← The Official SaaStr Podcast: SaaS | Founders | Investors
The Official SaaStr Podcast: SaaS | Founders | Investors
SaaStr 842: The 90/10 Rule for AI Agents: What to Build vs Buy with SaaStr's CEO and CAIO
01:00:55
Loading summary
Transcript24 lines
- [00:01]
A
Welcome to the official Saster podcast where you can hear some of the best Saster speakers. This is where the cloud meets up. Today on the Saster podcast, you're either at the cutting edge of AI or you're getting displaced. And the question is just what's the speed? If all your customers are on five year contracts, you can hide if you want. There is no middle ground anymore. That's the best advice I can tell you. One last thing I would say to founders on this. I wrote this on Twitter yesterday and I'm a longer post on Saster. Ask yourself calmly use your own product. Is your jaw dropping? Is your jaw dropping? Almost every, if anyone that has even a couple million in revenue, up to billions at some point your product was a jaw dropper or you never would have gotten there. You, you utterly changed the lives of a stakeholder of a customer. They didn't buy you because you're 1% better. They bought you because you were you. You're jaw droppingly better. At least for some niche use case. Be honest. In the age of AI, are you as jaw dropping as cl? And if you're not, that's why you're decaying.
- [01:06]
B
Hey Caster, imagine having agents for every support tab. One that triages tickets, another that catches duplicates, one that spots turn risk. That'd be pretty amazing, right? Happy Fox just made it real with Autopilot. These pre built AI agents deploy in about 60 seconds and run for as low as 2 cents per successful action. All of it sits inside the Happy Fox omnichannel AI first support stack, chatbot, copilot and autopilot working as one. Check them out@happy fox.com Saster hey everybody.
- [01:39]
A
Saster Annual will be back May 2026. The world's largest SaaS and AI gathering for executives. Just as last May we hosted 10,000 attendees with 68 VP level and above attendees, 36% CEOs and founders and 25% were AI first professionals. It's the very best of S tier attendees and decision makers that come to SA Summit each and every year. But here's the reality folks. The longer you wait, the higher ticket prices get. They're cheap now. They're cheap, so just get them early. Lock in your spot today. Use my code Jason100 for exclusive savings. Get your tickets at podcast.saster annual.com or just use code Jason100 when you check out. See you there. Saster annual and AI summit 2026. It will rock.
- [02:25]
B
Hello everyone. Welcome back to another Zester AI Live. We wanted to do a follow up from AI Day we did a couple weeks ago. We did not make it through the whole thing and I actually think a lot has changed in those two weeks just with the wider, let's say release of cloud cowork to a lot more users including ourselves to use try and then obviously I know also cloud code got a lot of updates in the last two weeks as well as just like the model for that are core to Claude and OpenAI. So we wanted to do an update basically on where we're at just with the lens of our journey of like building versus buying all these apps for AI agents and now apps and kind of encompassing some of that with you know, everything that's been going on in the last two weeks because a lot has happened. All right. Just I always like to put this slide first for those who need it. And just to preface, as always, you can go to Sapsburg AI Agents if you want more info on any of the either third party tools or things that we buy. The apps we vibe coded today are all listed there. We have like a new community list, newer community listing that we have there as well that you can play around with and see what we use and how we use it. So that's that. Okay, so getting into our. So this is the 9010 rule Jason, that you originated. So love to get your thoughts here just to kind of kick us off. But basically our rule from the last, you know, nine months, 20 plus agents, six plus now, now maybe eight with the last. The two that I'll focus on today that we made recently in the last few weeks. So a bunch of apps we've made both for internal purposes and then actually now external facing. The rule we've had has always been, you know, buy 90% of what you need if you can, off the shelf, do it where, you know, there's already a solution for this. Like we're not trying to Vibe code our own CRM tool or anything crazy like that. You know, we have agents that we use for outbound and inbound. Those are all third party tools. Like I'm not even going to attempt to recreate those in what we're doing in bytecoding. So that's, that's an interesting mix there. So we've, we've pretty much bought 9% of what we've needed and only built the 10% where we either couldn't find a solution that really already existed pretty close to what we wanted off the shelf or we had, you know, in the case of our AI BPM That'll walk you guys through a bit more where we had a bunch of proprietary data that I had in one place. I basically had on my computer. We had like a bunch of years of data. I couldn't find a tool. And so that was also a reason where we went into the 10% of where it didn't exist. And then also more recently, I'll show you what we did with cloud cowork and vibe coding is I actually did replace a tool in the last week we were paying for. So this is kind of a new. I don't know if you would consider it a caveat to the roll book. This is a new one where, you know, we were paying for a tool. It didn't some of what we needed off the shelf, but it didn't do enough. And so that's where I decided it was worth the effort to build our own tool in this scenario. Yeah, we do use a lot of third party tools. I think in those evaluations it's typically been, you know, okay, are these vendors and tools we could trust Again, does it do like most of what you're looking to do? And if it does, like there's so many, I think nuances behind the scenes that are maybe unappreciated when you start vibe coding and learning how to do it. Really good things like compliance, things like security, data warehousing, like these are all things I don't, I don't want to tackle. Even as good as we've gotten of this whole thing, I don't want to tackle. The other thing I've already fallen into is the maintenance of all these apps. Right. Like Jason spends a lot of time in his day doing maintenance of these apps. I now have to spend a bunch of time with the apps I've created maintaining them and so it just kind of compounds on itself where then you're also now just spending time doing a bunch of app maintenance or agent maintenance. So it is a big time suck as well.
- [06:40]
A
I think there's a lot to dig into in here. The only sort of meta thing I'll add for folks is Emilio will show you. We rebuilt what I call an n=1 app. So the one I want to go deep on what the apps you guys might be used that we use but maybe we'll show you. We built this sponsor application for our sponsors. We have 150 sponsors for Saster A and 2026 and may come and what is interesting at a higher level is there is third party application for these what is called sponsor portal. It's a niche category of software. It's not big enough to justify significant investment and the products are not, they just have not kept up with 2026. They're terrible. So we didn't want to build one. We built our own version of something we could buy for 5k and 10k and have for years and we'll show you why. But I think there is a risk for folks building vertical apps, building niche apps. You should not build an agent if you can buy it or but if you fall behind too far, folks may build it out of frustration and that's more what we did rather than try to cancel our Salesforce bill. But this was one where we were driven to frustration by the mediocrity in this event software space to finally build our own. And, and it works. So it would be that that's a lost customer forever for those vendors. So we'll show you that though. Keep going.
- [08:03]
B
Yeah, I'll show folks this is my evolution of the 9010 rule of, you know this again we've, we've bought a lot of third party tools are on SAS AI slash agents. You can see the ones we use multiple for outbound, we use one for inbound. But like when we, when we go to evaluate things right. Again, if we can find a solution off this shelf, we're going to go for it, right? Like you, we are famously three humans, one dog and like 20 something agents now and like eight apps. So like we don't have a lot human manpower to keep up with these to go through sock to compliance to, you know, use a data warehouse other than Salesforce. Like a lot of our apps and agents all push back into Salesforce. It's a great system of record. I'm not trying to vibe code, you know, an inbound agent. Like I'm not trying to vibe code another outbound sdr. Like there are things off the shelf that do that really well. It's not a good use of your time. I think that's the other thing too. Like just because it's feasible, sometimes it may not be the best thing to buy code even though you could maybe do it because it's just going to be an ongoing time suck. Like I really had a way in the pros and cons of Vibe coding our sponsor portal because we had something off the shelf that worked. But I just got really frustrated that it was lacking a bunch of features that I knew would be possible to vibe code into an app and like make it better. I was like, you know what this tool that we're using off the show has no AI, literally zero AI. And this is not a knock on them. Like, this is maybe just a category of like event software has been unjust for a very long time. So, you know, hosting SAS 3 annual doing these events has been always a sore point of frustration that we're kind of hostage to a lot of vendors we end up having to use because there isn't anything better out there. But I knew it would be feasible. Um, but I was not keen on doing it because I knew it would take time. I knew to build in some like AI native features is like not my strong suit when it comes to Vibe coding. And the other biggest thing was I did not want to tackle single sign on. I was like, that is complicated. Like I am not an engineer. I, you know, have been a go to market person. I don't know how to fix that. If I was like, what if that breaks? Like I'm screwed. Like, I basically actually ended up giving myself a limit of like, okay, if I can't get single sign on to work in a Vibe coded app in a day, then I'm just going to use the off the shelf tool because I can't spend that much more time on it. Like I will give myself a day, see if I can do it, see if I can get the integrations to work, see if it'll work for real and then I'll do all the other features because all the other features are like fairly simple. And I was like, if that worked then I'll launch it, but if not, I'll go back to what we were using because it's just too risky. I was like, this is super risky. We have 150 sponsors, millions of dollars of revenue and customers at Saster that rely on this portal and this backend to work and get the information they need to come be a successful customer of ours and sponsor at Zaster. And I was like, there's just a lot of writing on this. So it was definitely scary. And I would say the thing that kind of tipped me over the edge was one we're nine months in, right? So I wouldn't say if you just started bypoding or keeping kicking yourself on. Okay, I need to start learning how to ICOD app so I can see what's possible. I would not tackle this day one right? This was after we've done, you know, the valuation pitch calculator, the pitch deck grader. We have, you know, Sasser AI is on a Vibe coded websites aster. Like this is after multiple vibe coded websites, forward facing in production before I try to even fathom tackling something like single sign on. Because I knew, I knew from those experiences that it was going to be complicated af, I was like, it's going to suck. I already know it's going to suck. So I feel like that was a lot of hard. I was like, okay, because it's going to suck. I know it's going to suck. I'm going to give myself a day. But I also think it ties back to, you know, what are, I think there's all this social media hype around. Okay, like what are people, you know, obviously loveable, replit on fire. Versailles, like all, all of these tools are on fire. They're all going to be a thousand. Like all these, all these guys are, you know, obviously benefit and brand and I do think there's so many great stories coming out of people who have buy coded things or and are now profitable from apps that are in production. But if you actually look at Anton's tweet that Jason wrote up from, I think this was in the last week like they posted what are most people vibe coding. And I feel like this is super interesting because it really applies to the 9010 role of like if you look at this, it's not till number four that he got to replacing, you know, and or rebuilding a custom built SaaS app the way that we did. Literally number four, number one is rapid prototyping and that makes a ton of sense, right? Like you don't want to wait for your engineer tape. Maybe you're on the go to market team so you're not as technical but you're like I just need you to fix this on the website or we have a new product or we have a new case study and I this is how I want it. Like it's perfect for that. Like you don't have to get all in the weeds. Maybe your engineering team will take it and actually make it real and forward pacing. But like that's a great use case. That's a great, I think also first use case where a lot of people come in to vibe coding and I was like okay, that makes a ton of sense. So it's just interesting to see that reflected in his data, what people are actually using. And by putting apps like global before I think to building internal tools, that's what we did with our AI bpm which I'll show you guys next. Again, that made a ton of sense for us because we knew it was going to be internal facing right? Low risk. We couldn't find an off the shelf tool to do what we wanted in marketing. Like too many of the marketing tools right now are either like actually sales tools that have a marketing wrapper around them or are just glorified copywriting tools and that's not what we wanted. So that's where we fell into number two and built that tool. Number three is something we do a lot of. We actually do it a little bit more automated but you can basically API into making slide decks real, right? So if you have something like you know, everyone's call recorders here, you could take a call recording, put that into, you know, pick your poison tool, Zapier, addita, whatever, API it into replit, lovable gamma, have it spit out a custom website for the prospect you just talked to versus, you know, waiting to send a generic deck to them. We do this all the time now. It's actually a really again really great use case for go to market. Really low risk because you're just going to send it to that prospect, right? Like you know, if the whole, if they share it around the org, great. But it's not for like the whole Internet to see. It's fairly, you can password protect it right. Fairly easily. It's pretty secure. Like there's probably nothing proprietary you're going to put in that deck on that website. Again, low risk. So it makes a ton of sense. So I think just interesting to see here that it's not until it's also number four where he was like then we see people starting to recreate simple and I appreciate he said simple. It's like not godly, you know, simple thing like the ones I'll show you that we build super great projects. But not again we're not trying to, we're not trying to rebuild Salesforce. I'm not trying to recreate our marketing automation platform. I'm not even trying to recreate the AI bought off the shelf. Like those all do that really well and there's, there's just no need to build those things when you can feasibly buy them. So I'll show you some of the two things we've built more recently and also why it's why it's interesting that we've built these recently in the last two. Well 10k is like a month old and the sponsor portal is literally we code and I honestly. And part of that build versus buy. Yeah, two very different reasons here. And also I think part of my hesitation to rebuild something like SaaS responsors would have been greater had it not been for Claude. Cowork coming out, right? So it came out, I think, January and beta, and you have to be on a max plan. So my max plan got it, like, fed one or something. So I didn't have it for that long, but I was like, okay. Between that being literally on my machine and then Claude also having Cowork in browser, I was like, oh, like, it can do a lot, actually. They can. It can see everything it can do. It can take action for you. So I was like, you know what if I get stuck Vibe coding this DAS app that I need for sponsors? Because I know it's a little bit more complicated with, like, single sign on, and, like, it has some unique data that's like, unique to each customer, let's say. And I was like, obviously, single final will pack a lot of that, but if I get stuck, I felt confident having used Cowork for a couple days to be like, okay, I think Claude could probably bail me out if I get stuck. And the other part of using Cowork so much was that because it's on my machine and I was recreating an app we already used, it was easier to do it because it could already ingest all that stuff because again, it's on my machine, it was in my browser. And so it basically took all that good stuff, which I'll show you, and helped me write a spec for Vibe Coning, and then also helped me when I got stuck a couple times with single sign on to get unstuck. So, like, I don't. I don't know if without it, I could have done the whole project and app to completion. I'm going to be honest. Like, I don't know if I could have because it just one. It kept me to my. My, you know, restriction I put on myself of, like, okay, if I can do this in a day, then I'll do it. But if I can, I'm going to go back to the third party off the shelf tool we built, which I think is a good restriction. And it was a whole day. I was like, I have no meetings this day. I'm literally just going to figure out. I was like, don't talk to me. Don't slack me. I'm too excited trying to figure it out. I'm like, in a dark hole. And so I think there's two very different reasons here. Part of it, to. To an earlier point you made Jason, was this tool in particular for SAPS, for sponsor that I decided to recreate with livecoding had no AI features in it. I was like, if February of 2026 and there is not one single AI, not even a lightweight AI feature, nothing. Not like, hey, put the sponsor's name in, or you put your customer's name in and then we'll use AI to backfill their address and their location and how many employees they have and their company size and where are they based. Give you a little blurb about that. And I was like, that's, that's AI 101. That's so basic for an agent to go into. Like, how is that not even in the app that we've been using for the past couple of years? And so I was super frustrated that there was literally zero AI in the tool we were using. And I thought, you know what, I think it's just time to rip the band aid and recreate it ourselves because I know for a fact I could get AI not only into this tool, but it will make life easier if the single sign on works. And so I was like, you know what? There is no. And to be fair, I also did look for a different third party tool that maybe had more AI in it that would do this, like unique sponsor, but it's fairly unique. Like this is again, event SaaS apps are fairly behind and all this AI stuff, like there was not one I could find off the shelf to do this. Similar to the way that we built our AI VP of marketing, which is internal pacing only. So I'll show you guys some screenshots of it and then I'll show you SaaS for sponsors, which is external pacing. Again, same thing. We could not find a third party tool more specifically for marketing. Obviously there's a lot of sales tools out there that are really great that we're using and sometimes I do use them for marketing, but I couldn't find one that was all encompassing, that was like, okay, I'm going to take all your data from the last 10 years. Like, this is something we also couldn't hire a human to do because asking a new marketer to look at 10 years of data, of, hey, what's worked for Saskar, for sponsors, for tickets, for customers, for, for media. We have worked the whole media business, like, would just, they would blow their brain. Like, it's too much, like it was just too much complexity. They'd be like, like, I don't, I don't, I don't know how to process all this as a human being. And so I was like, okay, clearly AI can easily actually process all this. Data which it did like it again, it probably took a day because I followed this rule quite strictly. I gave it a day to like ingest all of our old data, come up with like analysis of it, come up with what's working, what's not working, what are we not doing, what are, you know, do some competitive analysis. What are other folks like us saps are doing? What's working in AI? Like what are the benefits I can get out of inject like putting and ingest. Injecting more AI into this process and then I let the AI run with it. Right. I was like, okay, come up with an actual plan. And so that's what we did for these again. Two vibe coded tools. I think two very different reasons of like one was internal facing. So I think to Anton's point, like, that's a great reason to buy code something if it's just going to be internal facing. The other one was again to replace a simple SaaS app, but also it had zero AI functionality. So I was like, there's just no benefit to sticking with this tool. I'll also say maybe to take a different angle on it. If that's you right now, that's where I would be worried. Like if you feel like your app has zero AI, that was literally my first frontier to try and replacing something. And I was like my bar is does this tool already have AI in it? And it didn't. So I think if that's you, I'd be a little bit more worried.
- [21:38]
A
Build out something in AI in your app that moves. That really moves the needle. Yeah, you should give up you. The LLMs. The LLMs are available to everybody. Everything got really good around last summer fall with Claude 3537 and they're really good now. And it's not that hard. I mean here, listen, I don't want to spend all this time. I want to show folks what we build and answer questions that we didn't get. But honestly, here's the tough part guys. This, the. The. The vpm, let's get to it because we're. We're. But the V. The VPN marketing is niche to us. The sponsor portal if Emilia with two days of hard work or day and a half is able to build something better than it's on the market. You should be embarrassed if you have. If you can't do that. Right? So this is. I think it's time to be honest and ask yourself is your. Does your app still make your customers jaw drop? Is it still magical? Because. Because there's there's really no excuse today and you're going to be quickly disrupted and churned if you can't build something magical with your AI app today. Stuff so good. So. But let's keep going.
- [22:41]
B
Yeah. It's so good. It's. Yeah. To your point, there are some good questions. So I'll walk you guys through both of our more recent custom built agents and I think there's some good nuances here. Like, you know, could I build these two apps this time last year? Probably not. Like, everything got, not only have things gotten really good over the last, you know, more so nine months than the last year, but in the last two weeks with the new models and obviously with the work, you could do a lot more, lot faster. And I think that's a big unlock that people haven't quite fully realized yet. I know I see a lot of folks who have realized this about quad code and, you know, GPT coding and all that jazz, but I think just having cowork sit in your machine, sit in your browser, see what you're doing, be able to take actions. It's a, it's. Honestly, it was the first time I was like, oh, I'm a little scared for myself. Oh God. As the, as the, as the agent builder and as the person who has managed all these agents for the last nine months, I was like, well, you know, give them a year. They may not need me, they may not need me in the air. So we will say maybe we're more than a year off from that. But I was like this, it's gotten scary good crazy fast in the last two weeks. I'm like, it's only February 11th. Like, it's gotten really good. So going first to our custom AI bpm. So again, we couldn't find a tool that would do this. I did. The real problem here was finding a tool too that could not only adjust all of our proprietary data from the last 10 years, which we needed AI to do. We knew that AI would be able to, like an agent would be able to understand all the data, would be able to come up with an action plan, like an actual coherent plan of okay, here's what to do every single day from. And I did it for a short time. I was like, let's do it for the next six months, right? It was January now it was after. Let's do a six month marketing plan. And you're going to output from me first at a high level, right? So this was an iteration. It was all based on data of what's working what has it for, you know, our sales, for our marketing, for our go to market team to really push us and keep us on track. I was like, I need something that will keep us honest and is also other than the data doesn't have bias. Right. The agent is just going to look at everything across the board. I give it, it has real time API inputs of where we are today and it'll, it'll adjust in real time. It'll say like today it was kind of roasting me for being behind on our summits. I was like, I know, I'm trying again. You're faster than me agents. And so it's just really interesting. So we custom built the AI VPN more specifically using, you know, third party tools we already had. Because I basically wanted to take the data of what's working the best and then put it into this new agent. I took a lot of, you know, historical data, I put it into this agent. I used Zapier because obviously that powers a lot of other workflows to push back and forth and pull back and forth. So it's got web hooks, it's got APIs, it uses Salesforce as you know, our primary data instance. Again, I'm not trying to recreate the wheel on that one. It's a really great way to have data that's plus Salesforce already has a lot of data, right? It's got 10 years of chapter data. So I'm not going to try and recreate that. I'm just going to push and pull from Salesforce because it's already there, you know, push everything back in the cloud. And then we use Replit for this one to build an internal agent. You can use Replit or Lovable to build your internal agent. So we built 10K for two very specific reasons. One, because we did more of a marketing orchestration tool and we couldn't find one. But two, also because you know, we wanted something that would help us get to like the Keepathon. It's got us to the first 10k attendees for success rate annual this year and help us drive the first 10k of revenue. And it's giving us like literally actionable to do. It's like every day it's the first agent now I log into other than like our outbound and inbound agents. But this one is typically first because it keeps us honest. And I'm okay. Here are the campaigns you need to be running this week. Here's the copy for the campaigns, here's the email copy, here's the ad copy. Here's how much you should be spending on ads. When we look at where you're at today, like every morning, like agent, look at where we're at today. It'll say, okay, here's where you're at towards those attendance and revenue goals and sponsorship goals. Here's where we should, you know, adjust those levels and dials maybe a little bit more. Have you tried this thing? Or sometimes we'll be like, hey, I'm thinking of doing this. Which is not our game plan. Is it a waste of time or not? Like, it's a really good copilot for me to bounce ideas off of. So that's our AI vpm. It's more so internal facing, but I'll go back to what it looked like for a second. So this left hand rail is what our AI agent looks like. Again, it's got a lot of proprietary data. So I took some screenshots that were more high level. But this is, you know, it's like, okay, here's the setup week, here's the launch week. Here's what, here's how many tickets you need to tweak. When you click then, then these are also further clickable. It'll say, okay, here's what you need to do this week, right? It's telling me like, hey, create this promo code. Make sure Jason Jason 20 is on. Which it is. Make sure you're running ads to it. You know, make sure all of your AI SDRs are optimized for sponsorships and tickets. Like, here are the campaigns, here are the target audience you need to hit. Here's what you should promote to them. Like it's literally giving us all these things which has been a super handy day to day tool. And it's been again, just a nice copilot to have. Kind of off the shelf and glad to do it again. It's high stakes in the fact that like we rely on this now, but low stakes from the fact that like only our turtles need to access it. We had to hook up some APIs to it to have real time data, but other than that, you know, it's not trying to solve anything crazy. And so for all types of purposes, It's a simple SaaS app or copilot or agent or whatever you want to call it, but it works pretty well. Okay. Sasktar Sponsors is our newest one. This is the one that I use Cowork a lot more for if you, you can actually go to sassersponsors.com so this is external facing. If you're Logged out, which I'm logged in. You'll still see a bunch of the stuff. You just, if you want to take a look at the website, just to see how it works, that you can go to it. We decided we would just leave it open and then we would only gate. You would only need to sign in for things. Like if you wanted to actually submit a task, you'd have to sign in to pick your booth. You have to stop. Okay. You can't take your competitor's booth to be really far from you. It's got some gated features there. But for all intents and purposes we were like, you know, a lot of our sponsors have either third party agencies or other like contractors they work with when it comes to something like Sasser. So we're going to keep it open so that anybody can access most of the information at any given time just to see it at a glance. But if they need to take an action, that's when it'll prompt them to sign in. Interestingly, that is something that I wrote into the spec with Claude and I'll actually show you that and then I'll come back to the website. So as I've already said, we were paying for an existing tool here but I'll open up my blood chat. So this is the whole chat, I linked it in the slides. So if you want to read my whole chat with Claude on specking out the sponsor portal that I built, go for it. I actually think it's a super important step if you are going to vibe code. Your own simple SaaS app is to write the spec first and actually that was a pro tip from Jason is to write the spec first so that it's kind of like bounded, it knows its parameters, it knows what it's going to do and it's not going to go crazy on like trying to self validate and do like a bunch of crazy features that maybe would break in that you don't need. But also in the new spec make sure you have, you know, basic AI features and tooling. Make sure that anything I put in is going to be able to have that functionality. You know, like let's benefit from AI, like let's build that into the platform where the other one didn't have it and then certainly you can see it's here first. I was like most importantly and needs to have single sign on. So whatever you write for the spec needs to keep that in mind that like it's going to be a single sign on tool. I literally said to Claude, you know I want to keep it open since our, since our customers have people that are going to work on their work on their event sponsorships with like their agencies. That's super common, right? These are really big companies that we with sometimes some are startups and so sometimes it's also like the CEO or like their, their chief of staff is on it. And so I wrote that onto the spec. You can read it all there, wonder board later. But I think it's a good format to always write the spec first to be able to do and then like, you know, have a conversation with it of like, okay, does this look good? Mine looked pretty good. B1. So I was like, okay, this looks like what I want. And so that was the first thing I did before I went into. In this case, these screenshots are from replit. So then I took what I wrote with plot as a set and I put it into replit and I, you can see when I very first started, I said, how can I add true single sign on? Not like faux single sign on. Sometimes if I'm coding apps, like it'll do stuff that's not real, like true signal sign on, make it actually real. And I wanted to do this because as simple as it sounds, actually our old third party tool we were using for this didn't have persistent single sign on. So like, if I sign on, if I sign into our previous application of this and I, you know, submitted a bunch of stuff, disaster. And then Jason logged in, let's say, to the same instance. He wouldn't see what I did. For some reason it wasn't persistent. So that part had always kind of annoyed me. And so I was like, oh, I want to do true single sign on across the org so that, you know, we only gate things that we need to. Anybody from our agency can do it, but also, you know, make it smart enough where like if they join and it's their company, domain Org, that anybody can join. Because the other annoying thing we had to do with our old platform is like me or Wendy who's on our production team or David would literally have to invite somebody anytime they needed to add somebody to their instance. And I was like, why? This is the age of AI. Like, can't you just look up their domain, see if they're a sponsor and then like let them in the form? And the answer was yes, yes, it could do that. And so that's where again, I used a mix of cowork to help me think through this as well as the replit agent that is just in the agent to say, okay, what can I use to tackle single sign on? They have a, it's not a native integration yet, but I think it's coming. They use a tool called Clerk to tackle single sign on. And I was like, okay, let's do it. So I had to sign up for Clerk and then my chat's really long after this. But this is how I go through building true Single sign on and some light AI features into this app that we were previously using. So yeah, this is what it came out with. I think part of where I used Cowork a lot too, for this tool specifically. I was also like, okay, you've now ingested and written a spec for the new portal. Great, I already gave that to my Vibe coding app. Now take all the information from this year and put it into this, like, help me rewrite it so that I can give it to the replay agent to put into this portal so I don't have to copy paste a million freaking times. Like, just do it for me. And then the other thing I did was actually made this a lot smoother and a lot faster. I just gave Pond Cowork because it's on my machine. I already had all the contracts right of all of our current sponsors, all of our current customers for Samsung annual this year. And I said, work in this folder, go through each contract and like, this is what you're going to need to do to create this like, app. And I, I, I told Claude specifically, write it in a way that a Vibe coding app will understand it because I'm going to give it to the Vibe coding agent. Like I'm basically going to output with you as an agent and give it to another agent to deploy. And it said, okay, I'm ready. It worked in my, you know, on my machine, in this folder. It went through all the, it went through all the contracts, et cetera. And it, and it came out and I told it to look for a few specific things. I was like, why? You know, make sure you're capturing the company, their URL, et cetera, to make sure you read each contract to see how many sponsor passes they have, because this is important for registration. And that's like one of the first things people do when they log into the sponsor portal. I like to see how many customer passes they have, see if they have any speaking stuff, and then also tell me what day they sign so that I know what order that they'll be in for boost selection, because that's what we use for boost selection. It did that in I want to say it took like an hour because there's a lot of. There's a lot of contracts they have to read. They're a little complicated. And I checked it, so maybe a little less than an hour, but let's just say an hour. It took Claude go work an hour to do that. And then I told Claude, okay, great. Like, all the data looks good. Now go into our event registration platform. Because I had it open on my browser, I was like, just click into it and start creating the codes. It started making the codes on its own again. Probably took another, like, whatever, 20 minutes to do that. It went through. Everybody made their codes. I was like, great. Now make all of their customer nomination pass codes and all their promo codes and all that. Everything I just did there with, like, cloud cowork in a couple hours, we used to do with a human. We'd be like, hey, check the sponsor list. Check our contract. Like, go into each contract one by one, read it yourself, put it into Vinsabo, tell them how many tickets they have, create the promo code and then email it to them. Because back in the day, our old tool we were using couldn't ingest unique codes like this. So when I click this one, it goes to, like a semester code. When you know a different sponsor clicks theirs, it goes to their sponsor code. I'm not going to show you their code so that you don't use them, but it goes to their unique registration code for them. And so I was like, that literally used to take us a day. And Clyde just wiped it out. And, you know, he knocked it out in like an hour or two. Like, what a champ. It was super interesting, right? I was like, I know this is something that I could feasibly do a little bit better. It also, like, made this entire timeline of what we do. It made our campus map more interactive so that when sponsors come in here for the first time and get a lay of the lamb, they know exactly like, okay, what are the. What, What. What do these. Like, if you're a first timer, it's probably a little hard. So I'm gonna add all these widgets so that you could see what they are. Pretty cool. And so there's a lot of other stuff in there. Like I said, you can feel free to go tosash for sponsors dot com. That's what we built. May not be relevant to you. If you're coming to Saster, it's still got, like, the days and times on there. I think it's just interesting to click through it and see what is feasible and possible as a now forward facing app. I'll say my biggest fear I had when I was I like tested, I would say it took me like an extra half day on SaaS responsors to deploy just because I was testing everything so much. I was like, I have not done again, I'm not an engineer, I've not done single sign on ever before in my life. The first time I'm doing it so I want to make sure it works. I would like sign in as like different emails. I'd be like, okay, let me try to submit something there and do something there. I was like, I don't know, it worked. Like I kept waiting for it to break a little bit more. I don't know. I was like, this crinkle sign on is working. Like, could it be that we've built a better ride code and tool than what we were using previously and this is the first iteration of it, right? Like it's February, there's still some time to Saster so we're going to build a few more things into the Saster sponsors. But it was again, I don't know if I could have done it without the cowork element involved. Maybe it would have taken me a lot longer. I also got like a little bit stuck at one point with single sign on and it was like, try to tell replit to do this. Yeah, hold it. And then it worked. So it did help a lot as like my buddy for me to get this all done. But what's the catch, right? And then we'll do Q and A. So I think I mentioned it at the top but you know, biggest catch here is again, if you're just starting to explore rib coding, if you're in one of those three buckets of you haven't really built out an app or agent yet. That is to this level of sophistication. I wouldn't feel embarrassed in any way by that. Like this was a fairly. It's still simple on the surface, but was still a little bit more complicated to deploy. And so I don't think I could have done it, you know, nine months ago with only having built other apps into production. Different websites, different things that are now out there in the universe that I knew, I knew feasibly it could be done if I just solved this single sign on thing. And so that was the biggest risk. And so I spent a little bit more time on that. That was the majority of my time is making sure that would work and then programming it and then doing A bunch of extra testing because I was like, I don't want to send this to 100 sponsors. And then I get a hundred. I was like, my worst nightmare is if I build this thing, I send to a hundred people and they're all like, amelia, I can't log in or I can't do this, or my registration code is wrong, or it told me I was a platinum and really I'm a gold. Like, I had so much fear that it was going to get some of this wrong. Then I spent extra, extra, extra time testing it and I was like, I don't know, man, I've tested it all. I feel like it's working. Like it works. And so then I sent it to like 12 sponsors. I was like, let me send it to 12 people, see if it works. And then they were submitting. There weren't any issues. I could see what they were submitting because I also built, I also told the app to build me a what I call a super admin back end so I can see anytime somebody submits anything, which is super cool, which the other tool didn't have. And so this is really, again, it was a scary situation, but one where I was like, okay, I've tested enough. I just have to submit to the first 12 people, see if anything breaks and then go from there. But it was honestly my worst nightmare to be like, if this doesn't work, I'm just going to. I was like, you know what, I'm just going to go back to the other tools. It doesn't work because I can't deal with like a thousand emails that they can't get into the park because the benefits will not outweigh the cost. But yeah, the other thing is like, I do keep futzing with it, which I need to not. Like there are other things I want to build into this borderland where I'm like, oh, what if we did this and what if we did that and what if we like now that all the basics work and they're working fine and they work well. I'm like, there's all these other things I want to do now, but I actually need to stop messing with it. It's hard to do that, I think to some degree because it's also a little bit detrimental to my time. I'm like, I just need to. Again, that's not something I would be doing if I just bought an off the shelf tool. Like some of them are how they are, right? Like you can do some additional agent training and you can do new Campaigns or whatever. But you're not futzing with the actual like app and the backend itself. So I was like, I'm just gonna stop because I have to. So all of that to say can you really buy code and apps into production? You can, but obviously with a lot of asterisks and daggers there, it should be something simple. It should be something where you know, you have something either proprietary or stuff like this where you can't find an off the shelf solution. Again, if you boil down the things we've done with it, which are, you know, websites, valuation calculators, now this test responsors one which is basically a project management tool. If you think about it, um, it not super compl. Like they are complex but they're not complicated in the sense that like feasibly, yes, you can do these things again versus like trying to buy code your own. I'm not going to try and buy code my own CRM and like recreate 10 years of data touchpoints and who opened this email and when, like that's too complicated. I'm not going to. So it's just not worth the trade offs of, you know, buying Salesforce versus building something like this. I think too why we've bought versus built a lot of things we use is a lot of these third party tools, at least today are, you know, they're compliant, whatever you need, some, some of you will need LightSoc 2 or whatever or ISO, whatever, they're secure, right? Your data is safe. Like this is. Again, neither of these two tools I've built recently are storing like super proprietary data. So to some degree there's a, there's probably a little bit of risk there but it's worth the trade off because it's not the anything is not the end of the world there. Again it's not like table stakes and typically too these apps that we've bought and have, you know, we, we have an FD at each of these and that's not unique to us. That's just, you know, how these products work in this day and age. Most folks will pair you with an FD to get it going and to help you run it and deploy it. Which again for a time perspective has a lot of benefits too. Somebody who knows how to do this fully. I'm going to let them do it. Like I'm going to spend you know, once a week with them. Like I said, once a week with like qualified artists and agent force. But I'm like, hey, it's you know, 15 minutes here are things I'm kind of looking for an agent to do. They go off and do it and then we come back. Like, I'm not spending the rest of my week trying to fuss with these agents all day long because they can't. There's other things to do.
- [43:31]
A
1.1Captain Obvious point though. 2Captain all the stuff, obviously everything. The quality of these apps you can create yourself is accelerating. If you, if you haven't done it or don't get it now, actually now's the best time to start five Coding. Right. If you have to. Right. The next best time is probably next week because the pace of acceleration is so great. Stating the obvious, I asked Claude what the value of Amelia's time was and Claude said, depending on how you calculate it, between 1,000 and $2,000 an hour. That's her time. Now that we're so such a lean team with so many agents of humans. So there's no economic justification for software cost to building these apps. Right? You can buy them for ten grand. Like it makes no economic sense. So you have to really need something that you can't buy today. It's, it's almost beyond the 9020 rule. If your time is, is if you're, if you're, if you're working in, in Tulum or Bali and you're just futzing around, I don't know, your time might be worth $5 an hour. I don't know. But this is something we had to build for someone built that's worth $1,000 to $2,000 an hour. You have to be careful the boundary time. You have to be careful the not fuds. But, but, but that's, that's the bar for us to build something versus buying something. It is not economic. So anyhow, that's what Claude said. But it is worth it because these are high impact things where we get extremely high value out of an AI VP of marketing and a sponsor portal that will interact and manage 10 million of revenue and make that process infinitely better than it's been the last 10 years. Like infinitely better with AI. So those are two thoughts. But we can go, go to questions from there.
- [45:10]
B
Yeah, it's fine. Like somebody asked me yesterday, but like, hey, would you white label the SATs responsors? I'm like just.
- [45:17]
A
No, I think it's a good, I think it's, it is. But, but then you have to support all the customers. You have to do all the portion here. The future, the future of vibe coding, actually. And we've worked closely with The Replit team on this. It is early is now that you can really build these things for real. And when we started on Replit last summer and it was true of Lovable and Bolt and others, it was very hard to get something into production. If you've watched my journey now, if you, if you put in the time and you have some familiarity with how to build software, you can build most apps that you want to, it's radically better. So the next frontier is maintaining them and all the stuff. If you talk about Vibe coding, it's about building, getting going and we're going to have a whole replit Vibe code at Saster AI 2026 in May 12 through 14 and we'll help you build apps and come. It'll be fun. You can build your own B2B app together. It will be very unique and we're going to promote that all and we're going to have a whole day called Saster deploy on the 12th just about deploying agents. We're going to teach you how to do all of it and get you into production. But now that you can actually build these app, who's going to maintain them for the next five years? So white labeling sounds fun but for us if they couldn't make millions from it, it wouldn't be worth doing the customer facing side. And so let's get to more questions. But the interesting frontier not to get distracted is who will maintain these apps. And now that we have 10 or 20 agents running and probably 10 apps we've built ourselves, it's us. Right. And so the maintenance goes more and more to the top. So issues to work on in 2026 but very interesting.
- [46:44]
B
Yeah, I do hope one of my agents probably thought I can get good enough to manage the others a little bit more.
- [46:51]
A
It's a dream right now no matter what.
- [46:52]
B
Actually right now it doesn't. You know, sometimes it's it really the first question that we have. Sometimes he times out like even on my Mac I'm like what? Like I'm on like the beefiest MacBook Pro you can have. Like maybe I need to move to a Mac Mini in a closet. But sometimes I'll time out on my machine. I'm like, dude, yeah, yeah.
- [47:14]
A
But the app does. But these apps don't time out.
- [47:16]
B
Yeah, yeah. So it's just. Okay. The first question from Kelly is what percentage of SaaS companies are you seeing now get hit with the current market conditions? That's probably a better question for you, Jason, are you able to share anything about your Conversations with folks in the Saster community on what makes them successful to avoid some of the kind of market reaction lately.
- [47:39]
A
Well, look, the market reaction is brutal. I mean Shopify had an incredible quarter they just announced while we are doing this growing 30% at I think 12 billion in revenue. And it's the, you know, the market did, it's down 10.4% today. It's brutal out there. And what can you do? Well, first of all, on 20 VC with Harry and Rory, we have Michael Cannon Brooks, the CEO of Atlassian, coming to talk about it live. We recorded it yesterday. So listen on Thursday to that pod. It's pretty good because Mike talks a lot about this. Practically speaking, his response is there's nothing he can do. He's got to go with it. He's got to motivate the team and get them through it. It's like, you know, Atlassian beat the quarter. It's re accelerating at 5 billion in revenue and it trades at 4x ARR. Like our world that we know doesn't really work If Atlassian gets 4x ARR or if Shopify growing 30% at 12 billion falls 10.4%. The world that we've grown up in the last decade doesn't really work well. And you really can have two options. You can sort of hope and pray that we get back to the old days. I think that's what most startups are doing when I get investor updates and folks that aren't that AI forward, that are still single product, they're just kind of hoping that things will get better because they're not publicly traded. But the simple answer is you have to get AI budget, you have to grow faster. That is what the world's saying. It's hard. Cry me a river. You have to grow faster every single quarter. And if you're not in B2B you're not getting AI budget, you're not changing the world. We had to build the sponsor portal because the dated crappy SaaS B2B products weren't, weren't agent level, they weren't disruptive. But the one we've built is disruptive. Now hundreds and hundreds of human hours are going away because AI can handle the sponsor portal. That's what you have to build something that good or it's going to get canceled. Like this 10k product, we just aren't renewing. They may not even know yet because we've ghosted them. We don't need them. There's no chance to save the deal. We've already built Our own. That's happening to folks. You're either at the cutting edge of AI or you're getting displaced. And the question is just what's the speed if all your customers are on five year contracts? You can hide if you want, but everyone, there is no middle ground anymore. That's the best advice I can tell you. And then one last thing I would say to founders on this. I wrote this on Twitter yesterday. Then I'm a longer post on Saster. Ask yourself calmly use your own product. Is your jaw dropping? Is your jaw dropping? Almost every, if anyone, that has even a couple million in revenue, up to billions at some point. Your product was a jaw dropper or you never would have gotten there. You, you utterly changed the lives of a stakeholder of a customer. They didn't buy you because you're 1% better. They bought you because you were you. You're jaw droppingly better. At least for some niche use case. Be honest. In the age of AI, are you as jaw dropping as Claude? And if you're not, that's why you're decaying. And you've got to find that jaw dropping AI moment where your jaw drops. That you're a AI agent is so good you can't believe that was even possible six months ago. That's how you're going to build a $10 billion company today. Or you're going to end up trading at one times revenue. There's nothing in the middle. So ask yourself honestly, would you renew your own product today? Would you pay more for your product than last year? And would your jaw drop if you're exposed to your product? Are you trying to ram something through with a decaying sales and marketing team? That is not a jaw dropper anymore. That is not a wow moment. Where's your wow moment from AI? They're all over the place. All the Saster AI tools we put up on Saster AI slash agents are, they're, they're, they're wow that your jaw drops when you use them. Right? You know. Sam Blonde launched a tool Monaco. It's our next AI SDR tool. Like we don't need another one. It launched today. You can find links to it on our agent page. It's even better than the other ones. You know what it does with this Monaco one? Like it finds your icp. It calls them for you and sets up a phone call or a meeting for you directly. Do any cadences or anything. It literally sets up meetings for you in your calendar for real that have been confirmed with a human. I Mean, that's a jaw dropper, isn't it? Amelia didn't even believe how good it was until she saw it. And it's a jaw dropper. You want to pay 53,100k for a jaw Dropper? Yes. Do I want to pay $200 a month for another crappy project management tool? No, that's today.
- [52:04]
B
Yeah, I give them just to tie it all in. Right. Like, I think the big thing that took me over the edge the last two weeks was like. I was like, okay, it's time to redeploy our sponsor portal. But again, the one we've used has zero AI and that killed. I was like, I can't. I can't be sasser AI and deploy something that has no AI in it. Like, that's just one totally terrible that they haven't done at least the very basics with AI. But it's funny because after I showed it to you, Jason, I think you made the comment like, well, what about these other apps? And you named a couple other ones where I was like, honestly, I feel like I could start to rebuild it. Maybe not till after annual, but there's a few now that I'm, like, literally looking at where I'm like, they fall into the same bucket of. They have not benefited from AI at all. There's just no new AI features. There's nothing good about it. There's no benefits to it. I'm like, you know, I'll probably move off those two or three other tools we named next because there's no benefit to staying. Like, there just really isn't. So, yeah, Monica's the newest agent we're going to. We've. We've already started using it. We just got onboarded last Friday, so we'll check in in, like, another week or so and tell you guys where we're at. On it. Sam, who you guys might have known from Brex, who was the CRO there, launched this tool after, you know, going through this very frustrating process on his own of, you know, trying to wrangle human FVRs and VP who sales and other folks to do outbound. But also when you're starting out and you're less of a. Less of a, you know, maybe name Brandon's, Brex became and then got bought. Um, you may have problems at the early stage, filling your pipeline. Right. If you're at super early stages, you may have nothing in your CRM. And so Monaco is really cool. We're actually using it for a very unique use case that I think is not, it's totally not the recipe, but the tool's pretty good. Like, you gotta let me think, they're like, what do you have? Do you already have like three or four sales agents? I was like, yeah, but this one's so good. But you gotta let me use it because I have. I was like, I have another use case for it. Like, I have another part of sales and go to market that's not being tackled that I want to tackle with this agent. And then this agent can go deep on that. And so I'm pretty stoked. I'll go to the next question. That's Monaco. You guys should check it out. It's monaco.com, they actually own that domain. That's an interesting story, but maybe we'll have Sam on next time about that too. Okay, got the last question here. Get back to our chat. Someone asked if we've been looking at openclaw or claudebot. Do you want to quickly explain that Ren is on?
- [54:53]
A
We haven't. Listen, I get why engineers are interested. Let me simple answer. It's not interesting to us today. And breaking guardrails and having rogue agents running around, it's just not something we need. I get it's very interesting. And we built a. Yeah, we built, we built an app called Renai. You can use it. It runs autonomously and has extended memory. So I get the appeal of it, but it's really not for what we want to do, which is, you know, GTM type stuff. Anything that needs to run autonomously, we can just have an app run autonomously. We can just have these apps that we built on Replit Run 24 7. It's really not, I'm not saying it's not exciting, but we're not interested in breaking guardrails. Actually. What's more important for us is making sure the guardrails work. Exciting. But today we don't need this stuff. Like, it's not that I'm not interesting. We're all over mold book. You can watch the things our agent wrote about buying $400,000 of watches that we didn't ask for and other things which are all sort of fake that our agent wrote and we've built versions of it. But I know, I just, I just, I just think that what's much more exciting is just just having, well, guard railed agents that know your data cold and can run on their own. That's all we need today. So I'd say we move on to the next question today.
- [56:14]
B
Yep. This question from Steve was, you Mentioned that some of your webpage apps are public facing. What are you doing currently to protect the live coded apps and tools from attacks? How have you tested this?
- [56:27]
A
Look, I think it's a great question and I think it's something that the platforms, you know, first of all, this I'll give you, I'll give you two simple answers. One, the platforms have gotten better. They all do basic security scans when you launch an app now. They look for private keys you've embedded in the code, they look for other things. They do a lot of the basic stuff which is helpful. Two, one way or another, what you should do with the agents now is they're pretty good. You can do this in Claude code, you can do it in replit, you can do it in lovable is ask them to do the deepest security audit possible on your app at least once every month or two and they're pretty good at it. And it will, I'm not saying it's perfect, but it's better than most probably low end developers would do. If you had an outsourced dev shop and they'll find an infinite number of issues and it will take you hours and hours and hours and hours to fix them and undo it and it will overlock your app down and it will stop you from able to do uploads and it will block your internal APIs and it will drive you crazy. And each time we do an internal security update it takes about a week to recover to fix all the stuff we have to fix and unfix and refix and there are more and more third party tools. It's a real issue, it's a large issue and it is not an issue that's fully solved but the platforms are getting better and then before you release anything with any customer facing, at least with the agent, tell them to do the most in deep security audit possible. And I, I, I'm not a security expert but as a SaaS founder I've lived so many of the issues, I've been through so many security guards, we've done our own. I think you will get a decent outcome but it is a, it is an open area.
- [58:07]
B
Yep. Question related to just our time trade offs. Do we have enough time to do everything else that we're supposed to do?
- [58:15]
A
We're at the limit. It's a big, it's an interesting issue for the future. It's, it's an interesting issue for your team. I mean there are better people to build and manage agents than me and Amelia out there, but there's not like so many. And how many folks you have on your team, like who on your team could you hand these apps off to? Who can you hand off your agent management? I'm sure there are folks you, you are on your team that can do it. But I, I think honestly you're probably lucky if 5, 2, 1% to 5% of your team could, we could offload the management we have. And I think it's a very, it's a challenging issue of how many, you know, we are at the max. We did add Monaco, which we love, but only because it's disruptive, only because it fills a gap that we don't have. Otherwise we would not add another GTM agent. We're just overloaded with the inputs we have. Right. So I think this is an issue that most folks haven't contended with. But we're at, we're at our max and we're going to learn this year how to, how to do it.
- [59:11]
B
Yeah. Okay with that. So to do some point if you want to learn more, I use some of the Stafford annual branding for this slide deck, but staffsterranual.com will be there in May. We're doing a full deploy day that's new on Tuesday. So content just like this on deploying agents. Some of it we'll probably do but otherwise it's going to be mostly other folks telling you how to deploy different kinds of agents. They'll go deep, they'll do deep dives. So that'll be exciting. To that thanks all for joining here. Thanks for the nice comments and all that jazz and we will do another one of these fairly soon because the world keeps changing every it seems like every week. So we'll do this soon. But yeah, in the meantime if you want to know more about the market and stuff. Yeah listen TO it's the 20 BC podcast that Jason does with Harriet Murray every week. It is Thursday. It drops Thursday morning. You can go to the 20VC YouTube to watch it. There's a lot of really good insights. There's no politics much like this. So highly recommend listening watching that one too. With that. Thanks for joining. See you guys next time. Hey Caster, imagine having agents for every support task. One that triages tickets, another that catches duplicates, one that spots turn risk. That'd be pretty amazing, right? Happy Fox just made it real with Autopilot. These pre built AI agents deploy in about 60 seconds and run for as low as 2 cents per successful action. All of it sits inside the Happy Fox omnichannel AI first support stack. Chatbot co pilot and autopilot working as one. Check them out@happy fox.com.