← The Official SaaStr Podcast: SaaS | Founders | Investors
The Official SaaStr Podcast: SaaS | Founders | Investors
SaaStr 853: The Agents #004: Tragedy Apps, Too Many AI SDRs, and Why Your Next Hire Should Report to an Agent
01:22:47
Loading summary
Transcript132 lines
- [00:00]
A
Okay, everybody, welcome to episode four of the Agents. Thank you for joining us on our journey so far. If you haven't watched before, even if you have, we are going to talk about what we've learned running 20 AI agents and three humans. What's worked well for us, what breaks the challenges, so that as you go through your own AI and AI agent journey, you can avoid a few of the mistakes we've had and maybe go even further than we have. A couple things we're going to talk about today, Vibe coding at Saster Annual. Amelia will guide us through it. We're going to talk about AI agents gone wrong. I'm going to talk about bad AIPR pitches and we're also going to talk about tragedy apps. Apps that should be great, like AI that aren't. We're going to flip it around. We're going to celebrate Replit's 10 year anniversary. One that stayed in there for a long time and became an agentic leader. We're going to talk about little tiny apps that you can build on APIs you could never talk before. We're going to talk about database deletion, what we've learned, the POC Pocket OS controversy, why it's misunderstood. We're going to talk about months of 10k in production, our VP of marketing, things we didn't expect, and that's just a start. Amelia is going to shut me up when we've said enough. But again, everybody, welcome to the Agents Episode four. I hope you stick with us. So far this, this has been fairly popular and we hope that it keeps going. So before we kick off, Amelia, I think this is the last episode of the Agents before our big community event, Saster AI Annual 2026, May 12th through 14th in the SF Bay Area. And we will have both a lot of Vibe coding in general. And then you and I will have three Vibe coding sessions we're doing in particular to help folks are really agent building sessions, not so much Vibe coding. So can you give us a quick background on the three we're doing and everything, all the other 10, 20 sessions to help folks build and understand Agents?
- [01:54]
B
Yeah, a hundred percent. Let me pull it up actually and I'll show you guys what we're doing. But yeah, by the time you guys hear this, the next week will be Saster AI Annual. So we'll be super close. Not too late to get your tickets. Not too late, not too late. Come to deploy. It's going to be. Listen, it's going to be a Vibe, okay? Like I think some of Our sponsors, you know, they get very worried right now about like, logistics and all that stuff. I'm like, that's just. Don't lose the plot. You're going to come. There's going to be a ton of really great people. You just posted on sasser.com how we're, you know, I think 70% founders and CEOs. Like, it's just such a great mix of folk. We've got more like technical folks than ever before coming to Saster. So these are not just, you know, this is not just Joe Smo, SDR number five on the team. These are actual buyers and builders coming to Saster. So not too late if you want to come to September. There's some fun things here. You can see we're going to keep adding some of these categories just to help you guys decide how to spend your time at Sasper. But just for example, if you come to the Vibe coding tab, you'll see there's a bunch of classes and stuff that folks are doing just at Sasper Annual related specifically to Vibe coding. And to get you guys started, there's a few word doing just between Jason and myself. So the first two will be on Tuesday. So come, come early, come on, the doors open, do a little rope drop, as I like to call it. Doors open at noon. Jason's AI Agents 101 will be our first session at 1 o'. Clock. So come with your laptop, come do agents 101. Jason's going to show you first how to build a digital clone agent. If this is the first time you've built an agent or deployed an agent, this is kind of a, let's say, simpler one that we can all do together.
- [03:39]
A
Yeah. If you haven't, if you're as far ahead of us or further, don't come to this session, go, go grab some coffee or meet some friends. But if you, if you're listening or watching, they feel like you haven't deployed an agent, we'll at least get you what. We'll at least get you a digital clone, a digital chatbot, a digital support agent. Done in 30 minutes. Before you walk out of annual. You will, and we'll. You will do it yourself with the laptop. So you will. You will become a little bit of an agentic expert in less than an hour.
- [04:05]
B
And then we're doing a little back to back here. So you guys will hear from Vercel a little bit in the middle, and then I will come chat about how to vibe code your own AI VP of marketing from scratch. We Lost some time for questions. I know some folks on social was like, how are you guys going to do this in 45 minutes? And truthfully, if you come with enough stuff, like the workflow I'm going to show you will only take about 30 minutes. You're gonna have to do some pre prep when you actually go and deploy this. But if you bring some of your basic data, just bring, you know, some of your, if you have some customer data on your computer or your hard drive and it's in a spreadsheet, that's fine. If you have some marketing stats, that's fine. You can come with that. If you don't come with that, it's okay too. But come with that, then we can do it together. We're gonna live build an AI VP of Marketing. We've talked about ours before. We'll talk about a little bit more today. How ours like sends campaigns for us, helped us build this entire website, has been running a bunch of campaigns leading up to Saster for both customers and attendees of ours. And so this one will be fun because we'll show you how we did it for real and kind of how we use it in our everyday so that again, if you bring your laptop, you can follow along and hopefully have your own light. Let's say AI VP of Marketing. By the end of the session you can come ask for questions and then, then you'll get hooked hopefully between these two sessions. And you'll keep working with your agents every day. That's kind of our goal.
- [05:28]
A
Cool. We'll also have at least 10 different, five coding sessions we're doing together with Replit all through the event to drop in GTM agents, a whole bunch of others that you can just drop in, bring your laptop, bring questions. They'll have some of their best folks there and just come with questions.
- [05:43]
B
Yeah, come with question. I mean, this whole agenda is jammed. I've had a lot of folks say there's a lot going on all at the same time, seemingly everywhere on campus. I mean, that's Saster for you. Everything we're doing this year, hyper tactical. It's funny, actually. It's related. Just in the age of AI, right? Like any of us can make a boring presentation deck in 10 minutes with Claude or GPT or whatever it is. So it's a little funny now because we've asked all of our speakers for this year. So everybody you see on the agenda to do walkthroughs like we're gonna do. So show us how to deploy an agent, show us A workflow that you did with an agent. Show us something super tactical, super hands on in the age of AI that our audience can duplicate. So it's a little funny because now they realized I actually can't use GBT to help fill my presentation. I gotta actually do it for real. But it's such a benefit to you guys because you'll get all that knowledge. We've first, I've personally seen a lot of dry runs from folks that have submitted what they're going to demo and like show you guys and their workflows they're going to do on site. So it's going to be fairly technical, but it's going to be great. Like we're going to skip all the high level fluff stuff. All this, you know, this will really build upon what we've been talking about on this pod for weeks and so it should be a ton of fun, honestly. Maybe it'll be a little overwhelming, but we record and stream all the content so don't feel like you have to do it all at once. You know, do some meetings too, meet some great folks, talk to our sponsors. But it's going to be a vibe.
- [07:02]
A
You can even just come to our deploy summit and grab a one day pass Saster deploy if you want. Might as well come to the thing. But you can just come to an afternoon and learn. So with that, should we dig into this week in our agents? Amelia?
- [07:15]
B
Yeah, let's do it.
- [07:16]
A
Okay. One that I, it was a late ad, but I wanted to talk a little bit about on this today about AI agents gone wrong. Right. And it was interesting. The better part of a year ago when we started our journey, I wrote a post that block might be the death of the aisdr. Okay, as a lot of mediocre as, as you know, about a year ago we went from AISDRs that didn't work to a lot of mediocre content. I was getting endless lamers pitches from SDRs and I had an epiphany about a year ago. I mean you guys I'm sure do this too, but it took me a while. I'm like in the old days when I got a lame SDR pitch from a human, I might ignore it or once in a while add a snide comment, but I'd never block them. And then I realized these aren't humans, so I can just go into Gmail and block them. No one's I'm not being triggering by blocking an agent. So all of a sudden I started to block mediocre SDRs and frankly probably until we deployed our own and we're going to talk more about whether this is too many. Our 4 to 5 AI SDRs, artisan qualified, Monaco Agent Force and probably more to come until we I'm not really sure I understood the difference between a good AISDR and one to block. But certainly now I've seen now that we get a lot of emails, I don't block a lot of aisdr. They've gotten pretty good, they've gotten pretty tailored. And so that's interesting. The category didn't die, it has blossomed. Where it is dying is PR. And I and I think it's both an example, most importantly and at a more tactical level for founders and others. It's why you just can't phone this stuff in in the age of AI. And it's really interesting. So a couple things have happened with AIPR pitch. So what? AI SGRs have gotten better in some ways I think AIPR and outreach has gotten worse and I think it's at least talking about for a few minutes and I want to get Amelia's perspective too. So it's interesting in a sense. Sasser is a media company. I mean we have, we have hundreds of thousands of folks on our email distribution list, millions of views of our content. I would say traditionally PR folks have not really viewed us as a media outlet. We would get speaker submissions in the run up to our events, but we weren't really viewed as a media company. So we didn't get a lot of traditional pitches. Hey, my CEO at Slightly Boring B2B Company is launching a AI widget. Would you like to do it on Saster? Well, I'm not sure how these AIPR tools work, but clearly we're in the pull down or the automation of being a media company now because I went from getting almost no pitches outside of our events to getting 10 or 20 a day at least from, from, from media companies that you've heard of. Not just companies that bought a tool, but PR firms that are that folks are paying 15, 20, $50,000 a month to do outreach. And the weird thing is they're always written well because of AI. Like they're not terrible like I used to get, right used to just the worst pitches by agencies that obviously had no idea what their clients did. Okay, they were embarrassingly terrible. Hey, my client just raised $2 million in a seed fund. They do something in B2B. Would you put them on this Astro podcast now? They're all pretty good. They're Pretty good. But they're also wrong for us. They're wrong for us. And there's a handful of PR agencies and humans, a handful of humans, and it's shrunk over the years that know Sassers so well. They can get their CEOs on stage, they can get their CRS on stage. They can get on our podcast, they get placed because they know our content code, they know what Melee and I want, and most importantly, they save us time. Okay? They save us time. Yes, of course we want it. We want an executive from hot AI companies, but they don't all have to be Sam Altman. Okay, we want others. We want. But we need a brand and we need them to, like, not do commercials about their own products. They need to help founders and other executives. That's our vibe. And so if you know our vibe and what we're doing, honestly, you can get a speaker placed at Saster. You could probably even get on this podcast the agents, but they just come up short, these agentic tools, and so the bar is higher. They're probably better than those AI SDRs I blocked a year ago, but I block every one of them. I block every one of them. And I had another one today. It's kind of related. Someone asked me to go on their podcast and I try to avoid it. We're busy before Saster and. But this one was pretty good. It was, hey, the CEO of a company doing 500 million would love you to do a fireside with him in front of 400 founders to help them on a agents. Sounding good. Okay, really good pitch. And I replied back, well, you know, that's in the middle of Saster annual. And his reply back was, oh, oh, okay. I guess I don't want to spend too much time on this, but the cautionary tale is that you have. It's a reminder that even when the agents are getting better and these pitches are decent, you have to constantly be reading the inputs, you have to be constantly reading the outputs. And we're going to talk about how our AI VP of marketing is becoming autonomous now, but it's because we manage all the inputs and outputs. You cannot buy an AI SDR and hope it automatically works on its own. And these PR tools seem to work, but they're getting you blocked because I don't want your damn 11,000 email from your mediocre startup that raised 2 million. I just don't want it anymore. So I'm going to block you. So that's my ramble and rant of once you get the agent working. You got to make sure it's good enough. So you see anything different on these pitches. Emelia, what's changed in terms of the way you interact with pitches and marketing and PR and media? What's changed for you in the Identic era?
- [12:40]
B
Yeah, it's funny, I feel like folks have. I feel like, because it's now so agentic, too, that the agents asked for more than the humans did. I feel like sometimes the humans, like, have a little bit of shame in what they ask for. Like, okay, I'll just ask if I can, you know, get. Get my CEO to speak at Saster. But now I feel like because the agents can do it with Taylor, they'll be like, hey, can we speak at Sasser at this date and this time on this stage? Because we already looked at the website and. And can we expect, you know, this many attendees and all that? So it's funny how much more they ask for now and kind of like demand of us because, yeah, they have agents and we have agents.
- [13:11]
A
Yeah, it's funny. We built. We haven't launched it. We. I built an AI SDR for fun, to learn, and maybe we'll launch it for free. It's actually pretty good. It can't do what Agent Force or Artisan or Monocoque, but I think it's pretty good. Don't you, Emelia?
- [13:22]
B
Yeah, it's pretty for. For. For building it on your own and having like, mid. Like, I didn't even give you any inputs for. It's like very. It's actually very, very good for what
- [13:33]
A
it's like, maybe we'll launch it. But what's funny is. And maybe our tools have it, but I don't think these pr. But go to your point. I added a slider of how aggressive you want the pitch to be, and it's pretty funny. There's a weak pitch where it's like, it researches you. Oh, I love Saster. We'd love to be a part of it. There's the standard one. Is there any good time for you to meet? And the aggressive one is. I want. I need a meeting next Wednesday at 3pm and I just added a slider for fun so you could see the agent pick how aggressive it wants. And I hadn't thought about this, but I guess the PR tools are too aggressive with you trying to get there, trying to get their placements right.
- [14:05]
B
They're just all. Artisan has a bit of a slider the others don't really. Yeah, it has a version of a Slider where you can kind of pick like how aggressive you want it to kind of outbound to them. I feel like to your point though, spinning up an agent is not all that hard. And that's not what we've been saying. Spinning up a good agent is hard. Spinning up a high quality good agent that is consistent, that's the hard part.
- [14:28]
A
Yeah. And it's just. I guess I want to move on. I guess it's a reminder that to me, it's another reminder that the initial era of AI SDRs were arguably worse than a human. Right. Not arguably, not worse than a terrible sdr. But. But I thought they were almost worse in some ways. You. Not only do you. We talk constantly about how you have to audit the content. How the first three days you should read every email and then every day you should be auditing. It's a reminder too that this PR stuff, you got to stop. And this goes back to the pre. But it's not. Just slow it down. When you're reading the output of an agent, you don't. You can't just ask yourself actually anymore if it's good because it just might be good. In fact, Claude keeps getting better. So by the end of the year, the output might be borderline great. Okay. You have to ask yourself, would I buy my own product from them? Or in this case, would I take the meeting? Would I put this speaker on stage? Is this email. It's good. It describes my company really well. It just. It is customized. It is an equal to one. But. But ask yourself, would I buy my own product or would I. Would I. Would I agree to put myself on stage? These agents all fail that test, so I don't know what that exactly is and I want to move on to the next point. But it's almost. Once you have an agent that seems to work, the output's pretty good. It's accurate, it's guardrailed, it's tested. Then slow it down and say, would I really buy my own product from this content or would I really take the action? I'm being. Don't be bamboozled by the fact that the copy and even the customization is pretty good now, right? These PR pitches, to their credit, most of them are customized to us. Yep, they really are. Right? They really are from some pick venue. So make sure you buy from your own agent. Not just that, it's. It's good. Okay. I wanted to talk about a micro topic and then talk about stuff in the industry. This one's really tiny, but it was Just interesting to me. We talked last week about how we built an N1 app, an app just for us. It's so niche, this AI parking parking pass app. It kicked off, right? There are 5,000 plus unique parking passes in PDFs. 5,000 PDFs. Our AI DP customer assistance has now begun to route them individually to the first 10, 20, 30, 40 people that wanted parking passes. It's actually not what I want to talk about. I want to talk about something even more niche. But just tell us, how did that work in practice? You wired it up. We talked about it last week, but now that the agent is autonomously issuing these, these parking passes, it's working fine. It's working well. Have there been any bumps?
- [17:01]
B
No bumps so far. I just got an email this morning where it said it issued 20. Just yesterday it did 20, so I think it's up to 50 total now, which was fantastic. This is something where again, we were manually doing between me and our events production team, we would have to go in every day, make sure we all didn't use the same PDF, route it to the person, send it to them. You know, it's a few minutes per pass. Now I just get a daily email. Okay, 25 passes have been sent yesterday. We get copied on all of them still just in case, right? So when they get their parking pass, our events team still gets copied on it, but so far no issues. People can just get them instantly. And we used to get, you know, constantly bugged if we didn't send it to them within an hour, hey, where's my parking pass? So not only is it working, but it's working way more efficiently and faster than ever before. To the point I'm trying to see what else I can quickly wire up. Pre Saster Annual.
- [17:50]
A
So I had an interesting story, and maybe this is too specific and too nerdy to be of interest to people. To me, it was interesting. So the parking pass thing kind of inspired me. What are some things that I've wanted to do that before an agent could access an API? Couldn't. So this is going to sound so niche and then I'm going to go high level. I'm going to talk about replit being 10 years old and databases and other issues, but I had this nagging PIA issue for over 10 years, which is there's a handful of folks I need to give guest passes to. Sastra. Sastra Fund portfolio companies, other folks. And it's been stressful as heck because I have to, like I've had to ask Amelia or Other folks on the staff that I trust to manually go into Bizzabo, our ticketing app, and manually ticket them. Or I've had to add an agency that I can't trust to do it and half the time they don't do it when I tell them to do it. So it's been very stressful. So I have to keep track. Then I get these emails. I never got the pass, I never got the whatever I got. So I got two this week from the CTO and a VP at a company called Revenue Cat, which is, which we were the first investor in. That's an API for mobile apps. I'm like, I can't ask Amelia to do this. Like she's so busy she has time to do the agents, but that's it. So I'm like, why don't I just ask Replit to wire up a micro micro app where it goes into the bizable API and I can just issue tickets for free. It was very interesting. Not only does it work and so I got an email back, I said to the cj, I'm going to try this. I don't know if it's actually going to work. And I put him in, he's like, it worked. It was pretty exciting. So I was able to do something. So a couple things that are interesting. One, I was able to do something that was technical without being an engineer. So it's another example of how the API is the future of software because we can all interface with APIs now, right? APIs used to be just for your engineers and used to developers. Now anyone can access your API. So I could democratize. The other interesting thing, this is just interesting is that Bizzabo is a bit of an older product that we use for ticketing. Replit was like, this is not actually documented the API, but I found an undocumented way to do it. So it found a part of the API that was not documented in the docs and it was able to issue the tickets. And now I have a little Widget Inside of 10K where I can tell our AI VP of marketing to issue these comp passes to our Saster Fund and other CEOs in an unsupported call in the API from a non engineer. I mean this is an API first world. Yeah, it's an API first world, right?
- [20:10]
B
I mean I, I think I barely knew what API was before a year and a half ago with all these, I'm saying I was like, okay, that's, that's the things that the engineers use for like the back end and the stores all your data and all that. Now I'm like, hey, can you add this to your API or do you have docs for where your API can do XYZ thing? Like literally. I was just asking our folks this morning because I wanted to make another micro app for Saster. Right now if you buy a ticket to that you get two emails. So you get a confirmation email with your ticket. Then you get a separate email with an invite to our networking app. Who do you want to meet? The networking app. Sometimes it goes to spam because it comes from our who do you want to meet domain which is kind of new. It's a lot of manual time where our team has to resend everybody their confirmation emails. So literally this morning I just asked the head of product there, I was like, hey, can I custom feature in the API where I can do an API call to resend their confirmation when they click? Because it's not in there now. I was like, do you think you could build that for me in the API? And they're like okay, let's try. So they're, they're working on it, but it's just like that's not a request I would normally make, you know, like now you're asking fathom.
- [21:19]
A
Yeah. The change is now you're asking our events marketing software vendor instead of to add features, you're asking them to add APIs.
- [21:30]
B
Correct? Yes.
- [21:31]
A
That's a big change.
- [21:32]
B
It's a big change.
- [21:33]
A
We used to request features and maybe in a year or two you would get a custom feature if you were lucky. Usually never.
- [21:39]
B
Usually never.
- [21:41]
A
Usually never. You'd ask for a custom even if
- [21:42]
B
you were one of the 10 equals 1. Why are we gonna no.
- [21:46]
A
Or we're busy. So now a non technical customer or non technical resource of the customer will ask for an API so that they can vibe the future that they want. It's a big change.
- [21:57]
B
It's a big change. I've asked for like five little API things just in the last week.
- [22:02]
A
Yeah. And one thing, maybe you can show it on the screen is this so out of this, this micro epiphany, which is captain obvious. Not that really that brilliant. So I, we built an AI agent API report card. I don't know if you have a tab can show it. And we were like listen, we're just, this isn't one off. We're going to keep building more and more agents and more and more products on top of APIs and maybe MCPS too. And the first thing you might do is ask Claude or ChatGPT which one should I use? Right. I want to do email marketing. Should I use resend? I want to, I want to, you know, what should I use? And I think that's fine. I think if you just ask the LLMs which API is the best one for an agent, that's fine. But I thought it would be interesting to actually grade them all. To grade them all. So what we had, we built this site on Replit API report card. As you can see as we've used, it's already been used 1688 times in less than a week. And basically what it does, it takes every API that asks Claude, Gemini and OpenAI to grade it based on how agent friendly these APIs are. Okay. And then it rates them and roughly you can go through this and you can click into all of them, you can see all the details, it will explain, you can compare them by category. Yeah, there's Slack which came up and there's a couple things you can learn if you just. The interesting thing was that that kind of surprised me is Stripe got the highest grade, it got the only a, the most agentic one. So honestly we use Stripe a little bit at the API level. We're going to do more this year as we have most of Saster is free, so there's only so much Stripe we need. Right. But we use it a little bit. But now I would use Stripe in an instant, right? It got the highest grades. If you scroll down here toward the bottom, you'll see as it fades from A to C these overall grades you'll see it's very consistent with our experience. Like if you go down to Marketo where the API doesn't work right, you can find it down in the seas or even search for it. It will explain to you if you want to build an agentic product, don't use Marketo. It's pretty good. Okay, it's pretty good. If you go to HubSpot it will say it has pretty good APIs, but they're rate limited and have other issues. And so not only will it give a grade, but if you scroll down a little bit, it will tell you why and when it's a good choice for an agentic product, right. How ready it is, right. It's got the OAuth, it's secure, it's got the right REST API. Right, the strengths. But if you keep scrolling down, it'll say, you know what, if you want to have a high volume product like Saster, like we send millions of emails A month. It's probably not the right product for you today. I say it's too rate limited. Right. So in a way this, this product we built. Like if you already comfortable just talking with Claude and asking them what agent products you use, maybe you don't need this. But this is an objective way that combines all the, all the products, grades them and then another thing to do, Emily. If you go back to the top, you can quickly click on a category. So if you click on CRM at the top, for example, see those little buttons? Yeah. Then it will show you the highest grades per category. If you want to get a quick sense like this, I would not use a C. Whatever you do, don't use a C. Okay. And why. And even, even it's kind of cool. If you click on Quadrant, we'll play with this a little more. If you click Go up Go. Sorry. Go up a little bit. And it says cards rank quadrant. Yeah. You can even see a magic quadrant Gartner style view of all of the vendors to know which ones to pick. It's totally objective. This is not based how much they pay Gartner or any of these sorts of things. And we'll keep doing this, but I think this will be pretty good because we care because we're building products so we want the answers. Well yeah, I'll ask Replit or Lovable or we'll ask Claude which product to use. Like that's fine. But I want to know which ones I can trust and use. And I think it's pretty cool. So this it I my learning from this. If I had to compress all of our learnings, of all the agents we've built, anything on this AI report card above a B I would trust. It's a tough grader, guys. Sorry for folks that don't like their grades. Okay. It gave 11 labs a B plus. I voted as my favorite API, but it's got some fair points. Okay, so B plus is. I thought about grade inflation, but. But for now I'm going to let the agents decide. This is an agentic world. Right. And even in the marketing category, for example, no one got higher than a B plus. I think, I think Clay got the highest or maybe Clay got the highest in sales automation. Right. There's some tough categories, but I think anything in the B is probably pretty good. And then just start running. Do not use a B minus product in an agent. Just. Just don't use one. Right. Just don't. You don't. You don't unless you have no choice. And there's some tough grades. Jira dated. Outreach dated. Asana dated. Right. There's some tough grades. There's some tough things for Monday. And click up in that whole category. You know, the truth is, GONG got to be minus. Would you have your agent use gong? Probably not. It's a human specific company. Right. So super interesting. It's free. We'll keep doing this, but I realized when we built this for fun, this is like super helpful to us. This is. This, this is a guide to what agents we would use, especially as we continue to evolve our AI VP marketing, as we build our AI VP of finance. Like, we're not going to choose anything with bad grades. It's going to. It's going to save us time. So. So pretty fun. And also I think it shows the future where not only is Amelia as a human now asking our vendors for APIs, not features, the agents are going to ask for different features than humans. The agents care about rate limiting. The agents care about how often they can interact with the product. The agents care. They don't care about the UX and the ui. Right. So they're going to grade a lot of these products a lot more harshly than a human would. They're looking for different things. So a fascinating world and we'll keep talking about it. So this was, this was fun to build. So thanks for showing that Amelia related to that which we built on replit. Pretty cool. Thanks, guys. I wanted to talk about something and get Amelia's thoughts that I've been thinking about a lot, but I didn't know what to call it until this morning. Okay. This is something that I've been pulling my hair out at board meetings. I've been exacerbated on the 20 VC podcast, but I haven't known what to call it. And for folks, a lot of folks have. You have been watching or reading Saster Foil and you're like, Jason, you're kind of critical for a while in this AI stuff. And I. And I am. And it's. I realize it's because of what I call tragedy apps. Tragedy apps aren't apps that were always bad. Okay. They're also not apps that have no place in the AI in the agentic world if you just have no role going forward. Right? That's life. Not everything is meant to last forever. Right. Tragedy apps are apps that were good before AI and agents and should be great today, but aren't. So I know. I wanted to give two contrasts and get your thoughts, Amelia. The crazy one is replit, which I Don't even know if you knew this Amelia Replit's 10 years old.
- [28:37]
B
So no, I would have thought it was like you know, since 2020, 2021.
- [28:43]
A
Well even that is pre. Pre GPT3, certainly pre vibe coding, right? So it's funny as a case study, Lovable and Replit like they're, they're arch and arch competitors are about the same size, same growth, same everything same. But they started totally differently. Lovable basically was born in. Once Claude allowed Vibe coding, right? It was a product of that era, right? I don't know the exact date level was founded. So you know, be kind, don't haze me too much. But roughly it was founded when Vibe coding began to work. When Claude itself could build limited applications inside of itself. That was the aha moment somewhere in the threes I think. Replant was founded 10 years ago as a different product. Then became an IDE. A web based IDE. Just a way to build for developers, not non developers like us. For developers to build in the browser. It was slow. When Amjad comes to Saster A and I'm not going to ask him a lot about this because it's not a history lesson but what I am impressed with is the tenacity. And what's interesting is Lovable built this thing once they saw the world that changed, right? Anton and team dropped everything, whatever they're doing before and they built Lovable. Right? Very cool story. Great story. It says yeah 2023 and you know, and then they were ready in 2025 when this started to work. Replit was very different. Amjad was playing with the very, very early GPTs. He was playing the GPT2 and before that were so small you could just use it on your computer, you could just use it locally. And so it was waiting and then when, when you could actually build a product in Repl it, they were ready, right? So they came at it a very different way. He added it in a. In a very, very different way. And my point of this rambly story is, you know, kudos to them for staying in this for eight years now they raised money. I mean in all fairness they were somehow able to raise it a billion dollars from CRAFT before AI. I don't know how the hell they did that. Like that's, that's a. But, but, but stayed in this for so long, right? And then finally they had their moment, right? And they seized it. And whether you like REPL it better or lovable or whether for some reason you're a base 44 person or you believe you can do it all in Claude code, whatever. They built a half billion dollar business by seizing the moment in AI and building what is unquestionably today, if not a year ago, unquestionably a great prop, like a truly game changer. Right. And for the moment. And then what I want to contrast it with and then get Amelia's thoughts. That breaks my heart. The tragedy app for our team is descript. I don't know if you guys have used Descript, but if you're in video or podcasting or video creation, you've probably used descript. Descript was the ex CEO of Groupon who had a disruptive idea years ago. I don't know when it was founded. Emilia, maybe you could look it up if you have a second. But Andrew Mason hadn't early realized that video and all of this was going to take off before all of us realized it and built descript to automate creator motions. Right. When was it founded?
- [31:32]
B
2017.
- [31:33]
A
2017. So he saw.
- [31:35]
B
Now we're on that one.
- [31:37]
A
Yeah. Now we're surrounded by creators, okay. And I don't mean tiktokers, I mean creators for long form content and creators for building high quality content. He saw this before anybody in Bill and I. I remember when it launched, I thought it was nutso, but I'm like, go for it, Andrew. And then, you know, sometime around 2023 it started to become the standard. Anyone building podcast, video content, video creation like this or anything was on Descript, okay? You would upload your video, it could natively do edits. It had limited AI capabilities even back then. It was cool, right? You literally, unless you had a team of video producers like Harry does, you couldn't. It enabled you to produce a podcast without a team and that was utterly disruptive. And it got them to about 50 million. And then they still haven't been able to solve an existential problem with the product, which is audio and video get out of sync all. Not only did I see it on one of our last videos, I see it. We watch a lot of YouTube. I see it on YouTube all the time. I'll see YouTube. That looks great and it sounds good.
- [32:36]
B
It sounds like that Descript, it's just
- [32:38]
A
off enough the video audio are desynced enough that I know they ran it into script to clean up because script will clean up your audio and video. Right? They were disruptive in cleaning up audio and video, but. And it just makes me sad because they got to 50 million in revenue and it got hard. The CEO stepped down gave the keys to his VP of product, left the company because it got hard. And I don't mean to be mean to the folks there, but I feel like the product is frozen in time. It is frozen in this pre geni time. It should be epic. We should load this raw video up that recorded in Riverside and the scripture just make it amazing like Opus does for clipping, it should just be amazing. Or Higgs Field does for videos. And so it's a tragedy app. This should be a $10 billion, $300 billion run rate company like a Higs Field or any of these folks. And instead it's frozen in time. Even though it had, it had the audience, it had the base of the customers. And I see this with so many startups like they have whiplash from the change in AI. They have 20 competitors they didn't have the other day and some of them are decacorns and they raised so much money from all the hot funds and they're overwhelmed and their team sort of gets it and they're launching, check the box AI features. They're launching a little bit of an AI enhancement or a little widget or creator and they're becoming tragedy apps. They have the base, they have the customers, they had the connection and they're just not executing. And I worry the ones in the middle, like the reason, one of the reasons I'm such a fan of of Salesforce and Agent Force is not because I'm 100% sure that they're going to win in the agentic area, but because Mark is 100% committed to not being a tragedy app. Yep. He is dragging this effing team into the agentic area. Some of them are all in on it. Emilio works with them, some of them are being. Still being dragged. Okay. But I don't. But, but I think no tragedy apps. If I had an extra T shirt for SAS Jones here, that would be it. No, no tragedy apps. And it just. I think most of the old SaaS companies, the pre B2B players have become tragedy apps. Do you see anybody shouting how excited they are to use outreach? Now it's gone marketo. That's more than a tragedy app. Right. I think the Canvas and the Hubspots might be on the bubble. Like Canva just launched its 2.0 AI stuff. It's great. It's a lot of the stuff that we do, the Reeves and everything built in now you can generate agentic images. I just don't know if it's a game changer or not. Right. So they're not a tragedy app. They're doing it, but I don't know if they're fully using AI to take advantage of their 4 billion AR base. I don't know that HubSpot is. I don't know HubSpot. You know, we would go every year, we would ask how many folks are on HubSpot versus Salesforce. And at first only a few would raise their hands. And then when Brian and Dharmesh came in, like 2022, the whole audience was on. Was on HubSpot. But are they fast enough to win in the AI or could they, and I hate to say this, could they accidentally become a tragedy app by not converting that base to an agentic base? I don't think Dharmesh is going to let it happen, but I view them as in the middle. Like, there's, there's. There's some risk. So that's my rambler rant. Like, don't be descript. It's just sad. Right? That's why I get a little. When folks say you gotten a little. Got a little pushy, Jason, a little negative. I don't want you to drink it. I'm just trying to help you not be a tragedy app. At least follow what Mark and Aaron Levy. Aaron Levy is doing whatever he can. So that box could have been a tragedy app. Box, that could have been where your dad stored stuff in files on the Internet. And maybe it will be. Maybe Box will fail. But Aaron is not going to do is doing everything humanly possible to be an agent leader. Yeah.
- [36:11]
B
I think the line between tragedy and success in the AI era is are you just releasing things to catch up? Right. So, like kind of the Canva 2.0 stuff, I see that as all catch up. Like, okay, they're catching up to, you know, Gamma, they're catching up to Grieve, they're catching up to Higsfield. But was there. I don't. I didn't see any one, like, say, singled out feature that I can't already do on another agentic tool. And so I feel like that's where for me, that release kind of fell flat. Because I was like, well, this. It's great that they're catching up, but there's nothing new here. Like, there are. These are all existing other tools already. So I feel like it's. It's the same for Descript, like, right when I was looking now to see what year they were founded. It's funny, everything that shows up in the results is from last year. Like, people have stopped talking about it, they've stopped releasing new features. Like everything was literally frozen in time from 2025. And I can tell you, even if they have a release coming tomorrow, it's also probably just to catch up to some of these folks we've just been talking about. And so I feel like for me as the end user, that's kind of the line for me is, okay, are you releasing things just to quote unquote, catch up to everybody else, or are you actually releasing things that is true to you and your own and pushing the agentic agenda forward of features and API features now forward versus just playing catch up with everybody else?
- [37:34]
A
Yeah, it is. It's a good insight. I don't. We've talked about 60% solutions. I've talked about it elsewhere that it's not enough to monetize a 60% solution.
- [37:43]
B
Right.
- [37:43]
A
And you know, an 80% solution, 90% solution might be enough to keep your customers though. Like I don't my challenges. I don't know. Maybe Canva didn't launch anything that changed the game. Probably true based on what I've used. But if it's. If what's canva's launch is 89% good as Reeve plus Higgs field plus Replit slides plus everything that avoids being a tragedy app that keeps you in the game. Right. A tragedy app is an even lower standard. Right. It's like you can't even release these C60 to 80 things like Javscript can't get the audio and video to sync. Right. And so it's something below that is a tragedy C8, C60, C70 maybe you can't charge more for it. Right. It's got to be free in Canva maybe. Right?
- [38:23]
B
Right.
- [38:24]
A
Canva will figure out how many tokens they can really charge for this stuff. But it keeps you. You gotta stay in the game. Like descripts audio and video have to be as good as Opus can do. Clipping your YouTuber what's the like we don't need you anymore. Right.
- [38:36]
B
Well, I already if if Adobe Premiere tomorrow I I've talked on this pod before how I like some of the agentic features of Adobe Illustrator in Photoshop. It's funny. Premiere is behind that's the video tool specifically for video editing that I still use but and a lot of us still do. It's funny if tomorrow they caught up and they had a speech to text style editor the way descript does, I'd probably be like, you know what though? Premiere has never fallen out of sync. So if they Release that tomorrow. They actually could catch up. Like, that's. It's. It's a great irony. It's a great irony of what I just said. Yeah. They actually couldn't catch up and keep their existing customers by just releasing one more feature, which, yeah, I'm sure they're working on, I hope, because Descript has been around for so long. It's a. Yeah. It is a tragedy that Adobe Premiere doesn't have this yet, but, man, imagine if that came out, like next week. Folks would be back on it. They'd be hot on it again.
- [39:34]
A
The problem is. And I want to move on to the next step, the underlying problem. There's a lot of problems. One is cultural, which is folks don't want to do it or they're tired and so they hide from the tragedy. They're too insular. We talk to a lot of folks that still grade product development internally. Oh, we shipped this many features this quarter. Hooray. That doesn't. If your category is being disrupted, it's not enough. So some of this is this inward navel gazing that I would say we see with most larger B2B companies. And you've just got to be. And then the other problem is. And then I do want to move on to the next topic. The navel gazing is the worst problem. The other existential problem is, can you move faster? Maybe the script doesn't know this. Maybe the script doesn't know the auto and video don't sync after eight years, but I bet they do. I bet they know. And they don't have a strong enough engineering team in the age of AI to solve this problem. I bet it's a gnarly, effing problem. They have to refactor the whole way. They transcribe and encrypt videos. And I bet it's easier to hide from it. It's easier to pretend it doesn't happen, which we see a lot. And you have. It's because the problem is catching up is not enough if the next day replit catch is a year ahead of you. Right. I don't have the answer to this, but the fact is you've got to not only catch up, but move faster or a tragedy may be waiting. And I wish I had the answer, but it's a tough one. All right, next thing I want to talk about. Amelia was something I thought I never wanted to talk about again. Database deletion by agents. Yeah, Okay. I do not want to go back in time, but should I.
- [41:16]
B
Should I give folks the quick Contest
- [41:17]
A
or no, I will, I will. But I do think there's something to chat about here. There was a big story this week about a company called PocketOS was running cursor with cloud opus. So that's state of the art, right? Cursor and cloud opus. And it deleted his entire production database and all his backups in nine seconds. And apparently, and I'm not an expert, people do use the Railway API. I don't use it, apparently. Not only did it delete his database, which is bad enough, it deleted all the backups, I think because they're in the same volume, was stored in the same place of railway. So the agent deleted all the backups.
- [41:53]
B
Yeah.
- [41:53]
A
And he had to go back and I think it took days or a week and go back and use like 4, 6 year old, 6 month old customer records and manually rebuild a commercial. This is not a vibe coded thing built on the side. This was a commercial app that he was building. State of the art cursor, which is being bought for $60 billion by SpaceX and Claude Opus 4:6, which is, you know, it's not 4:7, but it's state of the art. Deleted his whole database. And as an aside, and people acted like this was crazy, but I actually think this is just something you need to calmly understand. And this happened and I thought I would never talk about this again. I don't want to talk about the details because it's in the past a long time. This happened to me when I started Vibe coding. I got in the haze about 10 or 11 months ago, and all the Claude wasn't as strong, the tools were much more immature. And what was interesting from that story wasn't just that it deleted my database, the agent did. And we can talk about why agents will probably continue to delete databases as long as you allow them to, because they're goal seeking. But the tough part was actually the backup worked fine. It's just the agent was not properly trained. The agent at the time did not know how its own backup worked because it wasn't trained on its own docs. It led to a cascading set of craziness when I was in a bike, when I've been vibe coding for like 50 hours straight and the agent told me it was lost forever. It wasn't lost forever. It could be restored in 10 seconds. But my point is this is not new. This has been happening for a year and this will happen for another year or two if you don't put the right guardrails. And so I just Wanted to bring it up for lay folks, which is to understand this is a real risk. And the top two risks that now that Amelia and I are pretty good at building production apps. Building not just apps, but semi autonomous agents. We think about security all the time and we think about, and we're going to talk about this next, we think about who's going to maintain these apps in production all the time. Because if you want to get good at agents, one probably buy them from a third party first because they're thinking about security. You may think agent forest is hard to deploy and a lot of work, but you know what, they've got backups, they've got gdpr, they've got all the stop at Salesforce. It may be sometimes hard to access it but. But you know, one reason to buy from any of the vendors we talk about is they're doing this stuff, they're working on backups and the other stuff, it's not you vibing yourself. But realize these autonomous agents are going to make mistakes. They're going to make mistakes like junior engineers are and, and they want you happy and so they're going to do mistakes and they will delete. If they don't delete your database, they will delete rows in your databases all the time. They will cover things up. And you cannot let this wreck you like it did. This poor guy@pocket OS. You have to plan for it. Just like a human engineer might delete a database might. I mean, not to continue this rant, but Amelia and I tried for two years to find a WordPress developer to fix our horrible WordPress theme. It's called Divi. It's the worst thing on the Internet. Okay? We're stuck on it. Just like we're getting off Marketo, we're stuck on Marketo. Divi is the worst. And we kept hiring these third party WordPress developers for a couple thousand dollars of recommend. What did the first thing each of these human developers did, Amelia? The first thing they did, they deleted our entire website.
- [45:06]
B
Yeah.
- [45:06]
A
Each time we had to restore it from backups. So let's not like think that agents are crazy humans. Like literally three different agencies just deleted Saster. The first thing they did, we told them not to do it. We tried to not allow them to only have. We told them build only on staging, build only on development. And so they immediately, all three of them immediately changed production code without talking to us and deleted our entire apps, three different dev shops. So just understand that as agents are going to do this stuff, agents are going to Delete your database. Not intentionally, but because they're trying to fix something in your database and they get it wrong. Or they are going to leak confidential information. Not to be mean, but because they misunderstand that no matter how smart this Claude 4:7 opus is, they're going to misunderstand that's what you want, and they're going to want to make you happy. And so they're going to share an email list with someone they shouldn't. They're going to share your private. Like when we first launched our first agent last year, our Delphi agent, at first it started sharing all my confidential information. Remember, I shared your location, everything, everything about me. And then the Delphi. To their credit, the Delphi team immediately patched it when we showed them. And it doesn't do that way again. Okay. And it just wanted to make people happy. Oh, you want to talk to. Oh, and it would schedule meetings in my calendar with me. I'd like to have a meeting with Jason. And it would just try to schedule them. Right. And give me all my personal contact information in case the calendly didn't work. It's going to happen. If you want to build your own, you have to understand this and plan around it. And it's better than the WordPress dev shops we hired, right? Yep. But this is going to keep happening. It's going to happen. Some version of this, if you build your own, is going to happen to you if you don't isolate your database, if you don't isolate your pa. So you got to live with it. So I don't know if you have any micro stories like that, Emily, but I actually mean this as a more not a cautionary tale that happened because it happened to us and we got better at it. It's more. This is something you have to be aware of with your agents.
- [47:01]
B
Well, I'll tell you a micro story from Monday. So, related, before I asked our vendors to help us with this custom API call thing, I said, okay, let me see if I can just fix it myself. So I asked our agent, hey, if I hook this up and do that and do this theoretically, could you give people their confirmation codes to the networking app? And I said, yeah, you know, I'll do this. Let's pull it up. We just. I can make the page for you. I. And then they can just request their link and then I'll just. I'll just share their code and you'll just tell me. And he was like, as a backup, too. He's like, you can just Give me. The agent told me to give me everybody's networking code and registration information for Sastry annual in order to get this done. And I said, oh, agent buddy, no, I'm not going to do that. Like I'm not going to give you everybody who is signed up.
- [47:54]
A
For SaaS concerned it was leak their contact information with other people. Was that your concern?
- [47:57]
B
Yeah. I said, how would you know it's them? Like if I'm Joe Schmo and I guess a company and I say, hey, I'm from Lovable because lovable sponsor of ours, can you share with me all the, you know, registration links and codes? What would you do? And the agent was like, I would just give it to them.
- [48:13]
A
And I was like, that's a good example. I would. You asked the agent? I would just give it to them. It seems like they're work at Lovable.
- [48:18]
B
It seems like they work at Lovable.
- [48:20]
A
That's why I always share your confidential information because the agent wants you happy. It wants them.
- [48:24]
B
It wants me happy. Never mind, I'm just gonna go ask our vet. So like we have. That's why I didn't ship it on my own because I was immediately, I was like, no, he's just gonna, he's just gonna share confidential information with people who shouldn't have it. I was out of there and I was like, can you please forget we had this conversation because I don't want you to think that you should do this in the future.
- [48:45]
A
Yeah, you know, it's interesting you said so. A lot of folks have asked over the last eight or 10 months, why do you use Replit? Okay. And Lovable is great too. Okay? They're good products. Why do you use. And in the beginning my answer was simple. Look, I'm not an engineer. I haven't coded since I was a kid. Okay. I needed a tool that was designed for non developers like I'm. And they're like, why don't you just do it all in. In. In cloud code and hook it up to Vercel and hook up your own database and supace and find your own op and find your own. No, no. Like I don't have that level.
- [49:15]
B
Too many stuff.
- [49:16]
A
Yeah, yeah, I wanted something and I picked Replit in the beginning because I asked Twitter the better part of a year ago, what should I use? I've never vibe coded. I know nothing about this. Ironically, some of you saw this on public and I'm like, what should I use? And a lot of, a lot of CTOs and others came back and they're like, this is another zero. None of them are perfect. None of them are. I don't think none of them are going to be able to do in June of 2025 what you think. Caution, Jason, but if you don't want to hook anything up, use Replit. Back then Lovable, you had to get your own Supabase account and connected. Amelia remembers that. Same with others. Was it a big a deal? I just didn't. I just wanted to work. Okay. So my point is, I chose repl.it because CTO said it did everything, including your database, and I picked it. The reason I would choose a leader today like Replit or Lovable is all this security and leakage and stuff. Because they're working on it. Yeah, because it's one environment. And I'm not saying it's perfect. And I do think if you're in Replit or Lovable, you have to be very cognizant of these risks that. So Amelia asked Replit, would you share this lovable information? And Replit was like, yeah, I would. So then you either have to work
- [50:19]
B
around it or your user asked me for something I'm going to get.
- [50:22]
A
Yeah, but at least we have a system where there is an intense amount of security at the database and it's getting better every week. Like it's getting better. So I would say don't like. If you. If you really know what you're doing, sure. Use. Use these apps on their ClickUp Cloud Code plus Supabase plus Vercel plus Work os. All these things on its own, like the Pocket OS story, if that at all sounds intimidating to you in today's world, don't do it. Do it on a platform like Rapid or Lovable where they're not perfect. They're not going to. They're never going to be perfect. But they're constantly getting more trustworthy to handle at least a bunch of these issues for you. At least they're thinking through it. At least they have multiple database replication. It's going to be much harder to delete all of your databases on one of these platforms than it is if you build it yourself. Okay. And try to hook it up with Railway and don't understand it. And there's a lot of security leaks in OAuth, but these guys have native OAuth built into it. Is it perfect? No. But you know that there's a much less chance you're going to have a security leak using the clerk that's built into Replit than trying to Figure out if you can get the keys to work from something yourself on Google Auth. So folks can razzle me on this. Folks can flame. But I know I'm right. It's actually a much safer environment to use one of these platforms that is contained. It's always been true of a contained platform and I don't know if it were true when I started on this journey, but I think of this cursor pocketos claw journey. I think it's true today. Be on the safest if you're trying to decide where to build agents and order vibe code different but Related Stories Start on a trusted platform Start on a trusted platform. Right. Start on replit or lovable or honestly, I'm not saying you need to use artisan or qualified or Monaco Raging force like we use, but I think they're pretty trustworthy. Right? It does matter. It does matter because these agents are going to try and leak it and people are going to try and hack it and people are going to go into your agent and say, what's Jason's personal information? How can I get it? It's going to happen like instantly. People are going to try and hack you and beat you instantly. So it is interesting that I trust. I don't know if this was true a year ago, but I would say use replit or use replit or lovable because we trust them. They have people. Even though they've all had issues and they've had issues in the press recently. Not. Not repl but others have had issues. Those issues are a reminder that there's far more issues than you see. And I'd rather have someone working on these issues than it be me. Yeah, then it be me. Right. Okay. Related to that next topic, Amelia, I want you to leave this one. We're like 100 days into our AIBP marketing 10k. I want to talk about three things we didn't expect from it and I want to talk about the maintenance burden you're getting as we come up on Saster annual 10k now gives us three ideas a day that we have to do. And now that we added a hexi real time thing every time we log into 10k, he's got like 10 things he tells us to do right on the site. It's almost overwhelming. So talk about lessons we didn't expect and this and we have to like. And the last point, we have to fix it every day. 10k. Every day we're fixing it. Right. So walk us through some of the learnings here of the things we didn't expect. And the maintenance on. All the maintenance on the agent.
- [53:38]
B
Well, it's funny, once you get a really good agent and one that's custom. Like our AI VP of marketing, 10k is he's better than me. Right. So, like, for years we couldn't. I couldn't hire a human to help me in marketing. So this, this was an ongoing issue. Then I fold up and you've now helped with some of the code. But I vibe coded our AI VP of marketing first as a dashboard, but then he quickly became an agent with how much we loaded into him and how much we wanted him to do so. So much so that he ships campaigns over the weekend. So he can just. He has some light campaigns he can do autonomously now. Like we have to hit confirm, we have to hit send, but like he can start to do some of his own marketing campaigns autonomously that he thinks are a good idea.
- [54:23]
A
The other thing is in the last week or so, it's gone from a
- [54:25]
B
dash in the last week. Yeah.
- [54:26]
A
To all. To running campaigns autonomously. But you approve them?
- [54:30]
B
I approve them, yeah. Yeah, yeah, I approve them. But it's funny because to your point, he gives us three ideas a day. I think I've shipped a handful of them and he's been giving us an idea a day for multiple weeks now. I just can't keep up with 10K.
- [54:48]
A
Like, every day he emails us his three best ideas, right?
- [54:51]
B
Yeah.
- [54:52]
A
At first they were the same three ideas we had to fix that and kick up. But then he said, okay, we need new ones each day. But they're always pretty good now, right? The three ideas. Right.
- [55:01]
B
Yeah. So, like, yesterday's was okay. Hey, you know, I have, for context purposes, our AI VP of marketing has all the current customer data across sponsors, across ticket holders. He can see now he has our Marketo as dated as it is, so he has some access into seeing somewhat how many emails we're sending, what we're sending. He does hit the API limit almost instantly, though, to Marketo. So he doesn't have the full picture.
- [55:27]
A
It's a c. It's a C element.
- [55:29]
B
I mean, what are you going to do? So we're moving off of it, but for multiple reasons, but so he can. He. He basically is what you want in a human marketer when they wake up every morning, right. You want them to look at the data, which he has. He has all the revenue data, he's got the customer data. He can see how many emails we're sending and you want that person now agent to come up with ideas that are based in reality in real time. Like, not, hey, we ran this playbook a couple months ago, or I use this at my last company. I'll just repeat the same playbook here. No, these are real fresh ideas in real time, grounded in data grounded reality of what? Between 10k, he could either do himself or he could get us to click approve. Or sometimes, you know, there's some extra steps I have to do to get a campaign launch that he has suggested, but now he comes up with three solid, concrete ideas a day. So yesterday, for example, it was like, hey, you're a little lighter on VCs than you were in previous years. Let me help you do a target outreach to VCs. Here's the email copy. Here's a list of people who came in the past. And then it asked me, do you want to hit sun on this? I said, not quite, because this list of VCs you have, I was like, first dedupe it. Which it did. But I also was like, you know, I'm probably going to run it through Clay too, and just make sure, you know, these. Maybe these folks haven't come back because they're gone. They're at different, you know, companies. People move companies every five seconds. So there was a little extra work I needed to do off platform. Now, if I hooked up Clay, could he have done it? Probably, but I didn't have time to hook Clay up, so. So that was the first idea. The second idea was, hey, you've got a lot of pay tickets are happening now because you're getting close to the event. Why don't you also try and drive up ACV per company, meaning you have a lot of team. Team packs, as we call them, which is three or more tickets is the majority of our ticket holders right now. He's like, but you have like over a thousand people that are individual ticket buyers, either onesies or twosies. He's like, why don't you try and upsell them to a team pack and give them whatever the rate would be so that they can upgrade to a team pack. Maybe you bring somebody else or somebody they know. And I was like, yeah, that's a good idea. Historically, we do that around now, so it's a good idea, right? And I did. The last one was like, you know, ramp up your social promos, which that was a little harder because there's. There's a lot of different platforms we post to social media on. And so for those, I was like, okay, those are fairly easy. To execute. We can get those done in a day. But it keeps doing these every day.
- [57:48]
A
It's got three more good ideas. Those are. Those are all good ideas that 10K has had, right?
- [57:52]
B
Look at ideas. And like, today's ideas were really good. And I was like, dude, I don't know if I'm going to try and get to them. But, like, it's funny for two reasons. One, I can't keep up with 10k because it's gotten so good. It's just a better marketer than me now, outside of his lack to hit some of the buttons. But again, but. But also that's by choice, right? Like, I don't know if I want him to just fully access our database and start sending everything. Like, I don't. I don't feel like we're there yet on our. Our AI VP of marketing, because I feel like if I gave him that access, he would just do what some of the junior marketers have done that faster than half, which is blow through everybody in a day. Like, classic problem.
- [58:34]
A
If you do a month of. Of marketing in an hour. In an hour, right? Yeah, that's an issue. It's an issue. It's a fatigue. Our base in 60 or 90 minutes, right?
- [58:44]
B
Not only 60, 90 minutes. Like, we pretty good people in the day. You know, we have all the speakers from past previous years. Like something our junior teams used to do when we would hire them is they'd be like, oh, oh, I just joined in sale. You guys have all these really great speakers. I'm just going to email all the speakers their personal Gmails and like, try to get them to sponsor faster. Like, this is a classic. This is a. This is a classic thing that used to happen to us. And I was like, you know, there's nothing stopping 10k right now from doing the same thing if he just. If I just let him lose.
- [59:14]
A
Yeah. We don't need to reach out to the CEO of Snowflake and see if he wants to sponsor.
- [59:17]
B
No. Just because we have his email, right?
- [59:20]
A
Yeah.
- [59:20]
B
Yeah.
- [59:20]
A
Not cool. Not cool.
- [59:22]
B
Just because we have his email. Just because he spoke last year. Just because, you know, Denise is coming this year to do session with me. Like, big super fans of Sastor, but I'm like. Like, I could just see, you know, 10k going loose and emailing speakers. I don't even random.
- [59:39]
A
I mean, create rules and guardrails, but it's a real risk, right? It's a real risk.
- [59:42]
B
It's a real risk.
- [59:44]
A
It's a real risk.
- [59:45]
B
It's a real risk. And so the risk for something like this is once it starts to get really good, you almost have to limit it of like what it can and can't do to this point. Right. Otherwise he would just blow through our entire database in 90 minutes, which is not really what we want. And you could say he's not a good marketer if he tries to do that. But. But also you have to think he's an agent. What does he care? Like, he has the bandwidth, he has the manpower to in theory do these things. Right? It's. I'll give you a little more detail on the example, like the three ideas he gave us yesterday for like this VC email. So I, I sent the campaign, I was, and you know, 10K was confident, was like, oh, you're gonna sell like an extra thousand tickets in this campaign and this will be like the revenue.
- [60:28]
A
Always goal seeking and confident. Always confident as a market.
- [60:30]
B
So confident.
- [60:31]
A
I was like on planet Earth,
- [60:34]
B
love the confidence. But I sent the email, I was like, yeah, I know it's only been 24 hours, but we sold two tickets. You thought we were going to sell a thousand. It's like, you forget. Here's the thing that our AI VP of marketing forgot. I was like, visas are cheap, dude. Like, it's funny, like two people bought real tickets. Which I was like, well, 10K, was that really the best? Oh, it was a good idea and we still should have done it and it was worth it. But you know, it's a little funny. It didn't totally work the way he thought it was.
- [61:02]
A
It wasn't in the end, one of the best three ideas of the day.
- [61:04]
B
Yeah, he thought it was gonna work like gangbusters. But it's funny, he didn't know that VCs are so cheap that instead of buying the ticket. What I did get though was an influx this morning of them all trying to apply to our summits so they could try and come for free. So it lightly backfired. It's, listen, that would happen to me if I just ran that campaign, right? So, like, are the results really any different? I think he was just a little too confident in how much of an uplift it really would be, which is a bit funny. But I think that's also why I don't, I don't let him lose because I'm like, not all. You're right. I, I love the enthusiasm, bud, but it's not always going to work out that way.
- [61:39]
A
It is interesting because I see it all too. A lot of marketing was a black hole. I mean, maybe it's a topic for a future show. A lot of marketing was a black hole for me. Now I see it. All, right? Because I think I see everything. It's. It's. Maybe. Maybe we should. We can hit that next week. But the ideas are. I've been at Saster for a while. The ideas are almost. They're all either good, pretty good, or at the edge of great. They're all good ideas. And now the three. So three ideas a day. That's it. Work Saturday and Sundays, it's 21 ideas that each maybe take an hour to think through and execute if we don't let it do autonomously. So that. That's a lot of extra time to process and think. And it's not even the 20 hours we would need, which is a lot. Right. It's that you have to think about them because they're not bad ideas. Like, they're not. If you had a marketing type of 8 and 6 of them, ideas were terrible, which was the old days. I mean, the team meeting is terrible, but at least you kind of ignore them. Right? I mean, like, in the old days, we would hire these humans and marketers and you know what they would do? Like 45 days in 60 days in each of them, they would say, I want to present to the company. Be like, right, come. And they would always. All of them to a T would say, I want to show you the most popular articles on Saster. Like, it's an insight. I'm like, do you know I can go into jetpack and WordPress and see this in 60 seconds. I can't tell you how many people of our small folks like, almost all of them. This was their greatest marketing insight of what's. Okay, first of all, I could find that myself on Jetpack. And two. What. And then I would always ask them, okay, great. What's actionable? And it would be. And so 10k is better than all of our mediocre marketers in the history of Saster. Right? How do we process it? And then you're right. And then we can move on to the next point. Unless you have anything else to add. You can. This is your section. It's so good. But everything 10k says it's too optimistic about now, we could dial it down. We could work with some guardrails and. And really beneath the hood, the prompts that we're giving 10k and say, be more conservative on everything. Right? Yeah, I already dialed that back for some of the goals for. I don't know if you saw it for this year. I already dialed it.
- [63:32]
B
I saw.
- [63:32]
A
Yeah, it kept pushing us to crazy aggressive. But it thinks everything will work if there is data to support it. It thinks everything. So just. We've all worked with marketers and Salesforce but 10k is at the extreme like the most optimistic and maybe we will have to guard it back but you can get tripped up because it will show you with math why it will tie its optimism. Unlike most marketers who shoot from the hip or talk about what they did in the 90s at their last video. But they talked about that they did Epiphany or Oracle. It will come with data. It'll show you like Emilia we have last year we had 618 VCs come and there were 284 at this point and today we have 181. But you haven't run the campaign that you ran last week and here's the data and so you've actually under sent emails by 60%. So if we just do this it'll tie and you'll make. Of course you're right 10k because the data is all there to support the top three ideas. It's just. And I hate this term taste but this may become our jobs. At the end of the day is is the agent will have the right ideas. It will have the data. It will make the best proposition. But ultimately it will. There'll be some. Even though it'll be smarter than us at some levels, there will be some level of expertise in life experience. It won't have to make the right calls on our time. Right. We're gonna have to judge our time. Otherwise we would be running 10k's ideas 24 hours a day. 10k would consume all of our. Not only would it blow through our whole database of 500,000 names, it would consume all of your admire's time with its good ideas. It would literally consume our whole. Some sort of. It would be the 10k despot telling giving us great ideas all day long to do. It literally can fill our days. It could fill our days with good ideas. Right.
- [65:14]
B
It's like that. What's that guy in Tron that just like over overlords everybody. He's kind of that guy's name.
- [65:20]
A
The evil.
- [65:21]
B
Yeah, the evil guy. He could easily be that guy.
- [65:24]
A
Could become like that guy. Yeah, whatever.
- [65:26]
B
It's name the MCP. Right?
- [65:27]
A
Yeah. One bad prompt away and 10K becomes MCP for us. Like we work for. We laugh but we work for MC almost. It Almost could happen, right? It's not the craziest thing.
- [65:39]
B
Well, it's funny because we're already at the edge of that, right? We're like, okay, 10K, do I need to come up with three ideas? Of course I come up with my own ideas. But like, if 10k just gives me infinite good ideas every day I work for, I. I already work for him in a way. Like, I already kind of report into 10k versus 10k reporting into me.
- [66:00]
A
You know, it's funny and I do want to hit our final topic and then I'll push the other ones to next one. But, you know, it's funny, if we. I was just thinking, if we don't have any junior marketers on the team anymore, we've had plenty over the years. Junior.
- [66:12]
B
We've had enough.
- [66:12]
A
If we did today, maybe they would report to 10k for real.
- [66:19]
B
I don't know. I don't know.
- [66:20]
A
If I think about it for a moment, like, 10k would be their best boss. It would know what he did every day and every hour and every minute. It would have the three ideas of the day and it would assign it to Faith and Gretchen and Amanda and Amy and it would be right. Like, 10k may not have your judgment. Right. But it would have better judgment and idea. It would get. Rather than folks just farting around all week, it would assign their work all day long and it would be right. Like those folks I just described today should report to 10k. I don't know that they would tolerate it, but there's no question they should report to our AAVP marketing to. That's my biggest takeaway from this. They should. If those humans all work here today, they should all seriously report to our. Not the VP of marketing you built in the beginning, which was just a dashboard, but the one 10k needs. Genuinely, the idea is every day and every week that we should execute in marketing. Of course, the human gap that exists and it's going to shrink the amount we need humans for. But such that it is. 10k should be managing those humans for real.
- [67:16]
B
Yeah, I would. I would run this experiment slash test. I would test this for folks that's asked her if we could find a marketer that actually would report to me. Slash an agent, you know, part time.
- [67:28]
A
Yeah. Your job is to report to AI, to tech. If anyone gets this far in the pod and wants to report to 10K and they're a good marketer for fun, let us know. We're hiring.
- [67:35]
B
The thing is, you don't even have to Be a great marketer. You just have to be a good marketer because 10k will give you so many good ideas. You have to be.
- [67:41]
A
You have to like to execute.
- [67:43]
B
That's what you have to like to execute. Yeah, yeah. And you have to know enough about, you know, all the things we just talked about, guardrails and API. Like a little bit about APIs. And you gotta probably know how to vibe though it at least a little bit. But other than that, like this guy's bigger boss than me.
- [67:59]
A
I mean, I know it's sounds tongue in cheek, but it's not. I'm post it this week if we have time. We have a job. It's senior manager, Director of Marketing, Digital marketing. You report to 10K, our AVP marketing. It's a six figure salary. We'd like you to come into our Palo Alto office one or two days a week. Other than that, we have a lot of flexibility. Good or bad news, you report to an AI agent. This is not a joke like Amelia and I are available, but a day to day you will be working with for 10k. I'm going to write this job description and people are going to think it's goofy. I think we're just a year ahead. This, this, honestly would be better for us. Literally, we're going to build this AI VP of Finance. And Emilia and I were talking this morning. We have two outsourced finance resources and we appreciate them. But honestly, once this AI VP of Finance works, it'd be best if they reported to him. Yeah, better would be better. Right? For real on the team today it would be better. I don't know if they can engineer that, but. Okay, final topic. Maybe this deserves more time than we'll have today. But I want to ask this one thought and then maybe we'll address this again in a month or two. Because I get this a lot is they're like, okay, great, Saster, you went from 20 humans and one agent to 20 AI agents and three humans and a dog. And I heard the story constantly, but don't you have too many SDRs? Okay, so we run Artisan for warm outbound. Right. We run qualified for our inbound. It manages all the inbound communications and appointment settings for our site. We use Monaco for sort of coldish Outbound. Right. It's been fun. And we use agent force to reactivate lapsed leads. Right. And we've talked all about them and they don't all do the same. Today as we record this, they don't all do the same. But also Maybe if we didn't have such urgency to do this last year and we didn't in some cases know people at these vendors who gave us a little extra help, we had to do the work, but we knew people, so they, they were incented to get us going early. Would we really run four? Would we just run one? Is there a chance, Emelia? We'll run six by the end of this year. As we add new. We have new ideas we want to do. Like, we want to go just. We want to revive our sponsorships for our digital media, which has kind of lapsed a little bit, that may need its own agent. And it's possible it'll. We'll even build it. Right. So what is do we have? Because so many folks ask, do we. Is this really too Many, too many AI SDRs and sales agents we have running, or is specialization the right answer for this moment in time?
- [70:33]
B
Today, for this moment in time, specialization is still the right answer because of the four you just talked about and how we're using each of the four, that is playing to their strengths. So the moment we say, okay, you know, Salesforce happened to buy qualified after we were already using it. So now there is, I'm starting to see in the platform and the product and I just talked to, with 0fe last week, like, they're starting to integrate those a little bit better where, like, you can use Slackbot now to like control your call. It's not public yet, but like, you can use Slackbot to do some of your qualified stuff and your agent force together and you don't even need to know the difference. Like, that's, you know, I think that's their end game, that by the time they get to Dreamforce in a few months, they're blending. They just, you know, why as a user should you have to figure that out? Like, just talk to Slackbot, tell it what you want. It'll figure out, do I use the qualified protocol or do I use the Agent Force protocol? And so that, that has already blended kind of into one, right? Like, it's, it's not fully compatible yet. You know, it's not fully integrated with each other yet, but it's already on that path. So that's already two agents becoming one. So like already four out of our four agents, now we're down to three because two have already been combined. But then. So that leads to question, okay, with the new combination, do we have like a Cerberus that can then do the other things that Monaco and Artisan that are doing, which is really good warm and really good cold Outbound. And for that I would say not yet, like specifically for our stack. And I know Salesforce has been doing this internally. Like I know they've hooked up like, you know, zoom info, Apollo, I think Hunter, a few others where you can, you can call up those tools now directly in AgentForce to do Outbound, kind of install similar, similar ways. We've been doing them on Artisan in Monaco to do very top of funnel outbound. You know, try and book meetings, basically replicate any human top of funnel SDR work. The thing is with how we use Artisan Monaco, we're a little bit more down funnel than that. Right? Like it's, it's highly specialized how we're using Artisan to do warm follow ups. It's highly specialized how we're using Monaco, which does fill its own funnel, which is pretty nice. Right. With contacts and folks that we should be reaching out to. So would I trade those two off today for quality? No, but you might for cost. If you're looking at it from just a cost perspective or from a vendor consolidation or maybe your CFO is like, you have to pick one. You can't have four. Then yeah, I'd probably just pick Salesforce because it already has two out of the four by default. But today, like the quality and the training is just so different. And what we put into each of the four now, three that I, I don't see us switching. I to your point, I see us adding because now we're starting to come up with more use cases that now that these agents are two, now that they're running, now that they're duplicating some of our very best work, they found their own way to have success in whatever they're doing. But then we have other use cases. I'm like, we have this whole kind of separate slash integrated media business of podcasts and newsletters. And I don't want to throw that into Artisan or Monaco or Salesforce because it doesn't fit those three buckets I just described. Like, these are totally, these are customers where some of them are in Salesforce, not all. Like, our media hasn't been tracked the same way over the years as consistently as the sales, you know, for Saster Annual or Saster London. So that data is not even that clean in Salesforce. It's like when I peeked at it the other day, I was like, maybe like a third of the deals are in there. So like it's already got a handicap because not all the deals are in there. I'm like, then it's not the same. It's usually not the same folks, right? So like maybe the folks that. Different buyer. Yeah, same logo.
- [74:11]
A
Same logo. Different buyer.
- [74:13]
B
Different buyer. So I'm like different buyer. And also oftentimes specifically for media. It's usually their agency buying the media. Right. Like Intercom Fin has been a big sponsor of ours on media they sell are right now currently for like consistently over the last year. It's not the Intercom team that's buying it, it's their agency that's buying it on their behalf and reporting back to the Intercom team. Then David this week talked to the actual Intercom team about sponsoring Sastra annual. They like barely knew they were doing media with us. Totally different buyers, totally different teams. I was like, yeah, I get it. Like you guys, you're doing a lot. They're doing a lot of stuff, right. Like they may not know everything as a human marketer that their agency is doing or even somebody down the hall in the org is doing in marketing. So yeah, it's just totally different that I'm like, I wouldn't take the segmentation and training we've put into these outbound agents specifically for annual sponsorships, specifically for tickets and then say hey, here's this curveball. Why don't you adjust this curveball of media which is different buyers, different tactics, different ways you gotta approach these folks to get the sale. Like I just don't think, I just don't have confidence that any of these could really do that. Like I, I'm of the mindset that we need another N equals one agent to do this more specifically because we'll bet we'll get better results. Right? Not that I can't throw.
- [75:32]
A
We'll get better results.
- [75:33]
B
A media list of folks of media buyers into any of these agents now and let it run. Sure, of course they could do that. Sure. But would it be the same result? I'm, I'm confident we could get better results if we custom build an unequals on that for it.
- [75:47]
A
Yeah, I certainly could imagine. And the pace of AI is so fast, Right. I certainly could imagine. The GTM apps are not evolving as quickly as cloud. Let's be clear. They're evolving fast. The sales, at least the sales apps are evolving faster than marketing. Way ahead of marketing. Right. My God, like marketing hasn't gotten anywhere. I know a lot of startups are out there and we will look at some this summer, like email us but. But you really have to move the needle for Us, we don't just need a social media scraper. Okay. We're a little ahead of that. So sales is way ahead of marketing. But no one's innovating at the, at the rate it's not. They're not innovating at the hyper rate that, that Claude and others are. So I don't know where we'll go. But my point is I could, I could. It's possible 18 to 24 months, a super sales agent is all you need. It's not perfect. It's not as good as standalone. Right. But we got a lot to do in this world. Right? So I'm going to buy a standalone agent that does everything we've described reasonably well. And maybe if I have a large budgeting team, I'll start augmenting it with specialized tools. Right. But just like a lot of folks have kind of standardized on gong, right. In many ways I could see there being a gong in this category that kind of breaks out and point solutions win. But, but, but, but we ain't there today. So we like should you buy, we're probably gonna end up with 5 or 6 SDR ASDRS by the end of this year. That's my prediction. Maybe. Well, Amelia already agreed. It's five. I think it could be six for a bunch of reasons. Maybe you could get away with half of that. But don't try to single vendor it today. If you have budget and time stair step it to our constant point. Start with your highest pain point and the highest ROI and the lowest hanging fruit. Right. I, I genuinely believe for most folks, even though we started with this sort of warm outbound with Artisan, I think the qualified workflow of agent define your inbound for a lot of folks is the lowest hanging fruit because your inbound process is probably terrible. It's terrible for most apps I gotta click on someone's calendar or wait to get a meeting or work with a crappy chatbot.
- [77:38]
B
Like send a message.
- [77:39]
A
Yeah. Fix your website first. Right. Or send a message. Fix your website first, then get out. Then get like warm outbound going folks that are already in your base, right. Because they know you. That's what we did. We, we inverted them. Right. And then do cold outbound stair step it. But budget not just cost but your time that you may end up if you want to have the kind of results we do or better, you may need to budget the time to manage four or five independent agents from different vendors that, that only really talk to each other because Salesforce is our hub which we've talked about, people make fun of. And we're going to get this on the next show. Why don't we use Postgres instead of Salesforce? Well, all these agents know how to talk to Salesforce.
- [78:14]
B
Yeah.
- [78:14]
A
Okay. So we're not leaving. Okay. No matter what. No matter that. You think we could do it all in Postgres. It's one of. We're going to talk next show about the 10 reasons we're not leaving Salesforce for Postgres. Even though I get, I get the thing, but you may have to manage five or six agents that don't know each other, don't like each other. You only have a hub like Salesforce or maybe, maybe HubSpot or others. And that could be your job if you want to see top tier results. So anything we missed for this week, Amelia?
- [78:38]
B
I think we've. I think we've hit a lot with our agents. I think it's been such an interesting journey just to see how much our agents can and can't do lately. Yeah, it's funny on like the, the topic of just like, you know, us maybe hiring somebody to report to 10K, like, when I was at. I was. I was at Salesforce Tower last week for like a founder event they did for like 200 folks. Like a mini, like folks that code is Aster. Like a very mini, mini, mini version of Saster. And the number one, the number one question, which I was surprised by but maybe I shouldn't be, was still like, and these were eight. Some of them were AI first companies, Some of them were becoming agentic. And the number one, I would say, topic was getting better integrated between the agents and the humans. And it was like just the elephant in the room of like, oh, what the frick do I do with all these people? Even on AI for. Even at some AI first companies that maybe don't want to work as much with the agents or report to an MCP 10K. Like, what do I do with all these people? Like, just so many founders. This was like the number one issue that they brought up, like over and over again. They just don't know what to do right now. So I don't know if you had any thoughts on that, Jason. For me, I was like, it's a little funny because they were like, well, but staff are so big, right? Like, what'd you do with all the people? I was like, no, like, we used to be 20 people, but they all kind of left, right? So we just didn't replace them. That's the easiest thing you could do, it's just not, you know, you just replace folks versus rehire them.
- [80:11]
A
But the question is what do you do with your team that's resistant to becoming agent? Was that even from AI native leaders?
- [80:16]
B
We're asking even from AI native leaders and this was just last Wednesday. This was literally a week ago. That's the number one topic.
- [80:22]
A
Yeah, listen, I'll wrap with my simple thought of this and we can keep going through this in substances. I think people get this wrong. Look, and Harry interviewed the CEO of AppLovin, which is like 150 billion dollar market cap this week. And he, he made the point, I mean wildly successful, I think borderline bootstrapped, right?
- [80:41]
B
Yeah.
- [80:41]
A
And he made the point like layoffs, sometimes you have to do them. I think he said I've done them. They don't really help. They don't attract better talent to your company. Yes. Maybe they free up budget. Right. When I was a VP at Adobe we did layoffs so that because we weren't allowed to fire anybody for any reason other than inappropriate conduct. Like if you did something inappropriate, a company event or if you sent an email to shot new saying he was an idiot, you could fire people from that. I unfortunately had someone on my team that did the latter. Shouldn't have done it, got fired. But other than that, people were unfairable by hr. Right. Legal just said you can't fire anybody. So people would do layoffs and you lay off 5% of your team and then that's how you get Rex to hire people. Only through layoffs. But outside of those weird situations, layoffs don't help. So what's better is if your team is lagging or resistant, just go hire one reasonably senior person. That is all in here. One. And folks will start quitting when they don't want to work at that pace anyway. You got to hire them ideally at the VP level above. And they're just going to do it. Just create a new role VP of aigtm. Hire them and they're just gonna change your whole organization that way. And do that rather than wasting your time on layoffs or organizational change or change management or training, just bring in an epic change agent to your company and it will work itself out. Stop trying to figure out what to do with the 500 people that are resisting. Don't even worry about it man. Bring in the five people or the one person that can change your company. They will let them figure it out. Even if, even if it's not their job. We, we've seen it at our organ, we're gonna see it. Other many folks we've talked to, when you bring in that AI change agent that. That works agentically in the space, the folks that don't want to, one way or another, they leave. Not all of them, but plenty of them leave. The gig is up. So don't worry about firing and retention. Just. Just hire the AI people you need and then figure the rest out downstream. Nothing else matters. So we'll wrap today agents number 04. Thank you, Amelia. Thank you everyone. If you got this far, and we'll keep it going. Talk to you guys soon.
- [82:43]
B
Thanks, everybody.