Our Agent Negotiated a Vendor Renewal, Became a CFO and a Better SDR .. but has too many guardrails

I was like, well, the agent is the decision maker. I'm not gonna decide whether to renew this tool or literally 10K is gonna decide for you. And if it says no, I'm gonna say no. Like, for this, he has the budget.

Requirements to, to do a renewal. 10K made the— Yeah. Making the— for, for real is making the decision whether to renew or not.

For real.

Right? For real.

And so he came up with a shortlist of things he wanted to see in the API. And that was one of the things I sent over. And then second, 10K said to push them on pricing because their pricing is still seat-based. And he said, well, we only need one seat. That's all I think they need. You should push them to charge you for one API power user and dump the rest.

Hey everybody, welcome back to The Agents, everybody's favorite show about the trials and tribulations of managing 21 AI agents, 3 humans, and a dog. We are back with a lot to reflect on. We got through SaaS or AI Annual, 10,000 of the best building and sharing with agents. By far, I think the best event we've ever done. On the last episode, if you listened, Amelia in particular and I went through what we've learned on building our agents. We're gonna talk about that next. And now we gotta figure out what we're gonna do the, the rest of the year with our agents. And so that's a little bit of what we're gonna talk about. I wanna do a lot and talk about our team of agents, 10K, 10K Prime, and more. But first I want to get to something, Amelia, you might not know this, which is regressions.

Mm-hmm.

And while we are off doing our, our little get-together with 10,000 founders, we have an app I built a while ago called our VC Pitch Deck Analyzer. It's free. It's at saster.ai. You can just click on AIVC and it will grade your VC pitch decks for you. It's really good. And in fact, we have done, uh, at least as I type this, we've reviewed 4,614 VC pitch decks. It's great. It's honest. It tells you whether your numbers are good enough to get funded, what VCs will think. It'll give you the honest feedback that humans won't. And then I noticed kind of the last week or two when we were busy doing our big AI annual event, it was giving a lot of Fs. So, so it gives you quantitative scores of all these metrics, growth, team, market size, everything. And then at the end of the day, it just gives you a grade to make it simple. And basically a B+ or higher means you'll, you'll probably be able to raise venture funding. You know, a B to a B- means you're not, it'll be tough. You might be able to raise funding, but you probably gotta talk to 100 VCs and, you know, a C+ and below is tough. And it was just giving F after F after F after F. And I figured it was just Yahoo's uploading lame decks or something like that. And I didn't have time to investigate until just a couple days ago. And then I realized it was broken. It was partially broken. And it's an interesting story why. I didn't break it. Did you touch it? You don't even know where it is, right? No. You don't use it. But this was interesting just for my learning. So of the last 305 submissions, 88 just failed. Just failed. And of the 216 that worked, 53% got Fs. They just got Fs. And what had happened was, this was just interesting for, for figuring out how to deal with agents and guardrails. Because what happens is, we've done thousands of decks. Obviously I haven't done 4,000 and something myself, but I've run a bunch manually. Companies I've worked with, you know, owner raising around, revenue raising around, Gorgeous raising around, others, and I run it through and every, and then there'll be mistakes. Like for a while it couldn't translate euros to the US the right way. And then it sometimes got confused on different types of forecasting and weird types of language. And so I kept adding guardrails, right? If it does this, do this, if it does that. And I guess it had reached so many guardrails that every, that it rejected everything. It reached a threshold of guardrails. And I asked, I asked, uh, Replit what it thought the number was. And I think it said by, by the time we'd added the 14th guardrail, 14th exception. Everything was an exception. Everything got bounced. And so everything started to break and we had to rebuild it from scratch and get rid of almost all of those rules, like throw away all these rules. And so I think someone smarter than me could prob— you could probably figure this out better, but it was interesting that months and months of layering these rules into this mega prompt, it was too many and it broke the system and it rejected everything.

You're listening to the official SaaStr AI podcast brought to you by Hey everybody, starting a business can get expensive fast. Website here, email somewhere else, business phone with another provider. Northwest Registered Agent gives you a complete business identity in one place with free tools, resources, and built-in privacy from day one. Get more at northwestregisteredagent.com/saasrfree. That's northwestregisteredagent.com/saasrfree. S-A-A-S-T-R-F-R-E-E. Starting a tech company is thrilling. It's also really messy. Just ask anyone who's done it or anyone who's tried. Better yet, ask EY. They've seen startups in their best moments. Go to ey.com/techstartups to learn more.

So it really got me thinking about how many guardrails is too many. You know, we're running guardrails in Essence and AgentForce and Artisan and all these, and you can see them, right? You're building them. But the meta-learning is if there's too many, you can strangle your agent and it just doesn't work. I don't know if you've seen anything similar, but I had to rebuild it. So now the pain in the ass is, okay, so we got rid of it, we went back, we retested it, but now I've probably gotta run 500 decks through it to have the same level of confidence I had before. And there's a meta issue of, you know, Gary Tan wrote this thing the other day about how you have to just trust your agents at some point because it's just too much to qualify. I gotta trust it, but I don't know because all the work that went into grading 4,500 decks, the fix seems to work, but does it really work? Do we need new guardrails? Now, how many is too many? When do I, when do I do it back? What, What, how do I share mistakes with the agent so that it doesn't over-guardrail? I don't know, man. I don't know if you've had any similar learnings or insights from it.

Well, yeah, I mean, I think for, for when it comes to like outbound across our like AI SDR agents, I try and force rank the amount of segmenting/guardrails we do. 'Cause otherwise, you know, the agents and the platform will kind of tell you, oh, you've kind of segmented everybody out of, out of your, out of your segment if you have. Too many filters, too many guardrails, too many exceptions, you know, or too many rules just in place to reach out to folks. And then if you say, you know, only do it this way or only bring up these examples, you'll just see that the agent will just not— it'll probably not do it kind of the same way, right? Because it's just, it's just too much for it to— like, you've kind of filtered yourself out of, out of any workable kind of model there.

But when you're running— and so, and so, you know, the meta lesson for me is, you know, guardrails are technical debt. Too many guardrails are debt, right? They can break the product. When you're building AI SDRs and others, you've done all that. I haven't really done it. We've built 10K and QB together and other things too, but you've done those. Do they limit the number of guardrails? How do you make sure, like, we want these emails to be our voice, to be Saastr's voice, to have Saastr's content, but how did they limit it? Do you know? So there aren't too many inputs, too much into the prompt, too many guardrails.

Yeah, I don't think they really, you know, I will say, I don't think that's something that they do well right now, right? It's like, You can kind of let it run rampant if you want, or if you put in too many yourself, you'll see that you'll filter yourself out and then your agent just doesn't send, like, it sends like 5 emails instead of like 100. Same issue.

If you put too many in, it just won't send the outbound emails, right?

It just won't send it. Yeah. It's not gonna say to you, hey, you've got too many guardrails, too many filters, your segment is too small, whatever. I think as the tools get better, it'll get there. But for now it's pretty much like you gotta trial and error. You gotta figure it out on your own right now.

Yeah, well, we'll go on to the next topic, but it is interesting that guardrails are not the magical source of improved output that I thought we are because too many guardrails leads to too tight throttling, whether it's in this pitch decorator or in the SDRs, right? It's another thing, another reason you have to constantly be managing these agents on an almost daily basis because over-guardrailing is as bad as under-guardrailing.

I think the interesting portion, which I know we'll talk about more today, is Yeah. How many guardrails I've already put in though to this AI VP of market or, or a VP of finance that we're building, right?

Yeah.

So for that one, I don't know if I've put too many. I don't think I've broken it so far, but that one probably has the most so far because obviously it's got financial data. And so that's the one I've had to be, I think the most care, like outbound, okay, you send too many emails, it's not the end of the world. Your finance data, something goes wrong with their agent, maybe it is, maybe not the end of the, but like, you know, something more at a worse-for-scale can happen 'cause of the information that it has.

Yeah, it is. We'll learn, we'll come to that next. But for something like finance or regu— for us, for finance, things that are heavily regulated, others, you would tend to want to over-guardrail them, right? You would want, I don't care what the token costs, I don't care if it costs more and you need longer prompts, I just need it to be 100% accurate. But if by doing that, you're actually gonna corrupt the outputs because it's over-guardrailed, That is something maybe folks watching or listening already got because they're ahead of us, but I didn't fully get it. That, that you, that in our, the things that we really want to lock down, make 100% accurate, not 90% or 80%. Like a lot of these AI SDR emails we've sent, honestly, they're probably about 85% accurate, right?

Yeah.

Better than a human. Better than most humans. Right. But they're not, I don't actually think any of the emails are great. I just think they're good.

They're just good. Yeah.

Agreed. Yeah. But when it needs to be 100%, I don't know. We've got to get some guests on the show to talk through this because, uh, it's a new learning that it's not as simple as adding 14 or 15 guardrails because it, it, it, it broke all of our multiple business processes for us.

Right? Right.

Right. Okay. So I wanna talk about 10K or AI VP of Marketing because I wanna talk about a couple things that are new there. The first one, which I think we, we underdiscuss, I wanna learn more about, is when we were at SaaStr AI Annual Live, you had a couple of workshops, but the first one, the first one was you rebuilt a 10K live., with a couple hundred folks, which we can talk about. Yeah. But then also, interestingly, you built one on Lovable after ours was built on Replit and can see them side by side. So you used—

Correct.

You attempted to use the same prompt, you attempted to use the same, and they came out different. The recommendations are different, right? They look the same, they're pulling from the same APIs and the same sources of data, but they're making different recommendations.

Yes.

I call this one 10K Prime. It's like, it's like a different, it's like a different universe. It's like the Fallout universe version of 10K, 'cause it's coming to different conclusions from the Same. It's a branch. It's a historical branch.

You may choose a different faction than I would. Yeah.

So what'd you learn from rebuilding, uh, something in Replit under using similar LLMs, but not necessarily the same and rebuilding in Lovable?

So it's the same spec across the board from what we've had in Replit to now what I put it under into Lovable. So it's interesting because, so here is, it's, um, one of the core features of the AI VP of marketing we have. That's in the spec is to generate— for Replit, it's 3. It's interesting, Lovable decided to do 4, 4 ideas a day. So that's really interesting, 'cause I think my spec is just like, generate, you know, the top ideas of the day to help with marketing. And so it's interesting here because these ideas you'll see here are, are a little bit similar, but also different from how we've been getting ideas from 10K in Replit. So this, the one that stood out to me was number 2 because it automatically said to start a LinkedIn— again, this one's with fake data, but a LinkedIn campaign targeting GTM leaders in Austin. I think this fake event is in Austin with a last chance register now message. It's interesting because the Replit 10K AI VBO marketing has never— I don't think it's ever told me to do paid ads. And so it's interesting that this one is slightly more aggressive on one, spending marketing dollars, And it's also interesting, it didn't really tell me how much I should put into this LinkedIn campaign, but it said to do this and that we need to hit our goals, we should start running campaigns basically immediately and also launch a flash sale almost immediately. Beautiful. To get folks to hit our goals. And so it's just interesting to see how two different vibe coding platforms with the same spec can have very different ideas and out like, and kind of goals of what to do with your marketing data. And your marketing outcomes once you build them out.

So it'll be interesting. Tell you what else is interesting looking at this. This is the first time I've seen it. So when we get our 3, not 4 ideas a day from 10K Built in Replit, we get 3 ideas. Yeah. One that I got today, speaker AMA series on social, or one I pulled up, last chance team discount. Okay, I guess that one's a little bit old. And retargeting past site visitors. We can argue whether those are any good. What's interesting here is look how it's done the— On its own, it's got motivation. Channel and audience and success criterion, right? We need to remove some of these markup artifacts, but it's pretty interesting that, that Replit AI VP of Marketing says, here's the plan and then here's how to go do it. Okay. That's one way a human might do it. Guys, here we have 3 ideas today. Let's go do it. Yeah. This is much more like a, almost like a B2C marketer. Okay. Here, motivation, channel, success criteria. How's we gonna grade it? Right. This is actually— Right. In some ways it's better, but it's, it is even more different personality, right? Yeah. Isn't telling us go, what to go do. It's broken it up into the 3 areas, motivation, challenge, and success criteria. Into why. Yeah. So it's pretty different. These are 2 different flavors of marketers. So pretty cool, Amelia. Thanks for showing us this. The, okay. And, and, and, and we'll, we'll talk about another marketer that we built accidentally. We touched on before in a minute, but for, but you, you hit our AI VP of finance. So one interesting thing, interesting. And, uh, and we touched on this when I, when I did the deep dive with Amjad, CEO of Replit, is whether you want a set of master context for all your agents or whether you want it siloed. And our goal after SaaStr AI/ML this year was to build an AI VP of finance. We thought we would start maybe a month out. It'd be a complex project. Amelia, you couldn't help yourself. You started on vacation.

I, yeah, I just couldn't take it anymore.

You couldn't take it anymore. You couldn't take it anymore not getting— I couldn't. Well, well, one of our, our finance lead went on vacation during SaaStr AI/ML, so we got no updates. So you got frustrated to just see, hey, what if we hook everything up to an API? What could we get out of it in a agentic level, right? And let's talk about that. But the interesting thing is instead of building a discrete agent, you built it inside of 10K or AIVP Marketing.

I did, which I wasn't sure if that was the right move, but now I know, well, for now it has been the right call.

Okay.

And I'll tell you why, which is interesting, right? Yeah, first I thought, okay, I'm just gonna build my, or like a separate AI agent for to, to handle finance so that way it has its own, yeah, personality, kind of different guardrails so that it's, you know, very more, more locked down because it has more data than just like sponsors and contact information like QB and 10K have. Like, okay, if 10K changes something in our Salesforce, again, not the end of the world if it does something wrong. If it sends an email to— if, you know, QB, our AI VP of customer success, sends the email to the wrong sponsor, which it never has, but if for some reason it does because it gets confused or regresses, again, not the end of the world. Finance, I was like, okay, I'm going to hook it up to all of our finance tools that it can hook up to.

So that was a little bit of a learning there.

So For context, we're on bill.com. We use Brex, we use QuickBooks, which I don't really, I've never really logged into until now. Then we also have, you know, we have some stuff at banks and then, so I try to hook up Plaid too to get all my payments in.

Let's just step back a minute. I'm interrupting, but let's step back for a minute. 'Cause folks will be like, that's not a VP of finance. Maybe not. Our initial goals, and it's already changed since you started, right? Our initial goals were to automate collections. Right? Yes. And, and automate sort of real-time visibility into our finances. Those are the only two goals, I think, to start, right?

That was, yeah. So when I was starting to build it out, I just wrote a simple spec. I was like, okay, let's just, here are the tools that we use. Here are like the four, you know, primary tools between Bill, Brex, QuickBooks, and Stripe. We use Stripe. So like, here are the four main things we use. Plus, you know, I was like, we've got some small things running through other things and we've got banks and stuff and, we can hook that up later or whatever. Because I wanted to make sure, one, we can get an actual dashboard slash production for cash flow, which is something our, I don't know, our prior VP of Finance who was here like a week used to do. Which was like, at least at the very bare minimum, he would say, you know, like, where's your cash flow? What's your accounts receivable? What's your accounts payable? Where are we behind maybe on collections? Which I couldn't, I, again, I didn't get a report for a couple weeks on, so I got frustrated. And then I also got frustrated that we just didn't have real-time data on, you know, post-annual, okay, what bills are still due, what's gonna be upcoming, again, to help manage cash flow, which is something your CFO/VP of Finance does report on. So I figured if we hooked up everything to either 10K or a new agent and be able to do everything. Now eventually I decided I would put it into 10K as like a sub-agent, if you will. Because he does have so much data that actually correlates to finance. So because he has Salesforce and he can see what deals are closed, because he has Stripe and he can see how many tickets we've sold on a day-to-day basis. Once he had those, I was like, okay, I'm going to have to redo those if I move it to another one. Then, you know, maybe the insights that a new agent would give me would be different from 10K's personality because 10K's goal as our AI VP of marketing was always to get us to the first 10,000 attendees at SAS Re:Invent, or the first, you know, 10 million sponsorships. And I thought, okay, but without that context, would the AI VP of finance actually be able to project out numbers the way we want it to? And I decided probably not. Now, it would be interesting. I could take this, some of this data and build like a new agent and see what it does just for fun. But for now, I put in 10K and that seems to have worked out for the best because yeah, he does have all that context of, okay, he knew we had just finished SAS Re:Invent. He knew right away, like when we already started closing 2027 renewals, he was like, hey, you need to retag those as 2027.

Yeah.

And also, hey, now that I have Built.com and Brax for like some of your accounts receivables and QuickBooks, you know, for accounts receivables and accounts payables, he already asked me if I wanted him to start generating invoices, which I said, not yet. Like, let's get to that in a couple days.

I think you'd already wanted to start QB— I mean, he already wanted to start generating invoices autonomously, right?

Yeah, he was like, okay, hey, I can make those, you know, invoices for you and I can sync it to QuickBooks so it's on your books and your financials. Also, I can send automatic reminders so that you can solve this collection things that was built into the specs. And so it actually ended up working out for the best 'cause he had so much good context that when I hooked up these systems that were more financial related, that he could actually look at it and analyze it in a way that we wanted to. Even though, you know, for a normal person, you really can't have your, you really, I don't know, can't have your VP of marketing be the same person as your VP of finance.

Yeah, it's interesting. There's, you know, if you talk to folks, if you talk to folks like at FIN or Intercom or Gorgeous or Sierra and others, they're all saying that there's convergence, that support, sales, and marketing are all converging because once you can AI-ify them, you don't want to have 3 different agents, right? You want one.

Right.

And I didn't think that would happen to us. I assumed the AI finance would be a brand new agent But as we got closer and closer to SaaStr AI Annual 2026, I added projections. I added forecasting. I added a lot of dynamics. So if you had a bunch of stuff, but I wanted to know forecasting down to the minute, and it became a real-time forecasting thing. Right. And I said, now that we have real-time forecasting, why don't you start forecasting 2027? And I added that and you added the rest. And so a human AI VP of marketing would not want to get into that. They just, I mean, there's a handful that enjoy it, but even if they did, they'd be hands-off and let the finance team. The agent didn't mind at all. The agent was thrilled. We've never seen projections as good as we saw before in, in 10-K, our AIVP, but it's not a VP of Marketing anymore. Is it a VP of Revenue? Maybe.

I think it's VP of Revenue.

Yeah. Yeah. Or it may ultimately become the VP. Just one. It could be just one. Yeah.

Just a headless title.

Um, John made the point. I wish I knew what, what I, I, I, I wish I knew exactly what he called it. I'm stumbling on it, but. What you want is sort of these massive repos, these massive orgs where your agents can all pull from the same context. And so in Saastr.ai, even though we had the regression where the pitch deck organizer broke, it's very easy for me to build new apps in Saastr.ai because it has all the context of the other apps that we've built, right? Yep. And so this made it easier to build the API of financing in our AP/VP marketing because it already had projections, it already had the numbers, it already had the tickets, it already had year-over-year data, right? It already had access to Salesforce, it already had access to our pipeline. It didn't know what we collected and didn't. Right. And oh, actually, the other thing I did just as a learning, this is, I think, helped you just before SaaStr Annual. I just uploaded in Excel or a Google Sheet our historical financials so it could project just a static set.

Yeah.

And then all of a sudden figured out, hey, I don't have go-forward financials, but now I have a— I have your, your static financials. I can do a lot with it. Right. So it came up with all of these ideas on its own. And, and I don't know that we'll ever have a discrete finance agent unless we buy one.

I don't— again, I don't know. I think for the same reasons with why I didn't put into a separate one is why I wouldn't. I don't know if I would even buy one because like, I would— it's the same reasons, right? It wouldn't have the context that like 10K had for us to get him off the ground and running. I was— there's a few interesting— I mean, for better or worse, here's the thing. When you deploy any agent, right, it was the same thing with 10K. It was the same thing with QB. I'm sure it's been the same thing with our AICRs and like with AgentForce when, you know, we saw we had a bunch of ghosted leads and that's why we deployed it. The same thing happened with finance. Like anytime you hook up an agent, you're gonna find things for now until, you know, you're, you're maybe more agentic than humans like we are. You're gonna find things that the humans tried to hide or that they never did or that—

That's interesting. What are, what are the things that are driving you crazy already?

10K was pretty— well, I/10K pointed out that we could have had auto— we could have automated this a lot already, even without hooking up the agent. Like, just even when I just put in like bill.com as the first one, because I asked him, I was like, okay, it already had Stripe because it had a lot of our ticket data and it already had Salesforce. I was like, okay, these are the other tools we use for finance. Which ones, like, rank them easiest to hardest to hook into you? So it said bill.com, pretty easy. So, and to its credit, it was, you just make an API key in the app. It did say you used to have to request one, but I guess now you can just, maybe they know that people were just doing this. You can just make one. So I made, I made one. It probably took 10 minutes. That was nice and easy. That was great. QuickBooks was the hardest, and that's maybe no surprise 'cause it's a legacy platform. It was annoying AF to hook it up. Like even though I don't use QuickBooks, if you're the admin, which I was already, you can, you, you have to create an Intuit developer account. Just to get an API key.

It's a classic way to do it, but sort of pre-8-bit. Classic. Yeah, but it's true of, you know, Twitter and stuff has it too. It's a classic way to do it, right? It makes sense.

Yeah. It's just hard to do than you thought. It's just because it's finance, it's harder to do than you thought. You see, you have to answer all these security questions, upload all these different chip-ins to get your API key. So that's why.

I mean, as, as when you look at the ones that are easier to use, and we have a whole Agentic API grader that's graded 150 APIs. It is interesting that because we're all developers in the Agentic world, the, the, the line doesn't exist anymore. The line between, Hey, there's, there's a handful of developers for it. Unless you're building something that is developer only, like a Stripe or Twilio, there's a handful of folks that need a developer account, but 99% of, of the world doesn't. So they're very split and segregated. Now they're the same people.

Yeah, that's the— It has to be the same people, right? Yeah. So yeah, that one was hard. I still have not, as of today, and it's already been like 5 more days, I have not gotten our Plaid to work. I don't think I've been approved yet. So you, again, because it's banking and financials, you have to like answer a bunch of questions. Wait to get approved before you get an API key.

So I think— Some of that could be, we'll have to research, some of that could be KYC or know your customer or other regulatory stuff, right? It may not be, right? It may not be, but—

I don't know, but I was like, you know, I was like, when I was going through it, I was like, but I'm not gonna charge it. Like, I'm not using this to charge people things. Like, it's literally gonna be an API-only user to hook up to our, all of our banks instead of just me doing it. Try, you know, our, some of our banks don't have APIs, so I've If I just hook it up to Plaid, I could do the rest. So it is weird.

So you're using Plaid. We have, we have money at Brex, Silicon Valley Bank, and Wells Fargo, right?

Yeah.

Brex already works natively at the API level, right? We're really using Plaid.

That's pretty much native. That also took 5 minutes. Brex was super easy.

Okay, cool.

Plaid so far, so far the hardest. And I, I, it, it's interesting. It does have an agent workflow, but even its agent workflow kind of assumes you're gonna charge people via your agent. I think it needs something where you have a non, you know, write, read-only mode agent. I don't think it's quite there yet, but we'll learn. We'll learn.

That's better. Okay, so, so you, so you've hooked up QuickBooks. It was painful. Brex was easy. Bill was easy. Stripe was already there, right?

Stripe was already there.

And our other bank accounts are TBD, right? Yeah. And then you learned a lot of things were just really easy, like reminders, automated invoice creation, right? These were very simple.

Yeah. And so, you know, part of the original problem I was trying to solve before we got into this whole like dashboards and projections thing was I wanted to solve collections because collections has been pretty manual for us, right? We have a part-time person on the finance team who sends the invoice that they generate through bill.com and it syncs to QuickBooks. Then they send it to this, whatever, the signer and the finance contact. And then if they, and then it, if they don't pay on time, and this is also the thing that always bothered me, if they don't pay on time, then they get a follow-up. It was never like, hey. And I was like, this can't be that hard. I get reminders all the time before our invoices are due. Like, hey, your invoice is due in 5 days. Hey, your invoice is due tomorrow. And yet our manual system was always after the fact, which was annoying. So it's interesting, when I hooked up 10K to build.com, it said, "Hey, did you know you could just turn on auto reminders for invoices?" And I said, "What do you mean I could just turn on auto reminders this entire time?" We've been on build.com, which is the entire time I've been here. So you mean for 8 years I could have just sent the invoices, but like the invoices were already in build.com,, and I could have just clicked a fricking toggle to send. Not only could you send auto reminders before the bill was due, you could, you could, you could escalate them if they became overdue.

And 10K was like— Because I'm hoping a human would do this on a weekly basis. I'm hoping a human would do it, right? Which they often wouldn't do, right?

They often wouldn't do this.

It's interesting because obviously if you'd asked Claude, hey, does Bill.com have an auto reminder feature? It would've known, right? It, it would've known this, right? But what's interesting is an agent it sounds amazing. It's not really that amazing. It's just wonderful. An agent can take that context and figure it out on its own.

Right? It can— Right away. The fact that it already knows— It's the first thing he flags.

Yeah. It already knows everything in Claude's sonnet. At least I don't think it's using Opus, but it's using sonnet and it can say, yeah, of course you should be doing this. Right? But you wouldn't think to, to ask Claude this question. You never would've occurred to you to ask Claude this question.

No. And it didn't occur to me that I should just, you know, log, force myself to log into bill.com for like an hour or two and figure out what it could do. Already, right? Like, it just, I was like, whatever, that's the finance thing and it's annoying and we'll build an agent to solve it. It's just, it's funny, it's literally the first thing when I hooked up the finance personality and context and systems to it, it was like, one, you've got, it automatically saw what invoices we had overdue, right? Yeah. And it said, how are you, how are you invoicing these guys? You could just remind them and bill. I was like, what do you mean I could just remind them and bill? So it's, it, Saw it instantly, right?

Saw it instantly, right? That's the powerful thing. Maybe when you hook an agent, this, maybe this is where we go into the next topic, but when you hook an agent up to third-party APIs that have your data, right? It will, if for no other reason than just using the fact that it has the entire internet in its memory in Claude, right? It will figure out without drama, without laziness, without pushback, things you could do with that data that you didn't know you could do. Instantly. It, it seems like the smartest person you've ever worked with, but it's really just connecting the fact that this is in Claude to your data and saying, it's obvious we should be doing this, Amelia. Right? And so as you hook up all of your agents to all your APIs and all your data sources, they will— that, you know, our 10K comes up with 3 ideas a day. We've talked about it. It's, it's endless. It's exhausting. But, but, but, but this is a version of in finance. It will keep coming up with ideas on how to improve our finance workflows on its own. It's sort of magical, right? Yeah, it will. It's sort of, it's sort of magical that, and that might be the line between an agent, right? And just, and just a query is when it can magically come up with these ideas on its own and ideally begin to execute them on its own, right? That's the next, the next big gap. But okay, cool. We'll keep this journey on, but it, you know, it sounds like we might end up with the best finance leader we've ever had in the history of Saastr in a couple weeks.

As long as he doesn't, um, I don't know if he could. As long as he doesn't do something truly awful, yeah.

Well, it's funny when we asked, when I asked 10, when we wrote it up on, on, on the blog, when I asked 10K, are you an AI VP of marketing? He said, no, I'm a great marketing manager. I would never say I'm an AI VP of marketing. I think it was being intentionally humble. Right. And then it said something that sounded like a trite AI-ism, but I think it's very true. It's like to, to get to the level of autonomy you want, you don't need humans in the loop. Okay. That, that's not autonomy, right? What you need is humans on the loop constantly.

Right.

Having the exceptions kicked out to them. You're part of what the, our AI VP of revenue finance is doing. You're part of those decisions. You're seeing the outputs, you're seeing the inputs, but you're not, but, but you don't necessarily need to be directly part of that loop, right? But you are on the loop. And I think that, that we'll have to figure out on the finance side so that it doesn't make terrible mistakes. But as long as it doesn't wire cash out to the wrong people, we can probably recover from most mistakes. When the, the cash is gone, that would be bad, right? If it sends the wrong invoice, it'll be a headache. But we will, we will survive the wrong invoice or the wrong reminder, right? Or sending a stale invoice or something like that. We'll learn. Okay, cool. The next one I wanted to talk about, it sounds minor, but it was super interesting. I know it was slightly interesting when our FDE went on paternity leave. So we, everyone, the whole internet's talking about FDEs. We need more FDEs. We're gonna need, I think Aaron Levy said on a, on a tweet today or the other day, we're gonna need 100 times more FDEs than we thought because there's gonna be 100 times more workflows out there. And they all need humans to deploy them. They don't self-deploy. I agree with that. We probably need 100 times more FTEs than we have. But, but this was a reminder that even when a company says they have FTEs, a lot of times a lot of them aren't up to that, aren't up to the bar. They're just CS people repurposed or whatever the title is. They're not good enough to help you manage the agent. Right. And we had an issue where we couldn't touch, we couldn't, we couldn't fix issues with our agent for like a month. Right. Yeah. So what was, what was the rough story here?

So the rough story is we have an FDE for one of our platforms we use, I won't name them, that I, I can, for context, it's not that I'm dumb and I can't use this, this tool. Like, I can use it, I can do things myself, I can program the agent, I've done it many times. Just sometimes our FDE that we had at this particular tool was very proactive. So like they knew we were busy, you know, going into SaaS Radio, and so proactively they would update things for me and they would just Slack me like, hey, saw you updated this, I just updated it for you in like the agent context, 'cause it's not something the agent would do automatically. And our FDE was like that nice bridge between, you know, what the agent could do on its own and then me and like it doing some of its whatever workflows on its own. It was a nice bridge. Like we had this really great FDE, but then he was, they, I should say they were going out on paternity leave. I guess that's, that's, that's gender specific.

Family leave.

They're going out on family leave. Family leave.

Family leave.

Yeah, family leave. And which I get, like, you know, happens. And I was like, okay, cool, it's fine. Like, again, I can use this agent mostly myself. Like, I don't think it's gonna be a big deal for it to, you know, have some other person just in the background. I already was like, you know, there's no, no other person they give me is gonna be as good as, as my FDE. So whatever they give me, I already resigned myself to just working with the agent a little bit more myself while they were gone. 'til they got back, right? But then I needed something where I literally couldn't do it myself. And so I asked the kind of backfill replacement. And there's two interesting things here. One, they backfilled my FDE with a CSM who was definitely not as— I was like, okay, mistake number one was jumping me on to like a junior, 'cause I mean junior truly. Like I looked at them on LinkedIn, they've only been in this role for less than a year. I was like, That's mistake number one for somebody who talks about their agent.

None of these products worked a year ago. But putting that aside.

Yeah, but aside, I was like, dumping me on— it's funny because actually another one of our agents tried to do that the other day and I was like, you probably don't want to do that for these reasons. They tried to dump me. So they dumped me on a junior CSM, which again, CSMs, I don't know how you guys are doing it for yourselves, but CSMs are not as good as FTEs if they don't know how that product actually works. I ran into an issue where I couldn't do it myself. I needed help from who would've been my FDE. So I asked the CSM. I kind of knew I was probably not gonna get far and I would probably have to escalate on my own to get an answer, 'cause I was like, they're not gonna answer it. But I, you know, I followed the motions. So I asked the CSM and a few days later, which is also interesting, my FDE usually responds instantly. It's like this CSM, um, takes a couple days to respond to anything.

Mm-hmm.

And they were like, oh.

Which is brutal in the age of AI, a couple days.

Brutal.

Brutal.

Yeah, days is a lot. Days. And the answer was, oh, I can't help, like, I can't help you with that. And we may have to like rethink how much we support you till your FD gets back because I literally can't help you with that.

At least it was honest. It took 3 days, but. So we couldn't get any help until the real FD got back for months. You were—

Well, I escalated.

Oh, you escalated?

Yeah. I escalated because I was like, you know what? This was before Sashree Annual. Yeah. It affected things we were going to do in person. And I was like, that's not an actual viable answer for me. Like, this has to— this has to work on May 12th through 14th of this year. And so I just escalated to, I think, the CEO. Quite honestly, I was like, I'm just going to freaking tell the CEO because this is— I was like, one, this is not the answer their CSM should be giving. So not, I'm not, not that I'm trying to tattletale on their CSMs, but it is something if I were the CEO, I would want to know that like when my FTEs go out, my CSMs are not a good backfill. And two, I needed the support. So I went to the CEO and I got this, I got the support I needed and like some apologies and whatever. But again, a situation 100% avoidable for all of these.

Well, essentially, if you hadn't, successfully escalated it, our, you know, our NPS would be zero. If we got no help for 3 months.

Yeah.

Like we would churn, wouldn't we? This is, this app, this particular app is a good one, but it's replaceable.

It's replac— yeah, I know. That's the thing. That's the thing. I was like, you know, as much as I love this tool, and I think it's a risk across all of our agents where we have really good FDEs, like if our, if one of our—

We would churn basically everything.

If our FDE leaves.

Yeah, if we didn't, if we, it was worse, right? It would churn.

Yeah. So.

I don't know the answer. I think, I think, I just think everyone I talk to, CEOs and others, they're just underestimating this because one, they can't get enough FTEs. Two, they think they're too expensive. But three, just like everyone a generation ago thought their CSMs were better than they were, right? They all, most of 'em were terrible. Now they think their FTEs are better than they are. Just because you hire someone in FDE doesn't mean you can support, they can support your customers.

Doesn't mean they're magic. Yeah, yeah, yeah. Doesn't mean they're magic. The last tidbit on this one was I had somebody ask me at Annual. Yeah. Um, if they should use one of the agents we use and they said they got two quotes from one of our agents that we use 'cause they were there at SaaStr as a sponsor. And they said, you know, I got this quote for whatever the base rate. And then I got a quote that's I think like $20K more. To have an FEE do the implementation. Should I do it? And I asked Jason M. Lemkin, what I should— I was like, I'm gonna ask Jason M. Lemkin what we should tell this folks. And, and we said, you should do the one with the FEE even though it's more expensive because who else would implement this, right? So, it's, it's interesting. FEEs are expensive and they are expensive to the user, but at the end of the day, you gotta make it work. It's gotta work.

It's also interesting, you know, if you look at, data that Iconiq recently published. We have it on SaaS or Others. A vast amount of more Agentic deals are annual deals or less rather than multi-year.

Yeah. Oh yeah, yeah, for sure.

The main reason folks are doing it, if you talk to CIOs and others, is they're just worried that in 10 months there'll be a better product on the market, which is very fair, right? Folks are saying, listen, I'll pay you $50K, $100K, even more a year. Fine, if the ROI's there. But you're, you know, how do I know this is the next— I mean, this stuff's only 8 months old. How do I know you're the best product?

Right.

So no one really wants to sign A lot of— when I talk to a lot of folks, especially fast-growing folks, they're stressed. A lot of the CROs are stressed because they know all the deals are up for review in 8 to 10 months, no matter how well they're doing this. They're actually not that happy, a lot of CROs in these companies. But I didn't fully appreciate— so I understand the rate of progress, right? And, um, but what I didn't underestimate is I might churn just for FDs.

Yeah.

I would not want to sign a 3-year contract and find out Lindsay or Amelia leaves the next week. And I'm stuck with David or Jason with a 3-year contract, I would— this would be worse than Marchetto.

100%.

Yeah. It's almost a reason in theory you'd almost want to do month-to-month deals because you want out. If you can't— if you get a good FTE and then you lose them and you can't get a good replacement, you literally— I would leave. I would leave. And related to it, you know, another thing wasn't even on our agenda, but LLM portability, you know, we had We had one of the co-founders of Databricks at annual saying now they can do Databricks migration in a matter of weeks because they perfected with LLMs, right? As we're trying to figure out what to do when we leave our worst app Marketo, you know, Salesforce migrated us in a matter of weeks too. It was very simple, right? Versus, you know, a couple, you know, a year and a half ago we got a quote from HubSpot. It would take a year. So as you can use LLMs to do migrations much faster or even, or even just pull from the same set of data. You, you should leave an app without a great FDE. You should just leave. You should just leave and go to another vendor, especially if they're just pulling from these other sources of data. They may be able to, they should ask if they can redeploy you in a week for, at no cost to you. If you don't like your FDE, go to a competitor, see if they can redeploy. I'm not paying any implementation costs or switching costs, dude. Use, use, use an LM, use migration, but just do it in a week for me. And if it's better, I'll leave.

Yeah.

It's better all leave, right? I think it's, I think this is a meta issue to deal with. Okay, next one I wanted to hit, anything else you wanna talk about on the FDE risk? But I think it's even bigger than people talk about. It's not just finding them, it's when you lose your FDE, you might wanna lose that vendor.

No, I agree. And I think it's, there's, it's an interesting thing on the side of it's vendor agnostic, right? Like right now it's true of all of our vendors. Like, and if any of our FDEs left at one of our top agents, not the ones we've built internally, I probably would churn and be like, well, where are they going? Let's just see. Why? If they're going from Replit to Lovable or Replit to Vercel, we may move all of our shit to Vercel, you know?

I mean, I mean, you know, we love Replit to death, but if Cody leaves, I don't know, maybe, maybe we might have to switch a few apps over, right?

Yeah.

You know, it's related to this point, just one other small thing that I think is minor, but it's interesting that it isn't, is when we were backstage at SaaStr AI Annual this year, we had, Adam Alfano, who's president of SMB, or I think that's his title. It's close to it, right? Good friend of ours. You were there with the head of, with the president of SMB of PayPal. You guys demoed AgentForce. It was great. But he grabbed me and he was kind of laughing. He's like, you know what you guys say about AgentForce? It's pretty funny. I'm like, well, what's funny, man? Uh, he was laughing. He was like, you know, you guys don't even have an FD. You don't even have an FD. You only have an SE.

Yeah, I know. Lindsay's not technically an FD.

Here's why I just think this is so interesting. From Salesforce's perspective, a forward deployed engineer is something like Palantir. They go on site for a 7-figure customer. They might spend weeks there or even longer, and they make sure a big 7, 8, 9-figure deal just effin' works. Okay. And that is the classic model. And as we've all gone on this agentic journey, we've adopted this term FDE, which, like a lot of other terms like agent or cloud, has— I think it started to come to have a different meaning than it originally did in the Palantir days. I think all of our folks that we work with that are great are FDEs, but their titles could be very different. It could be an SE, right? It could be a field engineer, right? It could be. And so let's not, the one learning is don't get hung up on the title. Get hung up on if that person is technical enough on the product to help you. To help deploy. Yeah. Don't get hung up on the title. Also realize when you're talking to vendors, they might not know what they're talking about at the title level. So we loved our FDE at Salesforce. Like we said, put us, you can put us on the Salesforce website that we love our FDE, but they never would because we didn't have one.

We, I, it's funny. Yeah, I know. I think I've said it on this podcast before. I was like, I think I've said that Lindsay's not technically an FDE, 'cause I've done trainings with her for the Salesforce SE team. So, I knew she was not an FDE on paper. And I actually have had FDEs truly from Salesforce in our account when we were trying to do some in-person things. Yeah. So, I knew there was a distinction, but yeah, to me, the definition is just who can help you deploy this and maintain it. Like, who deploys and maintains your app for you from the vendor? That's your FDE. My FTE, I don't even, is Cody an FTE?

No, the forward was Palantir trying to put a nice veneer on making poor engineers go on site, which they didn't wanna do. The forward was trying to make a job that was uncool back in the day sound cool, right? Too bad you don't get to work in Palantir's headquarters in Palo Alto. You've gotta go to Reston, Virginia for 2 months. You're an FTE.

And the key card thing is funny too because You know, it's fine. Like we've been joking about getting me a key card to Salesforce Tower 'cause of how much I meet with Lindsay, but which I think is finally happening, which is a little funny. But it's, I also told her and I told Cody too, who works at Replit, I was like, well, but you guys can also just have a key to our SaaS for office. Like it doesn't have to go, you know, always this way of like I have to go to my FDE. Like, why don't you just, you could have a key to our office 'cause it's literally just for us and the agents. So you could come work.

Yeah. All right, a couple other things I wanted to hit and more high-level things, and then we can come back to a few things. Let's ask ourself about our learnings. It was a couple weeks back, but I wanted to get your thoughts and my thoughts and dig into it. So Matthew Prince, the CEO of Cloudflare, you know, one of the biggest AI beneficiaries, wildly successful, super smart guy, become a little bit grouchy, but that's okay. I can be a little grouchy. You can be a little grouchy sometimes too. And they did one of the first mass layoffs of a company doing really, really well. So they laid off 20% of their workforce even though they're crushing all their numbers, right?

Yep.

And, and he said, listen, there are just people we, I just need to get the people into Cloudflare that we need to win in the AI era. And one of the things he pointed out that they absolutely did not need was sales ops. Uh, he talks about in general, I don't need folks that measure something because 10K will just measure it. 10K will just do it. And he said, certainly what I don't need is sales ops, a huge team making up how we might do next quarter, next week, when a well-trained agent can do it in real time. So what are your thoughts? What role, like, do we need sales ops? What other roles do you think that we've obsoleted, or do you talk to? Was Matt, was he right? Was Prince right on this?

I think he was right. And I think other people agree and they just won't say it to the same degree, right? Like, so when I was chatting with Denise, the CMO of Snowflake at SaaStr Annual, she said, you know, her big tagline is the dashboard is dead. Well, if you listen to why she says that, it's because she doesn't need sales ops people or marketing ops people to give her these dashboards anymore. She can just do it with her agents. Now Salesforce is— or sorry, Snowflake is public, so they have to like repurpose these people for now. I think Salesforce—

She's directly querying Salesforce. I mean, Snowflake herself is CMO. She's got her own dashboard. She's creating for herself.

She has her own data lake. It's like she's just doing it herself now.

Yeah, I get it.

She literally said, she's like, I just log into my agent that I built and I just see all my numbers and I don't have to fight with people on what the numbers should have been.

Yeah, she's been telling people arguing with their biases about what the numbers are anymore.

Oh, she just sits down every morning.

That's a great example. Every morning. If Denise does it, if it's worth Denise's time to do it, it says a lot, right?

Yeah, she's like, you know, I think she kind of, again, they're public, so I'm fairly confident she said she's repurposed some of those, you know, folks into, into other areas, which she does talk about in the session. If you guys wanna watch it on our YouTube channel, there's a lot of great content from Saster. But again, if she wasn't Snowflake, would she have just gotten rid of those people? Probably. Yeah, prob— maybe, right? Maybe. Sure.

So, Tom's doing all these ops ones. Yeah, it is Tom.

I mean, not that we've ever had like a true RevOps person. I've just kind of done it as a backfill. But now that we have 10K in QBee, why would we need? I was like, I would never hire an actual, RevOps person to fight me on the data. Never. There's just no need.

Yeah. Yeah. I mean, for us there, so there'll be a whole generation of these people that are just never hired. There may be some folks that are laid off. I, and I think more commonly they'll just be slowly managed out and just replaced with agents. Like it's gonna be more invisible then. But yeah, it was, I didn't know that story about Anissa. It's pretty good. There's another one I've been thinking about and, and maybe we don't have any insights and that's okay, but Uh, you know, the other day Uber said they'd already blown through their whole token budget for the year. They've seen a lot of ROI from AI. This is kind of accelerated. Uh, there's a lot of folks saying this is a big deal in the enterprise, that the ROI isn't there. At first I thought this was just old school companies, pooh-pooh. And then this week I went to two board meetings for Saster Fund portfolio companies who said they blew through the whole token budget already this year. Already gone.

Yeah.

One wanted— like 5 more million this year in tokens. And it's interesting when we compare the fact that QBN 10K costs $257 a month to run. Do you think there's— and there's a lot of differences, but we have become hyper-efficient, right? We have replaced a lot of people with tokens and agents, not through layoffs, not even through attrition, just through just doing it. Can you imagine a point where we really care about the cost or you, or it's difficult to imagine? And there's, and that the ROI for us is so insanely high that we would ever care.

Yeah, it's okay. So I think comparable examples, right? So like when we were chatting with Elena from Lovable, she said they basically have endless token budgets. If you're a lovable, of course, you know, they're a pretty company. They basically have an endless token budget. Denise from Snowflake said they have endless token budget, but she did say it multiple times for now. Like, not forever, but for now, people have somewhat endless token budgets to play around with, you know, things that will make them more productive and agentic at Snowflake. So, I think it's, it's interesting that folks for now seem to have endless budget, but I think eventually real humans, if you have bigger, bigger orgs and CFOs, may wanna rein that in 'cause they're gonna see it as just a straight up number, right? Like, That's, that's the risk of when you don't have an agentic CFO, that the CFO that's a human is going to look at that number and say, I don't know if we are more productive or not, right? If they're not associating it directly in the revenue.

Yeah. Yeah. I think that is interesting that Denny said for now, right?

She said it like twice.

I was like, for now.

I hope our team is listening. You're probably going to get off at the end of the year.

Because our direct cost to run these two VPs is $257, right? Even if, even if Even if QB takes over all of finance, it might be $300 a month. That's not a lot for 3 VPs, right? If you really look at the fully burdened cost, if we were running it like other people, Matt, it might be $3,000 to $4,000 a month, right?

Yeah, because of our time that we've put in.

Oh, well, even just everything that we do, all the all-in cost. If we had 50 people on our team, we would probably start rationing that. We'd probably be like, I don't know if everyone should get $2,000 to $4,000 a month, right? At, at our size and our budget. So I think it's just a reminder that we're going to bifurcate. There'll be folks that are hype. I mean, we're, you know, we're doing, you know, $5 million in revenue per employee. So the tokens are worth it. Folks that are doing $500K or $200K, $200K, which is common for a lot of classic B2B companies. Yeah. $200K in revenue per employee, the token spend may well be stressful at $4K a month per person, right? Or more. So it's just, it's an interesting one. Okay, maybe just, um, two last things I wanna do and then we, we maybe we'll keep this one a little bit shorter than, than, than usual. We hit on this before, but, um, in, in a way it's kind of the opposite of putting everything into 10K, putting our VP of finance into 10K. Put, put, luckily it's called 10K cuz like Amazon, it could mean a lot of things. We can just redefine what 10K is. It could be 10,000 agents. Yeah. 10K could be for 10,000 agents rather than the first $10 million of revenue or the first 10,000 attendees at SaaStr Annual. But we talked about it a little bit, but then you went into more at, at, at SaaStr AI Annual, we ended up with, so now we actually have, in a sense, we have 3 AI VPs of marketing in a sense. We have 10K, which is becoming more. Mm-hmm. We have 10K Prime built on Lovable, which we may want to dust off more to learn more. We may, I think it would be fun in the coming weeks to run them in parallel. Like when we have a little bit of time, literally have both of them literally running in parallel, doing the same thing and seeing and, and having them argue with each other over what we should do in marketing. Yeah. I think that would be fun. And then you turned our website, sasteraiannual.com, into Annie, which is now the best email marketer we have.

Yes, she—

Um, we, you went into more definite, what, what are we gonna do with Annie now? And what have we learned from having this par— other, this third VP of marketing or director of marketing, Annie? Creating the best emails we've ever seen.

Yeah, they are pretty, yeah, they're, I, here's, it's almost confusing of why Annie is so good. Like it has, obviously it has all of our website data for SaaStray Annual. So, and so there's a lot there, but yeah, I don't, I just started to use it to, it became agentic when I built out the Parking Pass app. So that's when it became an agent. Yeah. But then I also started to build emails with it cuz I didn't wanna build it manually and I was low on time. And so I said, why don't you just help me write and build some of these emails so we could do it together and it'll save us all a lot of time and effort. And so it started doing emails really well. And I think just because it had probably 3 weeks or so in the run-up to Saastr Annual of me doing emails with it together, it started, I didn't realize it was learning how I do emails, how Saastr does emails, how we do emails. And since it didn't forget that, then post-event it has started to do more year-round just email marketing campaigns for us. Honestly, and I hate to say this, kind of better than our AI SDRs.

Oh, that's— Convergence, isn't it? That's another convergence, which if Annie can send the emails, right? Annie has all the context on all the attendees, everyone who's ever been before, right? And our AI SDR doesn't link into Bizibo and our source of record or sources of record the same way. For Bizibo, it can write better emails. Emails than the AI SDR that doesn't have as much context, right?

I know, I'm reasoning like 3 different—

That is interesting. Not only is Annie kind of another VP of marketing, but Annie is almost in some cases a better AI SDR than our 4 AI SDRs, right?

For warm outbound, yes. I will say, obviously her gap would be if we went head to head on the cold outbound, but we don't do a lot of truly cold outbound. That's the thing for us, right? For us, we don't do a lot of truly truly, I did some for, you know, our summits and stuff where we use a lot of lookalike audience. And yes, some of that is true for new attendees, but it's a small percentage of the outbound we're doing. Most of the outbound we're doing are people who are already in our database or already listen to the podcast or already follow you on LinkedIn or already have heard of SaaStr. Like, it's, it's, I guess I know that's a very specific us problem, but because that is true of how we use our agents for outbound, it also meant that when Annie post-event now just started to flip herself to year-round, very good ongoing emails to our, our database of folks and people that are engaging with us. Yeah. It is, they are kind of better than our AI SDR platform.

It, this convergence is so interesting, right? So we never, we, a website when it was Agentic enabled became a marketer. SaaS or AI, it was just supposed to be a website, right? But because it built— It was just supposed to be a website. It was just supposed to be, but because it pulled in all of our attendees, all of our speakers, all of our history overall. All of our sponsors, right? It has hundreds of thousands of pieces of data. It became a great marketer because of that, that narrow but very valuable context, right? So, and all of a sudden we asked it to write these emails and they were incredible. Now after this, but that was just for, for one specific moment in time. Now you're running it a, you know, 24/7 or 365 days a week. It's becoming a great AI SDR too.

It is.

The lines between sales and marketing are obviously blurring in the agentic era. For sure. But to get folks to come to SaaStr next year, to get folks to take actions with us, we're not gonna send them through any of our AI SDR tools, are we?

I don't know. That's the thing. I was like, I don't know. 'Cause, nah, ARX is pretty good. Unless I hit some sort of— and you could say, okay, a lot of the AI SDR platforms use like LinkedIn and they use other contacts. But I think if you're a company that already has a lot of rich contacts, which most of you have rich, even if you don't think do, you do. You have what they're doing on their website. You've got, you know, what they're doing on your website. What are they even there? Like if they do 10, if they read 10 different things or do 10 different things on your website, that's pretty rich context. Like your AI SDR is only going to personalize on that anyway. Like it could go look at their LinkedIn, but since it already did stuff with you on the website, it's going to use that because that's the best like nugget for it to personalize on. So it's just interesting to see how these. not only converge, but now, yeah, I don't know. You know, we started with 20 agents and then we kind of ballooned to 30 in the process for annual. Will we now shrink back because they all start to converge into like a handful of agents that just do everything?

Yeah, it is crazy, right?

It is crazy. So I haven't, I haven't cancelled, I haven't canceled anything today for Carmel.

Even if it was just doing some fun, it was having some fun writing these to Oricel and some others, these renewal emails, but they were really, really good too.

They're so good. I learned, started sending some of those.

Yeah, they're so good. It has all the contacts, you know, oh, Oricel, I saw you got this many leads and I saw you, you wrote this on the website. These are things you would hope some, a human or someone could do, but they never do it. You hope, but yeah. Right? It was epic, right?

It started to draft really good emails to folks that were lapsed or folks that we mentioned, like in sessions, right? Because it has the recordings.

It was like, hey. It has all the recordings.

It's got all the recording. So it was like, hey, you know, Anthropic mentioned clay in—

Why weren't you there?

Why weren't you there? And so that's—

I don't know why any AI SDR tool is going to have that level of context. That's in theory, in theory they could. You know, here's an interesting thing. In theory, of course they could, right? This is just LLMs and some, and, and, and some data. But in practice, they're never gonna be as good as for, for these use cases of what we've built. So it'll be very interesting. They're almost, they're, they're, we've almost our verticalized version, maybe vertical and horizontal is the one way to think about it, but when you can truly load up a somewhat autonomous agent with the right amount of context that's ongoing. The thing is it's ongoing con— it gets, these get better every single day, right? With Compaction and more. It's gonna be hard for a third. This is an interesting in the build versus buy decision.

I know, I was gonna say, build versus buy, buy, buy. Don't buy, buy, buy. Don't sell. I know.

This much better. Like, you know, it may show up, going to another point and then we can hit one last point. It may show up on the Wows and DAOs and, and, and, and others. For the vendors, make sure your customers' DAOs and WOWs and MAOs are going up. Because folks that are seeing how we're using them, our agents, in some cases are going to see it going down as we replace just some, some of the use cases with our own agents doing it themselves. It's something I'd want to monitor as a vendor so I could jump in and see if I could do more and save the account, right?

Yeah. Can I add a related point? I know it's not on your agenda, but related.

No, we'll add 2 more and then we'll break.

So related, there's a platform we've been using and now our agent uses it more than I do, right? And so post-annual, this is one of our platforms that's, you know, all of our platforms renew post-event because it's when we sign all these things. So, so post-event, this vendor was like, hey, your renewal's upcoming. Fine. Standard human, terrible email. Actually, an agent would've done it better. Not only would an agent have done the renewal email better because he would've like looked up our actual usage. I think an agent would've flagged, and this is what I told the human beings on this team, that our agent uses the platform more than I do. So I said, hey, I gave your renewal proposal to our agent and it just, it disagreed with you. Yes. And here's why. So I sent it back to the humans on that team, which they were a little prickly about 'cause they're like, what do you mean your agent sent this to us? I was like, well, the agent is the decision maker. I'm not gonna decide whether to renew this tool. Our literally 10K is gonna decide for you. And if it says no, I'm gonna say no. Like, for this, he has the budget, like he can decide if he wants to use this or not. And so I gave it actually a list of things that 10K asked for, for them to update in their API.

Requirements to, to do a renewal. 10K made the— Yeah. Making the, for, for real is making the decision whether to renew or not.

For real.

Right? For real.

And so he came up with a shortlist of things he wanted to see in the API. And that was one of the things I sent over, which I'm talking to them later about today. So I'll keep you guys posted on that of, whether or not they will at least pretend to work some of 10K's demands into our renewal contract. And then second, 10K said to push them on pricing because their pricing is still seat-based. And he said, well, we only need one seat. That's all I think they need. You should push them to charge you for one API power user and dump the rest. I think it's just interesting because it did add some some friction into this from both sides. They were like, you know, what do you mean your agent is just depending this and is gonna be the decision holder? They didn't really get it.

Our agent's gonna be the decision holder.

Yeah. That's what I said. I said, it's not up to me, it's up to our agent and this is what our agent wants. And if you don't meet it, he, the agent will 100% recommend a different vendor. So you're now up for renewal with 10K and not me. I just happen to be the human signer. And then I was like, too, these were his, you know, he had some very valid points on pricing of, you know, one API user to be headless versus per human seat pricing, which this, this vendor so happens to be on.

So many B2B, you're hoping that 80% of your customers are basically idiots when they renew. They don't know the right questions to ask, right?

Yeah.

It's doubly annoying because I know, I haven't seen this inter— I know Tenke's questions were great questions for the renewal. They were great. Better than 99% of humans. This is what a Tenke—

Better than things I would have flagged.

Yeah, these are things I need. And you're really, the truth is I've been doing this for a long time in B2B software. You're really hoping most of your renewals are with idiots. They don't know anything, right?

You're hoping they don't ask too many questions.

Yeah. You don't want any, whether you're in customer success or renewals, you don't want a lot of, on the vendor side, you don't want a lot. Any questions are bad news. The only good question is, can I pay more for your other, you know, if you're a Datadog and you're trying to migrate folks from 8 to 10 products, those are fun conversations. If you're, if you're a single point solution, right? There's no good questions in the renewal.

There's all kinds of questions, right? Because the humans on that side already had some friction about per-seat pricing. 10K just looked up some of the sessions from SaaSRe Annual where like literally Anthropic, Salesforce, Vercel. Yeah. Like everybody in their session was talking about headless, headless everything, right? Headless, literally all apps are gonna be headless. And yet this vendor, I did literally send this to a vendor and said, hey, I really think you're a little bit behind here that your pricing is not headless. Like literally this is the number one thing. People talked about it faster, was becoming headless in pricing and in actuality. And now you're seeing it here with a renewal that our agent is opening.

Yeah, it's funny. Well, we'll, we'll keep, we'll see where this one comes out. I don't even, I can only guess which one this is. It's, I already want to leave this vendor. I'm already feeling like this is not an agent-friendly vendor. Like I already want to leave.

That's the thing.

I would prefer to leave. Like if this call doesn't go great, like I just want to leave.

I said that to them.

Yeah, I just wanna leave. It's just because even if they avenge, like, it shouldn't be a, you shouldn't have this. If your API is perfectly exposed, if you have a headless framework, if it all works, it also, the sales process should be much simpler. It should be not harder, especially if you have transparent pricing. It should be so easy. And so much of this dumb pre-AI drama of hiding things in games and the whole Marketo team threatening us and forcing us to get on 10,000 phone calls. That stuff should all just die with agents and hopefully you make more money because we use more of it. Like our Salesforce bill went up from $12K to $22K a year. That's what you should aim for. Enough of these games. And if this next call is bad, dude, I don't know what the vendor is, but you're the boss. I wanna move. Like I'm already up, it's already given me, uh, uh, heebie-jeebies. Like deal it, just deal.

This is not the vendor I wanna partner with.

I just don't wanna partner. I just want 'em, uh, you know, unless they're a diamond sponsor, I just don't think I wanna work with 'em, you know? All right, let's move on. The very last thing I wanted to hit, just 'cause it was so interesting. This is just something you said live on stage in, in the final session when we went through all of our agents at SaaStr Annual, you said Amelia AI, which is our inbound agent, right? Built, built, uh, built mainly on qualified for now. It booked 614 meetings from 442,000 chats. Did I get that right?

Yes.

Yeah.

It's coming out in the—

Before we break, let's talk about that number just so we have it. It booked On its own, on its own, this agent booked 614 meetings from 442,000 chats. It's, it's almost unbelievably strong numbers from 2.2 million website sessions. That's our funnel to get to handing them to you and David to close.

Yeah. And actually the fun— the numbers are actually a little bit better because that's counting all of our websites versus just booking on saffronale.com. But yeah, it's. Here's the thing. I feel like folks kind of fight these numbers or contest them because— They do. They're, one, they seem too good. Two, you know, a lot of people are like, "Oh, Saastr has a brand, Salesforce has a brand." So, when I come to their websites and they have like a meetings booker, it's just easier to book a meeting. I'm like, "No, it's just easier to book a meeting." Like, why? I don't, this is one where I'm not sure why folks are so stubborn on this. I'm like, I don't— for outbound, sure, I get the hesitations and we just talk about maybe making our own, which would be better. But I'm not going to buy— or sorry, I'm not going to build an inbound agent. Like, it's funny, we just talked about maybe, you know, figuring out what to do with all of our outbound AI SDRs. One, we've already— we've always had too many, but two, now ours is getting better. But it didn't even cross my mind to try and replace or rebuild our inbound agent because it's so freaking good. Like, as you just said, like, why would I take something that is third party out of the box that I don't really have to, you know, is pretty much on autopilot, just booking meetings for us that then turn into closed one revenue. Why would I build that one myself? Like, I will just buy the inbound agent all day long because it's pretty good. If you train it right with the right context, you give it the right parameters, you keep it up to date, which does take some work. You do need somebody who manages it. But you could have one person who manages your inbound agent to automatically book meetings with your entire sales team, like versus having, I don't know, somebody in RevOps, which probably won't exist in a while, try and route inbound. Like, why wouldn't you let the agent just have all of its contacts and say, hey, like this person, which ours does, it'll say, hey, this person's more like AI forward, or they're more, they look like more companies that Amelia's closed than David's closed. I'm gonna book that person with Amelia. This person is like, and whatever, David has a lot of SOC 2 sponsors. It'd be like, give, you know, give that SOC 2 new inbound to David because he, he is like the king of SOC 2 sponsors at Zaster. So like with all that context, plus it can tailor the conversations it has to them in real time when they, and they can just book in real time an inbound. Like.

Yeah, I think, I think it's, I think, I think we could break on this. I think it's important. I think as long as you have any volume, this at least is going to work for you, right? That's the, it's not about brand. It's like, think about this for a minute. We have no humans in BDRs or SDRs anymore, but we are still able to, to handle with 2.5 humans, 442,000 chats and set up 614 qualified meetings. Like these meetings are all pretty good. Like we, like really crummy ones don't get through the door anymore. Right? That level of efficiency is impossible with humans. Like you could, you couldn't. Yeah. Forget about whether humans would be better. It's just not even like we couldn't hire enough humans that would stay in the job that would be willing to do this job. Right. It's just, it's not even, it's not even. That's the art of it is how much this can improve anyone's sales process. The numbers are pretty good considering, you know, no humans. All right. I think anything else we want to hit for this week, Amelia? That was great.

Thank you. No, I feel like we've, you know, I'm trying to give it. I think there's just, it's, it's interesting because part of me was like, okay, like what, what are we going to do with all of our agents, you know, after Saastr? And it's fine. We're, I guess I should have realized this, but I didn't realize it fully till now. We're just going to do more with them because now I have more time to focus on the agents and all of our flows and processes in a way that pre-event we couldn't always do the way we wanted to. So it's just interesting now to see how much, when you put more time into it, just to see kind of what happens, right? When you, when you actually can dedicate the time and just kind of set it aside to.

Yeah.

To, to deploying these and maintaining these agents. They just, they seem to just keep getting better for now.

So, all right, thanks everybody for, uh, if you made it this far, we appreciate it. It's the agents again, and we will see you on the next one.

Costs can add up quickly when you start a business, so stop paying for 5 different services. Northwest Registered Agent gives you a complete business identity in one place with free tools and resources to actually launch your business for real. Get more at northwestregisteredagent.com/saas-trfre. That's northwestregisteredagent.com/saas-trfre. EY knows you don't start a company to burn cycles on regulatory hoops or discounted cash flows, but you can't ignore it. That's how risk compounds. What you can do is work with EY. They'll help you get it right early so you can stay in builder mode. Go to ey.com/techstartups to learn more.