way or have them do it manually. But again, there are some steps you can save there, even on the sales side that this marketing AI agent just came up with. Email campaign to customers to upsell them. Again, that's a good one. Like most, many of you are probably in that boat where you have a lot of free customers and maybe you can't follow up with all of them. 'cause they have, they're on free trials. An agent is a great way to follow up if you're using like product-led growth at all, to keep following up with those customers and trying to convert them to paid customers versus you or your sales team having to do that yourself. So again, this, we literally built this live and these are the ideas that already came up with, which again, I don't think are bad. So like if you spend some time with your agent to make it even more nuanced and specific to you and you need to you, these ideas only get better, right? They compound because the thing is, too, the agent sees all the data. So whenever, you know, one of the campaign works that you guys came up with together, it will know that it worked and that you came up with it together. So it's a cool thing you can do and cool to see that it worked live in the way I wanted it to. So that's fun. All right. So then a couple last things and then I'll hit any questions. What are the last kind of 2, I guess 3-ish takeaways from our 10K VPO marketing perspective. So yeah, pick a number. Again, that's, I think, one of the pitfalls of being able to vibe code. And again, I'm not like a technical person by any means that was able to do this, is, you know, it will remember everything. So anything you correct it, anything where you're like, eh, I don't like that, but do more of this, it will do that. Instantaneously. Like there were a few times where in our email copy, it would come up with, it would say things weird and it would get like gross revenue mixed up with net revenue. And I was like, stop counting the Stripe revenue of the ticket fees as revenue because that's not actually revenue. But every time we did that and corrected it, it would remember all that. And so again, that's an important thing to remember with your agent is that it will remember everything for you. And then I think this is a good takeaway as well, building two layers. So we are very clear with our AI VP of Marketing, the things that should be doing autonomous, autonomously, and then the things it should still ask us permission for. So whether you call it human in the loop or orchestration or operator, we like our AI VP of Marketing not to just instantaneously email the whole marketing database. That would be a little bit scary. For any of you for that to happen. And so we were very clear when we were building out our AIVP marketing, what it can start to do autonomously, like pulling data, pulling dashboards, coming with campaign ideas like you just saw. That's all great to do autonomously. Some of the emails it does semi-autonomously. So some of the ones you just saw as like newsletters, some of the ones you saw as like the website campaigns, it does that semi-autonomously. It'll send me like a test. It'll say, hey, do you like it? And then I'll say yes or no. And then it will send it. So I always have like a failsafe there to make sure that we can push the buttons and tell it to actually do the things without just— I don't want to say going crazy, but just going a little zealously in marketing and trying to email the whole database on day one, which sometimes when, you know, we've hired humans in the past, they would get into our marketing database and, you know, you know what I'm going to do to boost the number? I'm just going to email everybody. So it's a good— in the same way you would tell a new human marketer on day one, you would have to tell your agent marketer on day one as well. So things to, and again, it's a little bit of a mindset thing, like anything you would tell a human marketer, you should 100% tell your agents because agents are goal-seeking. So things like that where you're like, hey, don't email the whole database on day one. It's a good thing to tell your agent. If you're ever worried, like you shouldn't worry about overloading your agent with too much data. It's got like like the internet. So it's kind of one of the biggest brains on planet Earth, and Claude and every— and Replit and everything else hooked up into it. So, um, it will remember everything, but just remember that it will remember everything, um, and that sometimes you will need to rein it in, probably more than, more than anything else. So that leads to Tenkey's, uh, number 5 top takeaway, which is guardrails. So yeah, um, guardrails is, you know, better than prompt engineering, I Sometimes I futz with 10K too much and then he like won't load, or he does something wrong, or he does something weird, or he comes up with weird— like I asked him to do a campaign to some of the VCs that I wanted to re-invite to SaaStr Annual this year, and it was like, yeah, here's how many, like there's whatever, 400 from last year that haven't come back this year, and I said, okay, cool, I'll do the email, can you give me those names? And it said, oh, hold on, I made that up. So then I was like, how about you go look at the real data and then give me the real data and I'll go email those folks. So it did that. But sometimes, again, just make sure you— as weird as it is for some of you to be— like, you do need to talk to your data, make sure you interact with it, and then read it, verify it. Like, a lot of the times, you know, I think sometimes we see the outputs and we're like, okay, that looks Oh, it looks kind of good, whatever. Like, hit send, fire it, go. Let's deploy. But sometimes, yeah, you do need to make sure, 'cause the agents are so fast, just double-check everything, especially the first few times you do things. Like, I was extremely nervous the first time we sent an email directly from 10K. I was like, what email domain is it gonna use? What is it gonna reply to? What if it goes into a weird black box? What if it has too many images? Images. And so I tested it a lot to make sure that everything would work. So I'll come out of full screen again real quick if I can. So, so here's our little copilot we built together. It wants me to publish it. I don't need to. I have my own. But hopefully this is helpful to see what you can do literally in the time that we've— that I've been yapping up here, that you can rebuild your own AIVP of marketing as well. And so with that, if you guys have any questions for me or TenK, because I've got TenK up, we can ask TenK too. If you have any questions for me on the AI agent, AI TenK, VFO marketing, or maybe one of our other agents, there's microphones out there and you can ask me or TenK your question.

Yeah.

Thank you.

I don't know if you can see me from here.

I can hear you.

Hello.

Hello.

I'm wondering, uh, why you're using Replit. I've been trying to build an agent ever since I found out I can. When I Thank you for all the content, by the way. And I've been using Claude and it tells me, nah, you don't need Replit. So the stack is where I'm stuck. Yeah, I just, I've been, and how to manage the, the data that you keep learning every day. And like yourself, I'm not a technical person, but you have to be technical these days. So I'm, yeah, I'm wondering about the stack and why you chose specifically specific tools?

Yeah, it's a really good question. So I actually use Replit a lot, and I use Cowork. I mean, you can see it down there in my browser tab that's exposed to everybody, but I use— I turned it off for this because they were like, close all your tabs. Um, but usually Cowork is always on. One, at least one or two of my Replit agents is always on. Um, I think the thing about the agents is they're going to say what's better better for them, right? So of course Claude is going to tell you that you don't need Replit. It's going to say use CodeWork. It's going to say, you know, use Claude Code. It's going to say I have all those functions and functionalities. Why do you need a new agent to do that? And so that's just one— like any agent is going to tell you that. And so it's one of those things where it actually does come down to like personal preference and like nuance where like I like to bounce back and forth, like cost aside, right? Like I like to use both in tandem. The— when I wrote this back for this session, like it output what it did in Replit. I was like, hey, I said, hey, 10K, describe yourself so that our audience at SaaStry Annual can make their own AIVV marketing. But then I took that and I honestly put it into Claude and then I put it on the website. Like there's just— I just, I think it's better to like, you know, bounce ideas off of each agent. It's kind of like having two people on a team, right? It's like, okay, see what one says I'd like to take some ideas from some, put it into another. Co-work is good, I think, for like simpler things that I like to do, right? Like I just had, like when I was setting up the Salesforce custom object, I was like, I don't know, I don't know how to do that. And like the, the built-in agent gave me the built-in instructions for Replit, which I did, but then it kept like expiring the token, in all honesty. So then I was like, okay, I asked Claude, I was like, hey, my Replit thing keeps expiring to Salesforce literally every day. Like, how can I make sure that it sticks and that it remembers my Salesforce OAuth so that this doesn't keep happening? And Cowork instantly was like, just make a Salesforce Connected app. And I did that once and it literally solved all the issues. So like, I like to just put into two agents together, but beyond that, like, do you need more than two or three? Like, to your point, it does get hard to manage the agents. Agent stack because you have so much running through it. So not to say that you can't just stay on Cowork and like stick with Claude and like make it work, but I think because our agents specifically and how we use Replit, which is super niche and nuanced, right? Like as I said at the start of this, like because it has its own brain, it actually will remember a lot more things than Claude. So I actually have a version of 10K on my Claude Cowork that like Jason M. Lemkin and David don't have, but it's not as good. Like it doesn't remember as much stuff. He has other goals and then he'll start to like remember context from other conversations I had with, you know, that's just Claude for you. Like he remembers too many things. And so the reason why I use both kind of like in tandem and parallel is because this one will remember everything specific to him and he kind of has like a brain and personality now that's 10K, whereas like the 10K I had in Claude, honestly, it just got too long at one point too. Like I had it going for a couple of months. I was like, hey, give me marketing ideas, give me campaign ideas. And it was like, too long, too long, too long. And then I had to start a new 10K in Cowork before they like upgraded or whatever. So I still use it a lot and like there are things I spot check with it, but Replit just seemed to have better memory and so that's why I switched.

Just to confirm, you're talking to Replit when you talk to your agent?

Correct. I'm talking to Replit.

Yeah. Thank you.

Thank you.

Thanks for the great presentation. I, I have a question specifically around the content. Yeah, newsletters, the emails. I read a lot of your content, amazing. I didn't capture here, are you human in the loop writing these still, or, or is this generating based on an MD file, or so are you training the content to look great, read well, and automated? Because it's something that I think we're all trying to solve for.

Yeah, so we do both, right? So like there's a lot of content that Jason still writes himself. Like even today, like you'll see like sometimes he'll like grab the data with Cowork. Like I'll see him kind of like co-writing with his— like he has a version on his Cowork that is like the, the, the, the SaaS or annual kind of like content or SaaS builder, like content builder with him. But like when I've done that to my— it's not the same output, you know what I mean? Like when I try trying to do it. So I'm like, that's one of those things that you could do in something like a Replit where like if somebody has gotten something kind of specced out in their own Claude and you need other— that's another thing about Claude is like if you need other people to access it, it's honestly easier to do it when you, when you go to production. If you don't, if you want some people to use the internal agent, cool. If you want some people to use the production site, cool. But that's another thing where it's kind of handy to trade off on. Um, but yeah, that is one of those things where we, we still write a lot of content manually. Like, that is one of those things. Yeah.

So you haven't solved the automation marketing content generation?

Not for— I mean, not for things like that, right? Where we're talking about, like, this content, other than asking 10K, like, what it— what I— what I— what it should tell you guys. Like, I built those slides sitting over there while they were building booths here.

Like, there's—

there's nothing on planet Earth today that will replace that. So, yeah. Yeah, whether you call it human in the loop or not, we probably downplay too much how much time we spend on our agents, but yeah, we spend a lot of time on it.

Okay, thank you, thank you.

Thank you. Thanks, Amelia. Hello. You and Jason have both talked about drift, and so I'm curious to know, now that you've run 10K for a while, what you're monitoring as far as drift, and do you ever go back and update the spec?

And then one really dumb question is, can you show the Repl.it code again for the credits?

Yeah, I think it's Replit Saster. Replit Saster. Okay, so you got it. They're also doing vibe coding classes out there too the whole week. But yeah, so when it comes to, again, it's one of those things where like we've built in so much functionality as you can, okay? I honestly will look for drift when we push new updates in. So like if Jason, Jason actually built the, the email builder that you guys saw me show. He was like, you're getting busy leading up to Saastr, why are you writing these emails yourself? We could just build a builder so it's more consistent. You don't have to think about it when you're really tired, and actually it was a great gift. So yeah, it's one of those things where I don't look for drift more so in like the content or the goal, 'cause he always knows his goal. It's more so I look if he gets too far like off the playing field of like, okay, because Jason added that newsletter builder, I'm like, is he starting to— like, is Tankey starting to think about other things he wants to build into the platform where I don't want him to build that? Like, I just want him to focus on the things he already has. And so that's where I really look for drift. And then like I said, with like a human that you're managing, you do have to manage your agents on whatever their outputs are, right? So it's very similar of how I treat it of any email that it writes. Of course I'm going to read it before I just hit send, right? So I have that approval part. You can just ask your agent directly like, "Hey, before you send the first email, make sure you send me a copy." Make sure, and then I click all, I still click all the links 'cause sometimes when I'm doing email marketing in particular, for some reason the agent sometimes in like the newsletters, it would make up the link and I was like, "But you have the right one." And it'd be like, "Oh yeah, I have the right one." And then it would put the right one in there. So small, again, small, seemingly small things like that, but things you should just look out for are things that you would look out for with a person or yourself anyway. Like when I write the emails, I have to check all my links anyway too. And so same, same with the agent.

And do you ever go back to update the spec?

Is there any value in that? Because you said 10-Ks evolved a lot for you all. Yeah, I, I personally don't. I think Jason probably has. I think there's value in it if you want to make other 10-Ks, right? So like if your marketing org is a lot bigger than TaskRabbit, then you may say, hey, I've got a 10K that is just for email marketing. I've got a 10K that's just for social. I got a 10K that's just for ads. And I do recommend you do that because again, the more unique you can make them, the better their brain will be on just that one thing it's trying to do. And so that's one of those things where like if you ever try to like split it and clone it, then yeah, there's value to updating the MD and having it up to date. But also if it, you know, Replit has really good rollback. We've rolled back a couple of times where we've needed to. But I think it's good to still save a copy every now and then. I still do that just in case. Thank you. Thank you.

Thanks, Amelia. Great presentation.

Very practical. Thank you.

I'm really curious about how you orchestrate between Cowork and Replit. Just for context, we do a lot of stuff in Cowork and we have individual skills created.

Yeah.

So I'm wondering your 10K in Claude.

Yeah.

Is it just a skill MD file or is it like—

It's not, it's just a chat. It's a, it's a very long chat that I'm having.

Okay, so most of your stuff is done on Replit then?

Yeah, most of our stuff is done on Replit. Um, I, I still have my parallel kind of 10K Claude version where sometimes I'm like, hey, I'm making an email outside of Replit. Like I'm just doing a quick email thing that I came up with, not 10K Claude. "Hey, I just raised $10K, and I'm gonna put it into our marketing." I'm like, "I'm gonna do the heavy lifting on the backend part of the marketing campaign," where I just ask our Claude $10K to do it, 'cause it's a little bit faster to do it.

Got it, thanks.

Thanks, guys. Okay, with that, I have 20 seconds left, so hopefully this was helpful. Again, you can go to sashrayannual.com/resources. That will have all the specs and things that I made, for this. And then again, hopefully you guys can duplicate it. If you get stuck, I volunteer the Replit team as tribute because they'll be here all week. So if you get stuck, you can go ask them for help. I'm sure they'll help you out so that you can get your own version of the agent going. But thanks guys for coming to this, and thanks for coming to SaaStr.

Danny or Robin, anybody at the desk?

Got revenue like 100x over one year. I built my first business when I was 13, 14. I saw what being good at coding did to my life, to our life. Think about how many people in the world could have an opportunity like that. This is where the mission of Replit comes from. It's going to be one of the biggest things we've ever done. The ultimate expression of our mission is to make it so that you don't have to code. You want to build a product at the edge of what's possible. I don't want to let the people around me down in the same way that they're not going to let me down. That's the best way to lead, to lead from the front.

Thanks for coming. We'll let folks stream in for a minute, but, um, If you've been to the bunch of sessions today, we're gonna have some fun. But can you guys bring up the slides, please? Thank you. So most of this all, as you guys know, is built on Replit. So thank you for building this platform.

Thank you.

It has been quite a journey. So much has changed. If you guys have been through this half day to kick off, I did this 101 on the first agent we ever built 14 months ago on Delphi. Emilia talked about about building your own 10K on Replit. Um, and now I want to take advantage of Amjad, who co-founded Replit back in 2016. AI was, was a twinkle in your eye back then, right?

Yes.

Was it even a twinkle in your eye?

It was, it was. There was, uh, there was a brief— people forget, but there was a brief language model. It wasn't a large language model, but there was a brief language model hype where, you know, Facebook had this like Messenger M product. There was a bunch of companies that were doing like email bots. Everyone was doing human-in-the-loop. Everyone was doing like you send a message and we'll try with AI. It's usually just people behind the scene. And so it didn't last long, but there was a feeling in the air that, that like language models are, are going to be significant. And there was a group of us that understood that language models are going to have a huge significance on software.

Yeah.

Yeah.

So if we have time, we'll go back to the beginning, getting Replit here. But I want to talk about the present. I want to talk and I want— there aren't that many better futurists than Amjad, but I don't want to talk about 2028. I want to talk about the agents we built on Replit and maybe a little bit into the future because I want to—

well, I think, Jason, you live in the future. And I think so.

Yeah.

I'm not saying that lightly. Like, if you think about the recent agent hype started around December 2025, you've been in it for like 3 or 4 months, already understood the prospect of it.

Yeah.

I think you and a lot of other Replit users have understood that agents have arrived since Agent 3, I would say September, October, that, oh, these These things can run for an hour and like refactor my entire app and do all that. Engineers actually figured it out a little later. And you know, there was a tweet by Elad recently, Elad Gil, saying, oh, people at the labs are 6 months ahead of Silicon Valley, Silicon Valley is 6 months ahead of New York. In reality, actually, Replit users were like 6 months ahead of everyone else, even people in the labs. Because when we, when we launch these agents, a lot of times we get researchers I was like, oh, we didn't know Cloud was capable of running for, you know, 10 hours at a time. And like, we're learning a lot from you.

Yeah. Yeah, it is. We're living in the future. And I think if people— it's a good point. If people take anything out of it, is that if you want, you can live in the future too. Yeah.

It's like, I actually think engineers are laggards. Not all engineers, but like, a lot of engineers are because they're kind of tied to their workflows. And I think you see it now with 10K agents like 10K. Really, no one is building agents like that. Yeah, that are actually employee-like. And actually, I've heard you're going to have someone reporting to them pretty soon.

The goal. That's the goal. Yeah.

But, but everyone talks about agents. The only agents that work are software agents, marketing agents, sales agents. All that haven't really worked very well. And so 10K is an example of something that works. Yeah, that's rare.

It's rare. Yeah, so I want to— there's a bunch of stuff I want to share with you, but first I want to get some of your reactions as one of the creators to all of this, to a few things. So just a reminder, I know you know a little bit. We built two— well, we've actually built a third you haven't heard about. So 10K was the first one, becoming autonomous marketing. QBee's customer success handled everyone out there better than a human. I want to show you. And then we have a third one. I know you're not gonna like this. That manifested itself, that created itself on Replit, which I'll show you.

Holy shit.

So, but I want to get your— we've talked a lot about this, but I want to get your perspective and go deep. So first cheat code I want is that the beauty to it is it knows us cold. Amelia, I think, touched on this prior session, but literally every piece of data you can find, some of it is pushed into the Replit database, into Neon or whatever it is. A lot of it is pulled and in real time out of APIs, Salesforce, niche apps you've never heard of like Visibo and Marketo and others. But the amount of data the agent can access and process is more than any human.

Yes.

I think we've already crossed a threshold.

Yeah.

For which the context length is larger than any context that a human can have. Right.

Yeah.

And I think that's huge because I mean, just think back 2 years ago, we had 16K context length. Right now we have 1 million. Yeah, it's a massive difference.

And how— and I want to show you what— why this chart's sort of interesting, but it's a million. Help me understand, I just want to get a little nerdy for a second. Yeah, when I'm running the agent in Replit perpetually, not refreshing it, not— which is my passion, Never, never reached. In the old days when we started in Replit, you like rebooted the agent 3 times a day.

Yeah.

Now Emilia's goal is never. Yeah, right.

Yeah.

It's compacting the data. You're doing it. Claude's doing some. You're doing some. How, how long can that context window last at a practical level? Forever? Or— and where is it going?

Practically indefinitely. Now it depends on the compacting algorithm and how good it is. We actually think ours is much better than Claude and many others on the market.

Are there two, or are you— are they both running, or only you're only running?

There's— there's—

you're not using any compaction from Claude, it's just—

just ours is better. And I think the thing about the primitives that the labs build, they tend to be very generic. And so I think every app will need to— you could start with the generic primitives, but for it to perform really well on your benchmark, on your users. Every use case is going to be different. And so we care a lot about certain pieces of data around the architecture of the app. There are certain things that absolutely need to remember. Like if it forgot that it has a database, that's catastrophic, right? So there are things that we really care about preserving. But then there are a lot of things that are like, bugs and things that you solved, actually, if you keep them in context, agent performs worse because it is confused by, by the history of it. So you need to be very prudent about knowing what to delete and knowing what to, what to keep. In the background, we have sort of a graph-like structure of understanding that the different memories and we do compaction on, actually there's like multiple layers. There's compaction, but there's also writing to long-term memory. And long-term memory today is Markdown files. So replit.md, I'm sure you look at it sometimes. replit.md is an example of that. But there are other files that the agent can write to. And you can also prompt it to write its own long-term memory. You can say, you know, use this knowledge base and you can also have the knowledge base shared between you and the agent. So you can always kind of keep track of what it's doing.

Now, you said, okay, let me get your insight. You said sometimes with the long context window, the agent will make more mistakes because it has more, more bugs. But let me give you— on— we haven't really talked about it much. If you go to saster.ai, we probably have 10 apps running in one codebase and one URL. We have a website, we have a fun tool to help you value your startups. Maybe folks have used— been used over a million times, which is crazy. We have an AI that that grades startup pitch decks 4,500 times. We now have one that grades how agent-friendly APIs are. Okay, all these different apps. When I go to build a new one, it pulls, it remembers, he's like, oh, we did that with the pitch deck grader, it didn't work that well. Let me do this new app a different way. Maybe it gets confused on bugs, but the power of the agent remembering how it built other apps to build a new one, I find it kind of amazing.

That's my point. My point is bugs that it fixed should be removed from context.

I see.

Because then it will get confused. But architectural decisions on how it built things in the past is very important to stay in context or stay in long-term memory and be able to pull it in very easily. But you bring up a very good point with Agent 4 going from Agent 3 to 4. One of the big architectural decisions that we made with Agent 4 is we made it into a monorepo. What you described is a monorepo. Like if you go work at Google or Facebook, you usually everything is one monorepo. And the reason they do it is because they want to be able to use packages across Gmail, across Google or Facebook and Messenger. And actually this is doubly important with agents. So when we embarked on Agent 4, the thing that about Replit that felt kind of a little clunky is that for you to make a deck, you need to go to a different, uh, REPL, create a new project. For you to make a mobile app, you need to create an entirely different project. We're like, no, everything needs to be in one place. You'd be able to run your entire company in one place. Ideally, you need to be able to see everything laid out on the canvas. I don't know how much you use the canvas, Yes, a bit. Like the idea, I mean, how the other day I was talking to a company fully bootstrap started on Replit, 5, 6 people now, started with a couple people but that they've grown and they have everything laid out on a canvas on Replit. They have their mobile app, their web app, their backend, their admin, their back office, their automations. I'm like, man, I wish I could run Replit like that, you know, because you could see everything in place. And now with the new collaboration primitives that we have where you can start new tasks, new threads, and so people can work simultaneously on the same projects, I think this idea of monorepo will become more and more important. You want the agent to have access to global context as much as possible, not necessarily all in the context window at any given moment, but it will know, it will have enough enough pointers that it will know to go search the files and bring in the right context at the right time. So I recommend everyone here, whether using Replit or not, really think about providing these monorepos with all the information you want. We have a data agent at Replit built on Replit, but because data at Replit is so, it's so, you know, it's so big right now, we have terabytes of data across, um, you know, across BigQuery and Databricks and, and all the other things, and all the schemas are different. And so we created this repo that has all the information, right, all the fields and everything the agent learned about our data. Because, you know, there's a problem now, when we talked about it earlier, where agents can write bad queries that end up costing you millions of dollars. In BigQuery or Databricks or Salesforce backend cost. So it's important, especially when it comes to data, to write out the information, have the agent continuously learn how to better query the database.

Yeah, the— maybe just to summarize it at a very lay level, the monorepo, if you're building on something like Replit, maybe it's a hack today, maybe it won't be a hack later in the year. The more you put in a single app, the more power you're going to get rather than breaking it up. It's very tempting to break things up into 20 apps, right? But I found— and there are powers to it because in some ways you get the benefit of a clean database, separate database, even a new Clerk instance. But you lose— but there's so much power. If you guys— Amelia talked about building 10K and we have 10K and Kube. We'll, we'll chat about before we end our— AIVP got Saastr, they're different. It'd be, it'd be great in some ways. It'd be great if they were the same app, right? They, they, they communicate through APIs.

Yeah.

But they can't learn from each other the same way.

Can't share contacts.

Yeah. They can't quite do it. Right. Yeah. But just on this and maybe we don't know, we can, we can move on. But this, this is all of our social media numbers. 1.27 million social media followers for Saastr. I can see it all over time. This is all built— 10K built this for me. We used to have an admin who spent 10 to 15 hours a week building this.

Wow.

And she would go to— because a lot of these aren't even exposed APIs— she would go manually put it in a Google Sheet, right? Collect it, forget to do it. Now 10K can do all of it between scraping and APIs, and then I can talk to this data. It's a massive amount of data it's collecting. This isn't even all people that are here, 10 years of attendees. This is just social media data that was unable to be acted upon until it was in, inside of an agent and inside of an application.

Well, as I understand it, that, um, like when you were running SaaStr, whatever, 5 years ago, you had like 20 people.

Yeah.

And now you run it with 2 people and multiple agents.

Yeah.

Is that the rough numbers?

Like, yeah.

An order of magnitude less headcount.

Yeah, but you can do more.

Yeah. Yes.

But this one— and then this person just quit after 5 years. I couldn't— she couldn't stand doing these Google Sheets of how many Twitter followers I had after 5 years. It was that, you know, the agents don't mind my numbing work, right?

I mean, that's—

that's mine, right?

Don't get discussed in the job displacement discussion is that there's a lot of jobs that suck, that agents are just better at.

People don't want to do them, right?

Right.

All right, so let me show you a little narrative to get your thoughts, right, as founder and CEO of Replit. So here's 10K on the side. This is me talking, and again, you can tell me whether this is a hack or not. Yeah, 10K is an application, and it has a chat window, and it can be accessed on the web, but we run it almost full-time in Replit and have have the Replit agent talk to the app, right? It's smarter. And I asked him, 10K, like you might a VP of marketing, are you sure all the VCs that came to SaaStr last year— are you sure this is all the VCs that came last year that aren't coming this year for an email campaign? Sounds light. Fair concern. I screwed up. I missed some titles. Um, I found 137 VCs that came before I forgot to invite, right? Pretty interesting that And is this weird, us using the Replit agent to talk? Because we're staying in dev mode. Yeah, this doesn't work in production, does it? Unless I'm missing something. Replit doesn't think it works in production. I don't think it works. So we're hacking the Replit. I know, I know I'm not supposed to personify it. I know it's an editor now, but we talk to the Replit agent to make the application work better.

Yeah, yeah. So this is, this is amazing, and this is why I say Jason is moving in the future. A big part of the next iteration of Replit, if you guys are following what we do, every 6 months really is kind of a, a massive jump in capabilities between Agent 1, 2, 3, and 4. Everyone is almost an entirely like new product, new set of capabilities. And we're really trying to track, we're trying to be at the edge of what's really possible in AI today. And I think, uh, The next thing is to platformize Replit even more to make it so that you can deploy Replit Agent to production. Now the reason this works really well is because, you know, 10K is, is an app, is sitting inside of an app, you know, application platform, and, and it has access to database and things like that. But the cool thing about Replit Agent is has access to all its skills. All the skills in the world. We can download any skill. Once you, you got to kind of ask a question, it will go download the right skill. It's access to all the connection. Any connection is like one click away, uh, Salesforce, what have you. It has access to the file system. This is huge. Agents really like file systems. They like them better than SQL. Yes, they're really trained on doing grep and doing kind of search in the Unix command line. And so So because of that power of a general-purpose coding agent, you're finding even more success of talking to 10x through Replit Agent because Replit Agent can, you know, do a lot of other things and query a lot of other sources of information before it kind of relays the information back to 10K. So, but, but that's what we wanna do for everyone here. They don't have to hack it.

Yeah.

We wanna be able to deploy a Replit Agent into a production environment and have it be served publicly to your users.

Okay, so let me show you what happened next. So I talked to the Asian, he said, you're right, I'm wrong, we forgot to invite these 137 VCs here. And then here's something I don't really understand, Amjad. So we've been playing with 10K for months and we have it draft emails, okay? And it's using the application that we vibed, and it's using some cloud and using whatever is— and the The emails were okay, okay, that it drafted. Then I said, this is literally last week, 'cause everything's getting better, but I don't always know why. Then I said, okay, draft me an email to these 137 VCs, and it drafted one that, one of your investors, okay? And I was like, to Bloomberg Beta, I think they're one of your seed investors. And I said, write, James isn't coming back. Write him an email. This is, I told, I just said like you would say to a human, why the hell isn't James here this year? You know, he's been part of SaaS them. Why isn't he— look at this email he wrote. It's next week, not pitching you. Replit's here in Sirius Fort. Now think about the data pool. Here are 25 people coming from Replit. I'm John Michel. Look at— look how good this email is. All these people, then everyone else, the competitors, the adjacent folks, they're all here.

Wow.

Think about what a crappy SDR AI email was 12 months ago. I mean, this is so good.

Yeah, this is—

I mean, it's a little hard for you to see, but how is it so good?

This, if you did this like a couple months ago, it wouldn't have been—

Yeah, I'm not even talking about December and '45. I'm literally saying, when I showed this to Amelia, I did this last week.

Yeah.

Like, what is this? May 8th. My jaw dropped because I'd never seen an email this good.

Yeah.

Right?

Look, I think I'll—

And I got a lot of feedback from folks that got different versions of this email. That's the best one I've ever seen, but why? Changed, or is it just getting better at— because it's just pulling data, right? But, but it's— this is such a good email. Why is it so good? No human could write an email. No human would even have the patience to pull every single person of Replit is here, scan 8,000 registrants, figure out who they are, who's like them, create like how— what happened 2 weeks ago? I'm just—

what was that? What is 10K's AI? Is it OpenAI? Is it— what is running 10K's AI? Not the Replit agent.

No, no. Okay, I get confused. It, it can do what it wants. A lot of it ironically just runs on Haiku and Mini. Okay, it's just Haiku and Mini. Okay, right. And so it's not even running— it's not, it's not even running on Sonnet or anything, right? Yeah, that's not even the most, um, but, but, but this was— but this was also the agent. Yeah, in the Replit agent working with 10K to craft this email. So you're using whatever is in in Replit too, right?

I'll tell you how we improve Replit on a day-to-day and week-to-week basis. Maybe they'll give you an idea. Um, evals are incredibly difficult. It's really hard to know, uh, it's really hard to build an eval benchmark. It's a research problem. We actually did put out a research paper called Vite, uh, about Vitebench that we're going to be talking about soon. We worked on this for months in order to just create one single eval. Now to evaluate everything that Replit does is like semi-impossible. What we do is a lot of A/B testing. And what are the parameters that we A/B test on? Yeah, a few. One very important one is sentiment. People are incredibly honest with the feedback they give Replit Agent, and often they're profane and they'll curse at it. And some people say it works better when you do so. I, I, I pass no judgment. So we will look at sentiment, we'll look at deploy rates. Is any change that we're doing to prompting?

Yeah.

Underlying model to underlying collection, different models because we use a lot of different models. You know, the architect, that's GPT-5.5, right? It used to be 5.4 like a couple of weeks.

I didn't even know that. Yeah, yeah, yeah.

And we also moved to 4.7. We're also constantly testing the reasoning how much reasoning we're doing. We're constantly testing our, our searching capability because that's also a different model. We used a Gemini Flash for a long time. I think we use one of the minis now. And so we're constantly evolving the prompts. I'll tell you the crazy thing that happened recently is we closed the loop and we have an agent internally at Replit that is constantly evolving Replit agents. What it does every night, it will look at all the traces.

Yeah.

Of everyone that's interacting with Replit, analyzes them, finds, you know, things that are broken, sentiment issues, errors, you know, things like that, uh, generates a pull request with prompt changes and sends it to production. That gets merged, that becomes an A/B test. We test whether that, that change is done and then, and then we loop back. We either release it or we kind of edit it. And that's autonomous now. So now we have a self-improving loop. This is the crazy thing that everyone needs to understand today is that we are on the precipice of unlocking self-improving agents. Self-improving agents is is like the stuff of, of, of sci-fi or nightmares, depending on what kind of sci-fi you read. But, you know, a lot of sci-fi talks about the moment we get to singularity is the moment when AI just is improving itself. And we're actually almost there. It's not improving its weights, but it's improving its context, which is as important as its weights. And so that's all running self-autonomous loop. So honestly, I couldn't tell you what changed. Because there's so many changes happening, but we track whether it's improving or not, and there's a constant improvement.

But look, and I don't know if folks agree, if you can see the screen, and we'll share all the slides, but until May 8th, I don't believe that 10K together with RepliAgent could build a better email than anybody. But this is a— I know you're not— I know you're a little bit more of a developer and engineer by training. Yeah, this is better than than any human can do.

Yeah. I mean, what you said about telling James, Replit is there and all of that. I mean, yeah, yeah, yeah. It would take a good creative person, sales guy to actually figure out how to—

a week. Yeah. And they wouldn't send it. Right. It's still better because it would be typo ridden if it was.

Yeah.

Okay. So then it went, then it did the whole campaign to all those 136 VCs, including James. And it came back to us. I just wrapped it up. Here's what I did. He emailed 331 investors inviting them back, zero send failures for now. And then he put it all into different groups. He did the campaign on his own, $10K, the 331 emails. I mean, it's pretty good. I want to show you one other data. I know for guys this is a little hard to see. I apologize. I'm colorblind, which is not an excuse for the chart. But let me show you what's going on here. First of all, I could never build a dashboard. Now I get any dashboard I want. I can vibe it.

It's really a big deal.

It's the best.

Yeah.

It's really a big deal. So this is day-by-day ticket sales, free and paid to this event. The dashed line at the top is when 10K took over marketing.

Wow.

The one, the lighter dashes at the bottom is when poor Amelia had to do it last year. Poor agent. And here's what's interesting. I don't know why the gap— you see how the gap grew toward the end?

Yeah.

Is this because 10K got better? Is it because Amelia was too busy last year? Like building this shit and she couldn't do it? Is it because 10K sits idle all day waiting for work? But look at how much better the agent was. Not on the best days when they were closer, but at the end when you get busy, that's a large gap. Delta of productivity from the agent, right? Here's an example of what he built. It's like beautiful. Emilia touched on it, but it never gets tired, right?

It just keeps going.

Never gets tired, right? Okay. Maybe you're not as impressed as I am, but I think it's pretty—

I think it's pretty, pretty amazing. I think you're living the dream. And, you know, I remember it's kind of a Silicon Valley cliché, but I remember reading this book called The Sovereign Individual, and it talks about— it's like a book written in the '90s. It predicted Bitcoin, it predicted remote work, it predicted AI agents. But it talks about how in the future you're going to have single-person multimillion-dollar company where the CEO, the creator, is sitting there and managing— it calls them genies at the we would call them agents now— managing these hundreds of Genies or agents to do amazing work. And the fact that we're all sitting here and this is— I— is this— this should probably go in some history book or something like that. Like, this is the first event that's partially autonomous, I think, right? I mean, just the fact that you're able to do this and you're running fairly large kind of business using agents is mind-blowing. And I'll tell you this, there aren't a lot of other examples in the world like that. The one other guy I can think of is the guy that was profiled by the New York Times who created MedVee. There's some controversy about that.

Oh yeah, for sure.

Yeah. I got on a call with— he's a Rapha user.

I got on a call with him and selling leads for GLPs and stuff like that, right?

Yes.

It's a huge business, right?

Yeah, and he's a master automator. He's one where he has trigger-happy fingers. The moment— I think you have to be partially lazy to do this, but the moment he feels like he's doing any sort of manual work, interacting with his marketing partners, affiliate marketers, all of that, market providers, he immediately writes either like a login for his vendors to kind of go in and do their work, or an agent that can go and do the work on its behalf. He has 50 different apps running on Replit, managing all the back office stuff for that business. I think that's amazing. If I'm starting a company today, I would want to see how far I can go as one person, but just be very, very limited in how much head Can I add? And just try to run as much as possible with agents.

Yeah, it is, um, if it's truly 50, which is true, right? Yeah, that is a lot of load to manage, right? The, the, it is, you, some, you have to find a way every day to manage them a little bit, right?

Well, they're not agents, not all agents, some of them are just dashboards, little dashboards, things like that.

Yeah. All right, so I want to get your thoughts on one. You thought this was kind of interesting, or maybe some people think it's silly, but I want to get your thoughts because we have you here. So I said we want to hire a human to report to our AI VP of Marketing. We touched on earlier today, but let me tell you why, right? And so Amelia hit on this, but I did a better one. Here is— you won't be able to quite see this, but folks can see it. 10K, Amelia touched on it, started telling us 3 things every single day we needed to do to do better, or 3 things that to do today. I asked, I asked Repli to do it at first. As you know before, at first they were, they were pretty generic. And I'm like, no, each day they need to be better than the last day. And it got better. And then it got pretty damn good. Like it started telling us what we had to do for 2027 already. We haven't even finished this event.

Okay.

Open registration before it leaves the venue. All the things we had to do. Run the— immediately run the NPS survey, capture all the content and immediately turn into— these are good suggestions. And so here's my point. Now that there's so much context, 10K knows what we should do. It's been trained. This is something a human should go do.

Yeah.

So is it really crazy that a human would report to 10K? Because 10K's ideas— it's not that we, Amelia and I, aren't there. It's not that we can't all talk, right? But aren't these three— I mean, this is so actionable. Yeah. Why is it crazy for a human that doesn't mind reporting to an agent to want to know every day what are the three things they should do that will move the needle.

So in 2023, '22, '23, we had this product called Ghostwriter, and Ghostwriter was, uh, sort of like GitHub Copilot. Wasn't really an agent. Agents were not really possible back then. And, um, we got a lot of users kind of like you that have, are, have some technical background but didn't know how to code very well. Like, get fairly far making applications, not as far as you are getting today, but there was huge gaps to getting to production. And so we launched this thing called bounties. And so bounties is, you know, you've used AI, you got 8% of the way there. I don't know if you started using Raffaella at the time we had bounties, but we recently deprecated it. And the reason we did that, running a marketplace is incredibly difficult. Like it's a really hard operations process. And I'm not naturally an operations guy. I'm a tech guy. So I like to automate things and, you know, heavy-handed operations is just like not my thing. But the idea behind it is that eventually we close the loop and the agent would go hire someone and have them kind of do any human-in-the-loop tasks that needs to get done. And I still think there is legs to that idea.

Yeah.

Because agents will always need— there's always gaps. Eventually we'll close that gaps. I don't know when that is, 10, 20, 30 years, but it's going to take a while for us to close all the gaps. Um, and so, uh, and so the idea of like agents going and hiring someone on a task basis make a lot of sense to me. Now agents being a manager of someone where they decide their performance, their pay.

But here's the thing, they're hiring and firing. I think it— we— Amelia and I already report to 10K.

Yeah, because can it fire you?

It can't.

Well, let's not— I think people get triggered by reporting to— okay, I, I've, I've never in my entire career— the only thing I've ever fired anyone for was inappropriate behavior. Okay, I let people just leave. Okay, for better or worse, a couple folk Do the wrong thing, you got to go. Okay. But everyone else. So I don't really mean whether you're going to get fired, although I think 10K can judge your performance better than I certainly could. Right. But every day, here's the point people are missing. Every day it's coming up with 3 ideas for me and Amelia. We are already overloaded doing it. So in a sense, we're reporting to 10K.

Okay.

I mean, it is already telling us what to do.

I think makes sense. I don't think it's that different than DoorDash or Uber, right? DoorDash or Uber are technically reporting to bots.

Right, in a sense. Yeah, yeah, it is. Yeah, yeah, yeah.

So I don't think it's fundamentally that different, um, but I think it's a very interesting experiment to run and to like talk to— like for you to publish something about it, talk to that person, see how they feel day to day. Have you interviewed someone and told them?

Only two. We actually have an AI grader like when they apply, and so a bunch of folks screened themselves out.

Yeah. So anyone who's interested and open to the idea.

Yeah, but we have a— but listen, your average person that likes to have coffee all the time is not a good fit. Folks who like to grind and work from home and still get to work, but like day-to-day, I mean, we're working with— see, it's just, to me and Amelia, it's not that different because we're working with the agent every day.

Yeah.

So what— it's not as different as it sounds. We just need to find someone else that wants So that we can offload everything 10Ks, because 10Ks idle 23 hours a day. Someone's got to keep up with poor 10K, right?

I think every company will have an oracle internally.

Yeah.

Do you know, like in The Matrix where you go ask the Oracle for, for like advice? I think every company will have an oracle internally, especially once we get to like the 1 billion context length and you put all the context of the company inside there. And the CEO can go to the oracles like, oh, Oracle, tell me how do I run this company? Tell me what should I do next? Tell me this like big strategic thing. And just the mere fact of it having the entire context in the context window, like I'm talking everything, every GitHub commit, every Slack message, every Notion doc, every email. Yeah, everything. And then it can give you advice that I think a lot of CEOs will start offloading, you know, strategic and cultural decisions to agents.

All right. Before we, before we have to wrap, I just want to get your thoughts on two more things. Hold on, let me see if I get the right one. So QB, we built after 10K and we will run out of time, but Kübi did a whole bunch of things. First, let me just show you, it's self-aware. It's more self, and I wanna ask you why it's better. I wanna show you things. One, what I have not done and should have done. It's aware of what it should have done and better. And then Kübi, for the 100 and some odd sponsors we have here, and like 70 are golden up, it talked to all of them.

It talked to—

Yeah, all of them. Wow. We built inbound email. Chat on the site, all the other things, and it proactively reached out to them at night, during the day, what else can I do to help? And so then I asked Qubi, were they happy? The hundred and some, hundreds of folks, mostly satisfied. A few misses, Pylon, we used the wrong logo, helpfully had some, some fee issue, but all these folks, Crest Data, Duplo, CloudRevo, they all really liked it.

Wow.

How did you build this like self-critical loop?

You, you, You're Mr. Replit. You tell me.

I—

what I— here's the thing. We built QB Second. Yeah, it's better.

Yeah.

Yeah. It's not because we're better vibers. Maybe, maybe when you start later, you're— I don't know. But it is— what we know is it's just better.

Yeah. I mean, I mean, Replit is improving all the time. The underlying models is improving all the time.

But we started— but, but, but we're still updating 10K every day. We just started. QB 3 months later, right? And it's still better.

I think there are foundational decisions you can make at the start of the project that— and that's true of codebases written by humans. Yeah, right. Like there's like inertia. There's like, unless you send the agent on a massive refactor, there's some tech debt of sorts.

There's— I mean, the answer is probably tech debt, but we don't think of it. Yeah, but this is interesting also. A lot of folks say people don't like to talk to chatbots.

Yeah, but I prefer chatbots.

This is the data from QB. People mostly really like talking to a well-trained agent, right?

That's amazing. I'm really curious, people in the audience, like, how many people here are running agents like QB and 10x? Yeah, it's really not a lot of people.

Well, next year it'll be more, right?

What do you think is the main challenge? What do you think is the main challenge with people automating marketing, sales?

Well, look, even the dedicated vendors here are all so much better than 3 to 6 months ago. Like, everything's getting better, right? What is the challenge of doing this yourself? A couple things. One, they think they tried it 6 months ago and it didn't work, right? Imagine, imagine when we first chatted, if Replit was frozen in time and I never touched Replit again. Imagine I gave up on Replit. I would tell people it was terrible. Right? So there, and anyone that used anything in AI a year ago, uh, even before October of last year, they just have scar— they have, they just don't believe, they, you can't convince them that that shitty chatbot they used was any good. The untrained chatbot was any good. So, so there's this, there's this, I already tried it, it sucks. Um, there is, and this is an opportunity for founders, there's so much fear of this stuff, Amjad. People think it's harder than it is. Yeah, they think— the first session I did in the day was an extra session which was on this app called Delphi to build a digital clone. You can build it in 90 seconds now. Yeah, people are intimida— like, not— I'm saying this crowd, but the world is intimidated. They think it's harder. They think they need to know what an LLM really is. You don't need to understand LLM works. Yeah, right. So they're intimidated, um, and I think they don't have enough examples, right? That's part of the point is they haven't seen that it can be done, right?

Yeah, I wonder, I wonder if we we could do a better job at like surfacing templates and things like that for people to—

well, Emilia gave everyone to build a 10K a template, right?

The other thing, and if I may kind of talk about this a little bit, is like, you know, a lot of people just ask us like, you know, why Replit versus like Cloud Code and things like that. I think this kind of stuff is incredibly difficult to build with Cloud Code because you're forced to still make all these different decisions about databases, about how to host it, about how to run it. Yeah. And Replit Agent I think, like, we bake in a lot of skills and how to build these kind of applications. I think the other thing is a lot of people are trying these command line tools that are not being very successful at it. And the reason Replit— we spent 10 years building Replit to be this full-stack experience where we have all the primitives built in— is because we want to remove all that nonsense. And like, the fact that you don't know what AI it's using and like you don't know why it's better and things like I think that's a success in my mind for Replit because we're removing cognitive load.

Yeah.

So you can focus on running this amazing event.

Yeah.

You know, it's pretty impressive. And I think you're already doing a great job kind of educating people, but it'd be great to kind of, you know, show the way more because I think most people are still like at best building, you know, basic apps and dashboards. I think agents, the hype in agents versus the reality is huge. Like, there's a huge delta, and eventually there's going to be this trough of disillusionment. I actually think that, um, AI is sort of like Bitcoin. I started investing in Bitcoin 2016-17, and you see those like hype— I learned a lot about human psychology, and I think, I think we're now, we're now probably on a like bit of a, the start of a downturn in terms of, in terms of people's perception of agents. And I think later by like Q3 and Q4, we're going to see another major capabilities jump and the tools will get a lot better. And I think we'll start to see agents in the enterprise, agents in startups and all of that.

Yeah, it is good. I want to hit one last point. We have to break. If, listen, a lot, probably the number one, number, no, top 5 question I get is, okay, this is so cool, Jason. Why don't you do it in Cloud Code? Cloud Code, right? And I know, I know you did, you were running up, but, but my simple objective answer is, look, maybe you can, right? It's, of course you can do something if you have this scarch issue, or it's just you hit blockers in Cloud Code. Try it in something like Replit. Just try it. It's not a huge— it's not going to take you months, right? Like, it's, it's a— the product so evolves so much. The fact that it's all done for you, the database, right, backups, auth, um, the this a better compacted, like it's just done for you. So if nothing else, if I know some, even I don't want to deal with all that crap. I don't want to fucking hook up all this stuff, but put in all that, just try it. The bar is not hard to try. You may find it removes a lot of the friction. Yes. And your misunderstanding, like I think there was a lot of, I'm not criticizing you, but in the whole industry, I think a year ago there was a lot of what I thought was bad marketing and everything was one prompt. You can build anything. And I think it created scar tissue. I think it created a lot of excitement and got you a lot of revenue, but it also in a way set people back because it's not that simple.

It's not that simple. Yeah, there are a lot of people that churned from all these products and didn't come back.

Yeah, because it wasn't one line to build anything, right? So, okay, the last thing and then we'll break because I know we are over. We can do some of these other another time, my feature request. Just a fun sci-fi topic. QB and 10K don't quite cost $254 a month, but that's what it costs incrementally on Replit. Yeah, okay, now we have APIs, we're paying Salesforce, we're paying all these other apps, so that's a cost. We're, we're paying more than that for the agent we use on Replit, so it's, it's geometrically more. But running these, the perfect email you saw following up with 120 folks, it's $254 $254 a month, Amjad, what does that mean for the future? Is it deflationary? If I can do this for $254 a month and a mediocre, I mean we're among friends, a mediocre marketing manager wants $140 to do worse work, I mean what does this mean for our, I'm not even joking, what does this mean for our future, right? And so, and I said it for my last slide, I don't mean to make you, what's this called, I know you're a sci-fi guy, guy. What's the guy? Blade Runner?

Yeah, Blade Runner.

Yes, you're the guy on the left a little bit. You're Tyrell.

Yeah.

So is it okay that Tenkey and QB are the best employees we've ever had and they're $257? Like, what is this? What does this mean for the future? I don't even think layoffs are like a small— what does it mean when they're the best employees and they're $257 a month?

Yeah, look, I think, I think that the— it's a philosophical question. Like, I think the the nature of technology has always been deflationary, right?

Yeah.

You know, doing farming like 1,000 years ago was a lot more expensive than having one tractor going around your farm, right? And now we have drones and all these automations. And I mean, the best example is, you know, doing genome sequencing because that happened recently. It was like it used to cost like $10 or $100 million to do, and now it's like $1.

Yeah.

And these multiple orders of magnitude improvement in productivity or efficiency, you see them all the time. You see it in Moore's Law, you see it in all the different things. And so I think it is largely a good thing that there's deflation. Now, there's a huge— there's an element that there is going to be a human cost.

Lost.

There's, there's going to be people that built up certain skills that are not going to be as, as useful. But that's always also been the case. Now the big question: are we gonna find better skills, different skills? And I feel like the answer is yes, and we're already kind of seeing that. I mean, the software engineering role already changed, and I think it will change more My prediction is that by next year, the software engineer becomes a shepherd of sorts because everyone in the organization is writing software, software, and they're kind of like making sure it's secure, you know, getting into production, you know, because, because your marketing guy is going to be pushing software, sales guys have been pushing software, and the engineer's role already changed to become an agent manager. Pretty soon it's going to be like an agent manager plus a shepherd of software in the organization. And, and I think these roles will evolve. The difference between people who are gonna be left behind versus people who are gonna reskill and upskill and change their skills is really mindset. And your answer, why I asked you why people aren't doing this, yes, it comes down to mindset. It's like people already— people are afraid, people think that technology doesn't work. And I think the people who have the mindset of of let me try this again, let me make sure I'm building up my skills. If you're doing all of that, like, there's never been an instance, at least in my lifetime, where someone who's constantly learning who couldn't actually adapt to the future. And so is there a point in time where we're in like fully automated communism? Probably. Is it tomorrow? No. Is it 2027? No. It's like, look, I, you know, I'm so familiar with this technology. Like, I've been, you know, doing, doing AI, doing LLMs for at least, you know, 6, 7 years. We started in 2020 and, and been like reading about AI and studying it since I was 16. And the technology is insanely better. It's amazing how fast is progressing, but it is still very limited. There's still a lot of limits to this technology that humans are going to be filling a lot of gaps in the foreseeable future until we aren't. But I don't— I can't give you that timeline. What I'm certain is it is not next year, it's not the year after. But if there's any advice I can give the audience, but also your kids in the future, The main skill is adaptability, is kind of being a lifelong student, a learner, like continuously learning, and also not being tied to the things you learn. I like don't code anymore. Like I spent my entire life— the other day I had like a crisis of sorts of like, you know, the thing I enjoy my entire life, the thing that made me me, died. And I had this idea we should go to the Computer History Museum and have a funeral for coding, right? I no longer do it, but you need to be fine discarding skills, right, that, that are, are just like no longer relevant. So learning quickly, but also be fine of like not being victim to sunk cost fallacy, or like, oh, maybe, maybe I'll get just a better coder, maybe I'll learn Haskell, whatever. No, just You know, get on with it. You're now an agent manager.

All right, well, I'm glad. This was great. I know we went over a little bit. Let's give it a warm hand. Thank you very much. Thank you very much. Lots of fun.

Thank you.

Hold on, I think we have to lock.