The Secret Sauce to Scalable AI Agents | Agentic EP 08

Agency AI started from early 2023. We focus mostly on our developer tools. So Agent Ops is the observability platform that we focus on developing. And also AgentStack is a way for people who are not ready to get observability or testing going. But just like a very fast scaffolding.

Tool, the barriers to entry are becoming so low to building up these applications so you can go build entire sales agents. We've seen with this, we've seen research agents, we've seen doctors start building agents to analyze broken bones using Agent Stack. And so we're getting like all these crazy, crazy use cases.

We spoke with several VCs and they just sent us feedback like, look, you guys are incredible. I just don't think agents will ever work. Fast forward to 2025. The only thing people talk about are agents.

Any founders who are listening. VC funds are amazing, but think about who their Rolodex is. Think about who you can get access to from their Networks here in 2025.

Even though the lean startup has been out for almost 15 years now, the knowledge is there. Why do you think so many founders still don't go talk to customers and Instead will spend six months, 12 months a year just building the sol and really focused on product falling in love with the solution as opposed to falling in love with the customer and the customer's problem.

Sometimes people get so ingrained into what they're building that they're afraid to think that people don't want it. They think that what they're building, they must know the customer. Founders and engineers will say like, I don't need to validate because I'm the validation and that is catastrophic. It is vitally important to talk to your customers regardless if you are the customer or not, or if you're afraid to talk to them or you're afraid to hear that it's not a good product. You have to.

Welcome to the Agentic podcast where we dive deep with the pioneer AI founders reimagining business and investors betting on them. We bring you the war stories, insights and perspectives you won't hear anywhere else. I'm your host, George Van Darien, the general partner of Untapped Ventures, a VC fund that invests in early stage AI startups. Let's go. Welcome back to the Gentec podcast everyone. I'm your host George Bandarian and today we have three very special guests with us. We have the co founders and founding Engineer of Agency AI and they are building tools to make the next 1 billion agents fast, safe and reliable. This is going to be an exciting episode. The CEO has managed global teams and built numerous machine learning products used by top tier institutions. He also has a hack for winning hackathons. It's worked 17 times already, so we'll dig into that. The COO is a serial entrepreneur and has worked on campaigns for brands like Samsung and Beyond Meat and GoPro, leveraging his expertise in AI and marketing and technology. And the founding engineer is a former founder herself, software and machine learning engineer, and rides motorcycles and is a dj. So let's welcome Alex Reidman, Adam Silverman and Braylon Boynton to the show. Great to have you here, guys.

Thanks, George. Great to be here.

Yeah. And you guys are in your offices in sf?

Yeah, agency headquarters here in SOMA in San Francisco. The AI capital of the world.

Alex, let's dive right in. What exactly is agency AI and why is it so special?

So agency AI started from early 2023. Adam and I, we were working together to build our very first AI agent. And what we did was we built an AI agent that would solve a very common problem for accountants and people, anybody dealing with numbers and websites, which was filling out forms. And what we realized early on was GPT was exceptionally good at actually just filling out and understanding parsing data. The big challenge though was it was hitting a threshold of something like 90% effectiveness, but it would crash a lot. It would go in circles and just lead to very unexpected failures and hallucinations. And I think the broader idea we came away with it was agents kind of sucked and we had a lot of very angry customers. And to solve this problem, we actually built out our internal developer tool set to help debug what was going on. Why are they breaking? Why are they so expensive? Why are we getting so many angry emails from customers? And what we realized was just by speaking to other people in the ecosystem, we were not alone. There were actually countless people, countless developers who had the exact same issue that these agents really just didn't work as we expected. But we felt that these developers that we had made for ourselves were actually very, very good at diagnosing the underlying problems. And that's when we built out our went from our first agent to building a agent infrastructure player developer tool called Agent Ops. And Agent Ops serves as debugging, observability and testing platform for AI agents on the developer scene.

Excited to dig into that and unpack that and also understand where you play in the ecosystem. You know, just Yesterday we had OpenAI announce their operator agent and you have your lang chains and Lang graphs, you have your crew, AIs. So I think it'll be helpful for us to understand where you sit and you know, what, what problem you solve within that stack. But before we go forward, let's take a step back and just get to meet the three of you guys. Would love to understand a little bit more about who you are, you know, a little bit about your background and really what led you guys to come together and create this company, what the origin story was.

Adam Silverman I'm the CEO of Agency AI. I previously built and sold a startup in Canada. So we were doing Influencer, basically created a marketplace for influencers on TikTok at the time and musically we sold it to a public company and then I was an early employee, built rewards, leading their growth marketing efforts. I can give a quick background. Just like I've always been trying to automate my work. I think everybody should be thinking about this as they're going through their day to day lives. Start my career in private equity and then went to go build a very basic search engine using optical character recognition to basically pull Influencer data. And this was like a very manual process at the time. And I was always just trying to figure out ways like can I automate my job? I think it's something that I really empower everybody listening to think through. And Alex and I, we've been jamming on a bunch of ideas and I think we kind of just talked through how we got into this was building some of these data entry automation tools. But us kind of the background of machine learning and Braylon maybe give some context, kind of how you got into it. But we've all kind of had this passion for like figuring out what is the most boring work people do, what are the most like expensive, repeatable tasks that are low risk that businesses can actually go automate today with the current technology.

That's funny Adam. I had a background in OCR as well. I came from the document management workflow automation world where it was all around reading paper documents and making it digital and then automating the workflow. So very cool.

My name's Alex Reedman, I'm the co founder and CEO here at Agency. So before working on this actually I've a lot of experience working as a machine learning engineer at a variety of different companies and industries. So first started out in the world of cybersecurity where I helped build ngav, which is next gen antivirus. So essentially employing machine learning algorithms to identify novel malware. This was kind of during the era when Russia was hacking Ukraine with ransomware. We were building these detectors that could identify when a malicious file was in a system and essentially block it before it could do anything harmful. Later went on to the world of fintech, where I worked with a lot of large clients as a consultant such as Goldman Sachs. American Express did some research around how to do predictive modeling around credit card transactions to identify fraud, identify patterns in spending to basically better sell ads, and a lot of other interesting projects related to the space of just like banking and consumer data. And just kind of being in that world exposed us, at least me, to the obvious things that can be automated in these workflows, especially banking and fintech. It's like, wow, a lot of this work is really kind of just like asking to be automated. And then agents come along and wow, here we are. We can actually see it happen.

I'm Wayland Boynton. I'm the founding engineer at Agency and the lead in charge of a project called Agent Sacred, which we'll definitely talk about later. My background I've been in machine learning. Since I learned what machine learning was, I've always found it so fascinating that we can build algorithms that have a semblance of intelligence much more now than when I started. I worked in traditional machine learning and built a startup in machine learning as a service in 2017 that was acquired and really just kind of kick started my career being in love with machine learning and in love with being an entrepreneur. I actually found out about Agency because I was building my own startup. We raised money for building an agent and the agent was terrible. It was not super capable. We had a beautiful wrapper around it, but at the core of it it was the agent was not capable of doing what we wanted it to do and what we needed it to do for our customers and ended up winding that down. And in the process of doing that, Alex was a friend, knew him through founder circles in the San Francisco area. I realized that really what we were missing was a great developer platform to be able to learn what was wrong with our agent and then also understand in production where was it failing, where was the shortcoming so we could have better built it out, learned about what Alex was building, thought there was a ton of potential in that. And I've always enjoyed working in the DevTools space more than that. Agent that I was previously building was in influencer marketing and I like dev tools a little more than influencer marketing. And so I decided to join the team, be the first founding engineer and have really been just enjoying the process since, like building great developer tools, having so many agent devs who are telling us every day like how much they love the product that we're building, how much easier their life is and it's just, it's a very rewarding thing.

Very nice, very cool. Well, it's great to get to know each of you a lot better. And so, you know, why don't we dive in? I think given your background, you have some unique perspectives on where the state of agents is at today and where it's headed. So would love to hear from you on that.

Cool. Like I could give you a quick rundown. Is that when we Start Building Agents 2023 and like we were the crazy, you know, people at the end of the block trying to convince people that agents were coming. And 2024 we saw a lot of it's a predominantly startups trying to explore agents, trying to test things. We saw the rise of like platforms like Lanechain and crewai and Auto Gen and a few of these other ones. And now what we're starting to see going to 2025 is that a lot more companies are saying from a leadership perspective we need to start investing in agents. So we're starting to see a lot more, I would say like more significant POCs coming out. We're starting to see a lot more budget opening up for building agents. And the part that excites us more is like dedicated engineering teams for actually building agents. This is like always the thing. Like if you have a dedicated AI agent engineering team, I promise you're going to be light years ahead of the competition. And anyone listening to this, maybe if you're like with a bigger company, like maybe a Fortune 2000 company, I'd really start thinking about where your engineering resources are now and trying to see can we deploy them on the Gentex side Because I think it's going to be one of the biggest productivity gains can be made over 2025, going to 2026 and beyond.

Yeah, that makes sense. And what are your thoughts? We touched on it earlier. What are your thoughts about OpenAI's new announcement?

I think we've seen so many browser agents, it's cool to see we're good friends with just about everybody in the space who's building them. Alex and I even like, we're tinkering with browser use agents with a product called taxi AI in 2023, I think was it GPT3 back then that was powered by. So we've seen the evolution of this. I think it's nice to see that it's getting more reliable, but there's still, like, there's still a lot of kinks to work out. I think this is like, right now. I still think we're in like the toy phase of these types of browser agents. I don't think we're at a place where, let's just say I want to go make a trade for me on Robinhood. I wouldn't feel confident with it going and doing that quite yet. Maybe. Alex, Marilyn, you have some other takes, but that's kind of what we're seeing on our side.

I think it's a bit more infrastructure that needs to happen with the Internet in general. It's kind of this. There's actually a little bit of a debate between should the agents get better or should the Internet be better served for agents? There's a number of standards that people are trying to push, like ll stop. Txt, or having an agentic layer to every website where rather than having a captcha to prevent bots from getting in, saying, instead of using this side, use this side. Here's all the tools that we can make available to bots. I think that it needs to be somewhere in the middle. We have incredible tools that are great at scraping, great at interacting with the human layer, but in reality, everything that is served on the Internet has been rendered in such a way that it is best optimized for humans to interact with. That kind of means that we're abstracting things away from the data that the LLM wants to access. And so inherently, any type of browser use agent is going to have some difficulties because of how it's been presented. So meeting in the middle is where I'm really curious to see what will happen. I don't think the operator is necessarily the silver bullet that's going to fix browser agents, but it's still a very cool thing to see, like Adam said, that it's becoming further adopted and we'll see more people using it.

So one of the things we talked about in the beginning was this idea that, that we. You have different products, right? You have. And essentially, if I'm starting to build an agent, how do I start? Once I'm in production, some new problems crop up. And then once I'm really scaling that out, maybe you could talk to us now about your product portfolio. What is each product? What problem does it solve? How is it different than its competitors for what it does?

The main reason we call ourselves agency was actually a variety of reasons. So the very first is that we're Building high agency machines. These AI agents are essentially things that have their own volition and they have the ability to change the world in a programmatic but also kind of like self driving way. But secondary to that is we realized that early on observability was a big challenge. So we built the way to see your agents. And so we call it the Agent See so entadra intended. And then, but also lastly we also say that like look, you know there's going to be a sea of agents here, right? It's like there's going to be countless of them. And so what we've actually figured out is the big challenge right now and this is we're still very, very early in kind of the AI life cycle. If you live in maybe the west coast or Bay Area, you're thinking about this stuff a lot. But most of the world is completely disconnected in a lot of ways from what's going on. So we did a lot of things so very first we focused mostly on our developer tools. So Agent Ops is the observability platform that we focus on developing. And also Agent Stack is a way for people who are not ready to get observability or testing going, but just like a very fast scaffolding tool. Right. So you kind of think of it as a create react app or like a starter template to build out your own agents and make it really fast and easy. And that's a big challenging question. We get a lot which is like I want agents, I have a problem I want to solve, I, I just don't know how to get started. Second to that is we actually manage the world's largest like AI agent marketplace. We call it a marketplace, but truly a directory of hundreds of agents that we personally vetted and we know are very high quality. And the very last bit is I mentioned earlier, agency goes many layers deep, but we've also kind of expanded to consulting services. And that again there's a lot of people with a lot of budget and they have a very clear idea what they want to build. They just don't have the expertise to. And just over the, over the time of tenure of the business, we've actually developed this really wonderful network of expert developers that we work with to help solve huge challenges for companies that have budget to build out agents themselves. So we ourselves are an agency in a way.

You know, I think that when I look at the different agent products, including the dev tools within the origin story, there is a root kind of ICP that the company has in mind. What is that for you guys, of course, over time you want to scale and you want to have multiple markets. But right now, I guess give us an idea like where are we at with traction? And then what is the commonality with those customers?

Yeah, I can speak to that. So on our developer platform side, we're very aggressive about tracking analytics, about who's using us. And we have thousands of developers who are signing up every month just to use the product in order to say, hey, I have an agent that I built out and I have no idea what's going on. The logs are too verbose. I don't know how much I'm spending on LLMs. I don't know how to compare runs. Right. How do I know Claude is better than GPT? How do I know Quinn is better than deepseek? These are all questions that people are thinking about themselves. So what they do is they use our tool Agent Ops to plug in and we have just thousands of developers who use the product every, every month. It's kind of a huge growth. And then Also Braylon from agents.gov you can speak to how people are adopting that.

So before we go there, so these are thousands of paying customers. They're paying based on API calls and some other subscription or.

Yeah, to be as blunt as possible, we have like a lot of indie developers who are just signing up and get started for free, but once they start tracking sessions at scale, that's kind of when we start charging them. And then we also on the other hand have like large Fortune 500 companies, banks, regulated companies who are coming to us and they require a whole suite of different tooling. So they require like on prem hosting, we have an evals product just for our enterprise clients. We have a deploy functionality. We can actually deploy your agents just for our enterprise customers. And this is kind of different, kind of a more advanced toolkit that we've present or that we present to them. And these companies are now putting together almost like agent Innovation labs internally getting these SEAL Team 6 type leaders who are coming in just saying, hey, we want to go figure out agents. And these have been some of our favorite customers. Work with, work with.

As of recent, Fortune 500 is a big focus that we have. We also have a company culture that is highly or highly revolves around like hacker mentality. We have a lot of indie hackers, as Adam said on the platform, that's always been a priority of ours is to make sure that we can support large agents at scale, in production, but we can also support somebody from anywhere who's just Having fun, playing with agents and give them a great tool that we can provide to them for free. We do a lot in the hackathon space in San Francisco and actually globally. We participate in a lot of hackathons, provide our tool, but also just to understand what are people who are coming into agents, fresh new developers or experienced developers, what do they see, what are they excited about? Gives us a lot of insight into the hacker community with agents.

So give me an example. If you can name names and give examples of use use cases of what is a large enterprise, what have they done? What did they automate with the agent? And then I know that Alex touched on it a little bit, but let's just make it concrete for the listeners to understand based on some experiences that you've had of specific customer use cases that you solved.

So I can talk about two very concrete use cases that we've seen on the enterprise scale that we deal with. So for example, one is from a travel agency company. So you kind of think of like a booking.com style travel company where they have flights, hotels, transportation management. And what they built out was they had this problem where they want to make a chat interface so that their users were able to communicate, hey, I want you to change my ticket or I need to update my passport number, my ticket deals and your refund, etc.

Etc.

And originally what they had done is they said, okay, we're going to build an agentic chat support system and we're going to allocate six months and an entire engineering team towards it. And instead actually what they're able to you what they did was they use agent ops. And since they were able to go so much quicker, they were able to get adopted too. Now the question is like what did they use to build this thing? So they used an open source AI agent framework called Autogen from Microsoft. And what Microsoft did is they said, okay, what if we create a multi agent orchestration layer where you can actually have multiple agents work together. Each has a specialized role, specialized tools and different proxies by which they can communicate to each other. And that way they were actually able to build a more efficient system than just like chaining LLM calls into one single system prompt. And in doing so they've actually built a very, very effective product. It's live right now in production and it serves a ton of users. Another case study that we've seen is also in the support agent line of things. So this is actually a software company that Mostly they sell B2B SaaS that's paid only. But what they learned was that they wanted to expand to the free tier because there's a lot of customers who were interested in trying out the product but didn't have the budget quite yet. And they wanted some basic free tier features that they could basically upgrade into a paid plan later. The problem with the system though, was all these free tier customers are wasting hundreds of hours of human customer support.

Agents.

Right. So what they did was they actually built an AI agent orchestration framework using crew AI. Crew AI allows you to also use multiple agents that work together to solve problems. And what they were able to do was find customer data, answer questions, look up document information about pricing, how to use tools, etc. And in doing so, they were able to automate over 80% of their tickets that were hitting the system. And it was so helpful that it actually made their free tier substantially more profitable than it was originally. So support agents, as kind of like general use case, are seeing a lot of adoption, especially in the enterprise world. And there's dollar signs attached to it. Automating humans, or at least making their requests routed more effectively is one of the best ways to put agents to use.

Yeah, I can share a couple of weird ones too. Like we get about 50 requests per day of companies coming to us asking us to build agents. And these are companies of all sizes. And we had a company come to us recently that actually want us to track every obituary in the US So they want us to come and find, okay, every time somebody passes away, we want to figure out, like, who passed away, who's their next of kin, how can we get in contact with them. And they had an entire team of humans that basically were on computers all day doing this type of research, trying to figure out next of kin, going onto LinkedIn, trying to like navigate and trying to actually figure out who these people were. And they built an agent that actually automates this entire process, which is crazy to see. And we've been helping them over at agent Ops. These are the types of use cases that are like, there's a whole wide range. Like, we're seeing a lot in the research space, we're seeing a lot in customer support. We're starting to see a lot of demand for voice agents. So if anybody's like thinking about voice agents definitely reach out to us as well. Insurance. There's so many different verticals where these agents are coming up in. And we kind of have this framework for which we think through when you're going to build agents. And one of the big things we always talk through with our customers is we always think, has a human done this task? Like, if a human hasn't done a task previously, there's going to be no way to automate it. And we think about, like financial research, for example. Like if you have a banker who, you know, spreads comps on different equities or something like that, there's a very clear pattern as to what they do. You know, they go to Bloomberg, they pull their data, they go into Excel, they make their model, they take that model, they send it to their MD or something along those lines. And this is an entire process that once we document it and we get understand the kind of intricacies that go on throughout the whole process, these are things that are right for adoption or automation. Yeah.

So essentially, when you're developing these systems, a big challenge with them is that they're very slow, right. And you want to be running a lot of concurrent experiments and you want to share those experiments with your team members. So, for example, devising the right system where you have multiple agents together, it takes a lot of variations, a lot of experiments, experimentation. And so what you end up doing is you run a bunch of terminals on your machine, right? You have a bunch of different experiments, and you want to see which ones are leading the right way, which prompt is the right one, which LLM is going to be the best at being. Sounding most human, which LLM is going to be the best at doing procedural tasks. In doing so, you're able to basically work with a team. It's like a collaborative tool almost in that you're able to see the runs, you're able to see which ones are working, which ones are failing, why they're failing, and also manage cost, especially in the customer support world. We've seen tickets. You know, again, like human agents are quite expensive, but AI agents too, if you're using an expensive enough model and if they're going in circles enough, they also get quite pricey in a way like they actually might sometimes match if you're not careful the. The effectiveness of a human agent. So essentially just kind of building a observability framework lets you benchmark and monitor how effective your systems are. And it just speeds up the process rather than what we call engineering. On vibes, Vibes engineering can get you pretty far. Right. But eventually, when you're trying to build that kind of a flywheel effect where your system feeds your system and you're able to monitor and run experiments in a rapid fashion, like that's where you need a system Like Agent Ops to help out.

And you're saying that like Lang Chain, Landgraph and Crew AI don't have that, or they have some basic version that is not as good. And so you guys are 10x better?

We're definitely 10x better. But I mean the reality is a lot of companies are coming to us and they're using all these different systems, using Lane Chain for certain applications, using Create to maybe make research agents, using Autogen for customer support. And so when you think about the businesses of the future that are going to have multiple agents, maybe even eventually hundreds of agents running within the business, certain frameworks and certain methodologies are primed for certain task completion. And they want one platform to monitor everything in one place. They don't want to go to 12 different vendors with 12 different observability platforms. And one thing we also support is cloud custom agents. So if you're not using a framework like CREI or Auto Gen, you want to go build something completely from scratch in Python. We support that as well. So basically having one like data lake and one control panel for all this data has been a massive, massive reason that a lot of these, like larger companies are coming to us because they want to track everything that their agents are doing. Everything from who initiated the agent all the way through to, you know, what was the final command that went, went through to production.

Okay, got it. That makes sense. So then you, you do that enough times, you understand like the voice of customer what their pain is. You're solving that and you're also seeing a lot of people that are earlier in the process. So then you guys come out with this starter product, like Agent Stack, right? So now tell me, tell me what makes that unique and different and better for someone that just wants to start versus going to the other starter products?

Definitely. I think that what's unique about it is the fact that we're looking at it from a complete, completely different lens than any other, any other tool. Right now there's a little bit of a confusion where a lot of people think that we are building a framework, and that is absolutely not what we're building. We're trying to be completely agnostic to every layer of the stack. So the best way to kind of understand what Agent Stack is is we help you define your agent stack. So when you build an agent, inevitably you have a tech stack. You have a certain number of layers that reduce. Regardless what agent you're building, you will have these layers. You have your LLM provider, then underneath that you have your framework, which then Helps try to allow the LLMs to work together as an agentic system. Above that you have your tooling. So your agents need to be able to interact with the world in some ways. Those are the tools, your tooling layer and then you have an observability layer. At minimum, those are four layers that when you go to build an agent for the first time, say you look at Chrome and you think, okay, this is fantastic. Let me start with crew. That's just one layer. Beyond that, it gets confusing. There's a lot of providers to each. Each layer of the stack that kind of just are floating in the ether where you're. You say, okay, well I want to add observability to this. How do I pipeline my data into observability? We do make it easy with Agent Ops, but on top of that with tooling is what are the tools that are available to me? Okay, I found this tool. How do I build, like say I want to use the Perplexity API in my agent. How do I build that in as a tool without having to do some custom work? That's where Agent Stack steps in and says, okay, here are the layers of the stack, here are all the providers that you can choose from. Here's like all of the LLM options, all of the framework options. You can build the stack how you want and we pipeline everything through it. So it's entirely an open source project. Right now there's no, there's no profit motive. We're simply building a really, really cool tool to build agents even quicker. On top of that, we have a CLI that will scaffold your project for you. You can switch it out. So say you build out a crew project and you say like, how would this work? If I use Langgraph, you can hot swap your framework and transfer everything over to Langgraph and then evaluate Agent using that. So it's a really cool way to kind of unify the agent stack. And that's really the core of it, is trying to create a little bit of a standard as to how you build agents. And on top of that, lower the barrier of entry to make it even easier to start building agents.

And Bradley, maybe while you pull a quick demo on it, I just want to share a couple things that we think through quite a bit here. We had one company come to us, there's like a series B company and they've been spending weeks building up this agent and I'm not kidding, like around five minutes, Braylon had built like a demo for them that what they've been building like over a matter of weeks. And this is the power of like once you kind of unify everything under one kind of house and you say, hey, here's like, you know, we are, you know, we have the best tools, we have the best frameworks, we have the best observability. It's, it really allows people to now test really quickly. So sometimes we meet with Fortune 500 companies and they'll say, hey, we want this agent built. And in real time, Braillen is just coding it beside them and by the end of a 30 minute call they have a prototype they can actually start using. And this is the type of thing where like our speed to market is just increasing so fast. Like we can now, you know, what used to take us weeks or months to develop now is just, you know, a matter of minutes now with agents stuck.

So that's actually kind of a really fun thing to do in sales calls where it kind of blows them away, but it's fun to ask them like what agent do you want to build? So if you have an idea for an agent we can build.

Actually I do. I think we could have a ton of fun with this one. So let me frame it. We're an AI vc. We invest in pre seed and seed AI startups. We use AI obviously, but we want to take it to the next level. You know, we're a pioneer in agentic AI. I have a lot of thought leadership out there. We have a podcast called Logentic which you're currently on, but we're not, we don't have an agentic process. And so we want to, we want to build that out. We have an active project for that. What would we use it for? Obviously for the deal sourcing, screening, diligencing and investment process. So in my mind and kind of the way I've painted the picture is the best case scenario of this working is it's effectively a robo investor for small checks. So founders can hit the website, fill out the type form and then either instantly or within a few minutes get a decision based on the questions. The deck. Maybe we could ask them to take a phone call with the voice AI along the way, depending on what we need to tune, but at a very basic, you know, kind of a low risk, you know, 100k check, maybe 200k, whatever it is, because I'm sure it's probably going to have some false positives in the beginning that would be like the end state. I think that'd be really cool and interesting to be able to get to the point where if you're some hacker or some PhD student or whatever your background is, you can just be able to shortcut that whole process. I think that'd be cool and unique. But the main thing we'd be doing it for is not for that small check, but really building the systems for the big checks. And so they're just really improving the process of taking what we know. Like you said, there's humans that, that do the process of meeting with the founders, you know, doing the research, putting together the investment memoir. These are all very defined steps in venture capital. And so if we could take all the steps in venture capital and have agents that are helping us, you know, we would love to be a trailblazer with that.

A really interesting proposition. I think it's in all ways we're trying to automate workflows. And so it makes sense that if you're in the VC space and you're investing in agents, like, why would you not be dogfooding what you're buying? And so let's try to simplify that a little bit. For this example, I gonna go ahead and initialize a project that's a research agent called Easy dd. But let's come up with like just a task and an agent to do that. So you said people can reach out, offer some information about what they're building. Maybe we'll have a perplexity search that will then like look into the company, gather like what it can on the Internet and then create just like a short memo for you, just as like a, you know, just like an early screener for a company. Does that sound good?

Yeah, maybe look at the company. I don't want to make it complicated for you to look at the deck and all that, but look at the company, look at the founders, LinkedIn and then give an assessment.

Sure. Okay, so we just ended a project. We've got by default a crew project in here. Like I said, like we build on top of frameworks in that we have in our crew, we have one agent that is our researcher and by default has a tool perplexity. I'm going to go ahead. This is awesome. If you were building from scratch, you can add an agent with agents to stack, generate agents, whatever the agent name is, and then you can give it a task, you generate task. You could do this all in Lang Graph as well. We're working on adding pedantic AI and OpenAI swarms and a couple others soon. But by default, since we use the research template, we've got an analyst, researcher, perplexity. So we've kind of got a great start right now. All of our prompting is done in the these YAML files. So we can see in here we have a researcher. We've told it like you're an expert researcher. We'll give you a query and the analyst that will summarize this. We're actually like 90% of the way there already. This kind of feels like cheating, honestly. We'll go into tasks and we'll modify this a little bit. So instead of passing in a query, we could give a query as an input so we can have more scalable agents. We'll just type in here to make it simpler. We'll say, well, the researcher will go ahead and go in here and change it to say expert, B.C. researcher, whatever.

Yeah, like is it already going to pull from the LLMs what the expectations of what a good pre seed investment precede AI investment precede receipt AI investment would look like? Do we need to give it some variables or are we just trusting that the LLM already has that?

We, I mean we absolutely can give us some variables. I see. Right now let's just trust the LLM to do what we can just for brevity. But okay, we wanted. We're going to give it a company to vet and provide as much relevant data in a concise manner. We are interested to know if this company is worth a meeting to learn more about investing. Something like that. So we're telling it like generally what we should be looking for. Oh, we could also look at relevant data. Could include current amount raised, AR published, etc. And in our task, let's go ahead and say what's a example of a company that you'd be interested in hearing about or you think that could pitch you?

Would you guys be up for the Pepsi Challenge and put.

Yeah, let's do that.

Put yourselves in there and see what it says.

Let's do HNOFs.

Yeah, for sure. Okay. HNOps and AI observability platform. Looking to raise a Series A and we'll see what it gets from here. We've done some very, very minimal prompt engineering, but all the code's written, so let's do Visual stack run. So right off the bat you'll see we've got our Agent Ops link that pops up and this is the whole reason why we built Agent Ops. This garbage just fills your terminal on a more complex agent, it gets chaotic, it's difficult to parse through. So we have an agent that got this prompt. Here's all the information that we give it as well as the tool. So it has access to Perplexity. We have the task that it wants to accomplish and then it says, hey, I want to use Perplexity. This is the query Agent Ops AI observability platform, Series A funding. And this is the tool event that had the input and we can get to see what the output was in here and then all the way down. This is all of our agents in here. We have two agents, one that is gathering data, another one that's analyzing. The analyst was the last one and we'll see what it outputted. We can view it in here or in the terminal. But final answer, HNF's funding. No information about a Series A fund?

Yeah, not yet.

Yeah, I guess we don't publish that as recommendations for next step. I think with some prompt engineering we could get a much better result. But the key here, like, I mean we built an agent in a matter of minutes that did do research on Agent Ops. I think we probably need to define a little bit more like what we're looking for and then all comes down to telling the agent a little bit more of what it should be looking for. And that's the kind of thing that prompt engineering is like. You play with it, you work on it over and over until you find an agent that works. And that's the important part of evaluation too. As you can see as you're changing your prompts, how do evals change as well? Do you get a better success rate? Worse success rate? And there's some really cool prompt tools as well. That's one of the additional layers that you can add into an agent stack is dynamic prompting. Things that will slightly tweak what the LLMs are receiving in the prompt in order to get more optimal outputs. With a B testing, you can also.

Imagine that you said earlier, what if you uploaded a slide deck to the this. So Bramley, we could probably add an agent in one line of code.

Agent Stack tools add vision and right off the bat we can give it access to vision. Yeah, Total Vision Edit, Agent Stack project and if we go back to our crew here, you'll see that now we have the analyst has the vision tool here. So we could give it pictures of the deck. We can also there's a file reader too that we can give it. So all of our default tools, we also have a myriad of community tools. So I mean we can just do Agent Stack tools list real quick and see plenty of tools. So if you're looking for like image analysis, there's vision, search, perplexity, you've got storage, unified APIs. And these are just our, like, community tools. These are other companies that have come to us wanting to partner. So that way our agents hack users can have even easier access to their tooling.

And there's a whole bunch more coming. And if anybody's listening and wants to get the tool included, yeah, they make a PR and we'll review it as long as if we approve of the tool, it's good to be added to the stack. But yeah, from our perspective, like, we really, like one thing we love doing is going to companies and basically doing these like mini hackathons where we go in and four hours to show their teams what's possible. With agents, we do this in a matter of minutes. Like imagine we can do over hours and days when you really start kind of leaning into this. And so these are the types of things where it's like the barriers to entry are becoming so low to building up these applications so you can go build entire sales agents. We've seen with this, we've seen research agents, we've seen doctors start building agents to analyze broken bones using Agent Stack. So we're getting like all these crazy, crazy use cases. And this is like, it really excites us. Like, I still think we're very early the whole space, but we're starting to see more and more concrete examples come out.

Yeah, definitely. All right, so let's switch gears a little bit. We were talking a little bit about fundraising in that agent. So let's talk about your fundraising. So if I saw correctly, you guys have raised one round. Tell me how that went down. Give us the war story on that and then would love to find out kind of what your next plans are.

So our fundraising war story was we started fundraising kind of tail end of 2023. So we actually raised some angel checks from early investors in stability AI, some senior guys from Google Gemini. Back then it was Google Bard as well as senior ML researcher or ML applications expert at Airbnb. And we used that to raise. We spoke to a handful of VCs who were just complete thesis alignment, who really understood the future potential of AI agents. I remember distinctly we spoke with several VCs and they just sent us feedback like, look, you guys are incredible. I think you're what you're working on. You know, you have a very ambitious idea. I just don't think agents will ever work. Fast forward to 2025. The only thing people talk about are agents, right? We found Investors who really did believe in the vision of the future here, that it was not too far away. But you had to have kind of a lot of belief and conviction that the law of these models progressing the way they are is going to be a reality.

And just to maybe even call out some of them, like Kayla Kavanaugh from before, like, she saw what agents were going to be. She was amazing. Gaurav from before was just amazing. Like, he just, he got it, like, the second we fetched him, he's like, okay, this is. I see where the future is going. And then Matt Hawes from 645 Ventures and Aaron, they've just been like, you know, Matt came to us with like, I think, like 10 pages of research on why he thought, like, agents were going to be massive and where he felt like the agent infrastructure piece coming into play. And these are investors we just love, like, you know, from day zero. They believed in us. They understood where the market was going. They were, you know, kind of saw the vision in the future the same way we saw it. I think that's like one of the big things we always kind of look for, you know, as we're looking to the future, is just making sure that investors are very aligned with, you know, where you see the future going, that they're very aligned in terms of, like, making introductions and like, talking with, like, allowing us to talk with customers that they're close with. I think any, like, founders who are listening, I think the, you know, VC funds are amazing, but think about who their Rolodex is. Think about who you can get access to from their networks and these, I think a lot of the things that we recommend founders to lean in on as much as possible, just because there's, you know, you can be as high touch with your investors as you want. And I think if you're just, you know, one of the main things we get on is we get on monthly calls with these folks and we're just like, okay, these are the 10 people we want to meet. How do we get instructions to all these folks? And this has been one of the most pivotal things for our business by getting these, like, really, really high tech.

And that's great advice to founders. You know, at some point, the interview is two ways. Maybe in the beginning, the. The VCs interviewing the founder, but at some point you have the right to start asking them about, you know, how they do business and what their value add is and. And how they work with other portfolio companies and things of that nature. So that's Good advice. Did. So you guys raised 2.6 million back in 2023 or early 2024 by the.

Time it closed November, early December 2023. And yeah, I mean, just hammer on the point. It's like really, it's a two way interview, right? The very first question, this is almost the perfect disqualifying question. Very first question I would ask the VC is like, what's your thesis on AI? What's your thesis on agents? If I got a non answer, I knew almost on the spot that they.

Were going to pass.

But when people were saying, okay, I'm actually diving into the research, it was almost always a meaningful conversation that went either the last rounds or like actually, you know, majority of them closed. Right. So it's, it's very much like understanding, like who's, it's not just people giving you money. It's many partners on this journey and good partners is what makes the journey worth it.

So. So that was your pre seed round. When are you guys planning on raising your seed round?

I don't think we're going to talk about that on this episode. All I'll say is like, we're really like, we love chatting with VCs. We love talking like chatting with portfolio companies of VC firms who are building agents. So anybody who's listening, like definitely reach out. But from our perspective, I think we're, we're making sure that, you know, as we look forward into the capital needs of the business, we want to make sure that we have partners who are really, really aligned with where we see the future. We have some hot takes, we have some strong beliefs, but that's kind of the thing that we look at as most important. Gotcha.

Okay.

All right.

We can not touch on that. So not sounds like not actively fundraising or not looking to talk about that. So let's talk about the future. So where do you guys see this company going?

What we're seeing a lot of is just the developer ecosystem is starting to blow up and it's, it's gone from, you know, AI agents were kind of a cool experiment and the fact that like they're actually providing actual value and dollar sign huge millions of dollars saved for companies. So what we see is like it's going in a few directions. Number one, we really want to kind of help be kind of, you know, a midwife in a way and help kind of take these ideas from research to life. So we help companies that need help with this, with our agency kind of consulting wings. So if anybody's listening and actually has an agent idea that they want to build in a few minutes. You know, Braylon and I, we can absolutely help with that. But second is like we just want to help with the ecosystem, right? So there's so many developers who are hungry in the ecosystem that are trying things out, downloading new tools and we want to make it as effective for them as possible. It's like really this is some of the most groundbreaking technology that's ever been developed and we want to build a future as ambitious as, you know, the promise of artificial intelligence. So making sure the tooling is as enjoyable as possible actually gets the job done and really helps you take things from idea to, from dream to reality as quick as possible. So focusing on the developer experience is like our number one priority.

So you're playing a venture game. Is your, is your thought that this is, you know, you're trying to go public with this or build a hundred million dollar a year business or I guess if we pan out to the long term. How are you guys thinking about that?

Yeah, I don't think, you know, going 10 years down the road, there's going to be an agent that's built without agent ops baked into it. And that's really like, you know, we talk about empowering the next billion agents to be fast, safe, reliable. That's really where we're, we're seeing, you know, the future go. We have some like hot takes in terms of like where the future of agents will be. I think right now like every agent is very reactive in nature. So what we mean by that is that you have to like think about, hey, this is a business problem I want to solve. You have to then go figure out a tool to go build it out. You have to find an engineer to actually develop it for you. You have to set up evals and you have to have monitoring and you have to deploy the agent. And when we look at it like we see like thousands of developers on our platform every month doing this exact broken process. And what we're excited for is a future where these automations are actually proactive. So imagine like an agent that lives on your computer 247 and it's just recommending automations for you, saying hey, I saw you did this for the past week, I could just take this task off your plate. I think this is what's going to bridge the gap to see like billions and billions of agents being run every day. And it's a crazy world that like, you know, I think we kind of see the future. We see how like Janky this process is, and I think have good parallels to like hosting a website in 1998 was probably a long like, you know, similar complexity. There's probably like, you know, all these different hoops you have to jump through to get something out into production. Agent stacks make it super easy to just go and rapidly deploy these things. Agent ops make it super easy in two lines of code to get observability baked in. And we have some cool things that we're working on behind the scenes to see how we can actually make these, the whole agent experience a lot more proactive in nature opposed to reactive like it currently is.

Yeah, I love that. That makes a lot of sense. So, you know, the other thing that I'm curious about, and you guys touched on it a little bit, is some of the advice that you might have to founders that are maybe a couple steps behind you in the process. And I think one thing that's unique here is we have Braylon, right, A founding engineer. And, and so let me start with that and then I definitely want to hear from each of you guys on what were some of the key hurdles you overcome. What were the big lessons? What advice would you have? What would be your biggest advice to AI founders that are just starting out but having Braylon here? So you're a founding engineer. That's such a key hire, right? Such a critical member. You could have great founders like Alex and Adam and ultimately they could, you know, one or both of them could be technical and could code out to a certain extent. But ultimately you need that founding engineer and that's a really big quest that, that founders have to go through. So I guess how does someone find someone like you? What led you to take this role? Right, like, like walk me through becoming a founding engineer at a kick ass new AI startup.

I mean, you got lucky.

Very lucky.

No, I think I, I can say my story ending up in this role was I was very burnout. I had worked as a founder on, I mean my first venture, as I mentioned, was successful, was a great way to start my career. I thought they were all going to be that easy. It's not. I had like failure after failure, always like filled my way up, was able to like have a little bit more, more of experience, wisdom behind it, be able to hire quicker, raise money easier, and failed forward. But at the same time, it was years and years of my life where I was constantly grinding. I always wanted to be something bigger than what I was at the time and my entire identity was wrapped up in it. I had to Learn the lesson of being burnout. And after the venture I had mentioned previously, the agent had started to enter a downward spiral. I realized I needed to take a break and I wanted to take a break, but not completely disappear from tech and from the agent space. I also knew that agents were going to be the biggest opportunity I've ever had in my career. It lines up perfectly with my skill set, what I'm interested in. I couldn't just disappear from that. So taking a step back and entering more of a founding engineer role where I could have a little bit better work life balance, I could go party on the weekends while Alex set some hackathon or whatever, that was what fit me very, very well. And I think it lined up well that now that I actually had this conversation, I was able to get past that burnout and fall back in love with the job and with the community. And we actually have a term we use internally of influence engineer, where it's, you are not only just an engineer behind the scenes writing code, like there's a place for that. Somebody who you can just lock in a dark room and they can write code, they can churn tickets. That's amazing. But you really need somebody who can go out and speak for the company, do a demo at like a meetup night, talking about agents from a very technical perspective. And I think that that role is very vitally important in a team. I think in all ways we operate as just a team. There's not so much of a hierarchy as it is like we're all working to make agents better. We're all working to build the company and build something that's bigger than each of us independently. And finding somebody who understands that is very, very key in finding a founding engineer.

I love that. That's a great lesson there. What were the biggest hurdles, lessons learned and advice to the founders?

I always love selling unsolicited advice, but I do meet a lot of founders who are frankly just trying to go ideas that are way too ambitious. And I don't think ambition is a bad thing. But it's like, look, you're not going to understand what to build unless you grapple with reality. And the one piece of advice I give always is start with a very simple wrapper. We see so many businesses that are just some variation of chat with PDF and man, they are crushing it with numbers. Some variation of chat with my data, man, they are just crushing it. I see these companies that are trying to build like groundbreaking revolutionary AI framework platform developer, blah, blah, blah, blah. And no, they're not crushing it. They're failing miserably. And I say, like, look, if you really want to believe that there's. There's stuff in this Valiant space, build a very simple wrapper. It's not complicated. There's plenty. There's no competition, really. Like, there's. There might be competition, but, like, it's de minimis for the demand. So I just say, like, go into this and that's actually, we're going to understand your problem space a lot better. That's exactly what we did. We started ground up, like, we built a handful of agents. Like, we built this autofill agent. We understand the tools were necessary. We also built our own chat with PDF that made a good amount of money. It was like a week of work and we made $25,000. Doing nothing with really is like, make your first dollar before you actually think that, okay, I need to expand to something bigger and just challenge yourself. It's like, not hard to make the first dollar. But so many people want to go for something that's so much crazier, they end up going nowhere. So just like, you know, start simple. There's so many very simple problems to solve. AI doesn't have to be this insane system. It can just do something simple like formatting text and you're really good to go. Yeah.

I got a couple ideas and I think it goes kind of similar to what Alex was saying. I really encourage a lot of founders to, like, actually try and be a consultant first before going and building a product. We have some, like, engineers on our team who've kind of been doing similar type of projects. We know that. I'm not saying the wrong way, but they do have an expiration date. We know that, like, they're, you know, they're only going to be with us for the so long. But right now they're going and working on like, the most ambitious projects they've ever worked on. And this is something I can recommend, like, anybody who's going to start a company, like, get someone to pay for you to learn. Like, why not? Like, if somebody's going to say, hey, here's 20 grand a month to go build up this product for me. There's no better feedback cycle than doing that a couple times. And you obviously figure out the IP and kind of how that's going to work. But these are, like, things I really recommend everybody to do. I also think, like, one mistake and maybe just being in the bay is a part of this, but we see these founders raising these crazy, crazy rounds And I think a lot of companies need to realize that, like, you can. I had a business school professor, he had this theory that you should all twice be Jimmy Buffett, not Warren Buffett. So Jimmy Buffett before he passed away was worth like $600 million. Warren Buffett's worth 50 billion or whatever he's at now. And the reality is that you don't need to be 50 billionaire guy. Like, it's totally good to be Jimmy Buffett with $600 million and like, live a great life and own, you know, Margaritaville and land shark beer and all these things. And I really like, I think so many founders are trying to take these like, crazy, like going back to what Alice said, like crazy, massive swings, trying to make these multi, multibillion dollar businesses. And the reality is like, if you have a good business making 20, 30, 40 grand a month, you're in a better position than 99% of society. And we look at like, people like Peter Levels and all these folks who are just like building these like little micro companies and they're like, we're very big on building in publics. And you see the whole community is doing it. It's just so impressive to us that like, you could probably get started with agent stack, build an agent, put a front end on top of it, and probably in a weekend make a product that can, you know, generate a few thousand dollars in revenue. And like, this is the power of the technology, I think just lean into this. So whether you're doing it, start first as a consultant, whether you're just tinkering and seeing how you can put something out there, I think don't be afraid to like, ship something that's like half baked. I think it's something that, like, we've shipped stuff that like, at the end of the day, like, breaks and we have to go fix it crazy fast. But I think it's very important to get that feedback cycle because if somebody's telling you that your product is broken, that means that they're using it. And I think that's something that a lot of founders maybe, like, obviously you want to, you know, there's certain cases where you want to make sure that your like, infrastructure is not going to go down or your backups aren't going to go down or whatever it might be, but just throw it out there, see if, you know, throw a strike link on it, see if somebody's willing to pay for it. I think that's a huge thing that a lot of founders may be scared to do.

That's Great. And I realize in the beginning we talked about the. Your little hackathon strategy and we didn't come back to that. So I'd love to hear about that. That's. I don't think I've heard of anybody winning 17 hackathons. And so I'm sure there's some lessons learned there. Maybe you could share about that, Alex.

Yeah, I'm happy. This is kind of like the bread and butter. And I think that anybody who Googles us will also see a TechCrunch article that says that agency actually was kind of incubated in the San Francisco hackathon scene. So my background was out of. After college, I actually worked on my first startup. I won't go too deeply into it, but it was just a colossal failure. It was a dual lingo, but for accents. So what we learned is a lot of people want to sound more American or they want to sound more British, but they're raised in a part of the world where they don't have that accent. So they, they might have incredible English, but at least they don't sound good on the phone. So we built like this training app and I was so deep in the machine learning side of things at that time, I was like, okay, we've got to build this crazy machine learning algorithm. Them it's going to recognize AdSense and offer recommendations. We gave this app to people who spent four months building it and they all hated it. And they told me one thing. They said, alex, I don't need a machine learning algorithm to tell me I have an accent. I already know that. Such an obvious face palm like when you gave it to the thing to your customers. And this is actually kind of like a critical learning that I've embedded. Right. Which is you don't have to build crazy thing. You can validate without having to build the crazy thing. Just talk with people, really say, what would this solve your problem? How. How much would you be willing to pay for this? I can give this to you tomorrow. Give me $100. Right? Like really understand this kind of this value exchange. And I brought this. I was so burned out from that startup, I actually just started doing hackathons on the weekend to kind of scratch the itch. And what I do instead of coding for the entire time, I spend like maybe two or three hours talking with people say, hey, I'm trying to understand the problem space better. Can you tell me more? And I talked to everybody from ship captains to travel agents to insurance brokers and just say, like, I don't know anything about your industry, can you tell me about it? And just doing this process was like a really good idea generation tool to say like, hey, have you thought about this before? If not, what if I built it? And then that was actually kind of the key to winning all these hackathons. And frankly it just looked really fun for me. San Francisco, there's hackathons every weekend. I can just do this all the time. I won some pretty unique prizes. And you know, it's also good for getting on VC radars, which is I think what made the round a lot easier for us.

You know, one of the things there is like in sf, we see this hack like culture and it's so prevalent. We actually do this like hack night in our office, usually like Wednesday or Thursday nights where we bring out like the best agent hackers in all of San Francisco to our office. We play music, we have lights go on. And it's one of these things where it's like there's this culture, people just want to build cool stuff and it's like if you are thinking about building an agent or building an AI, there's no better place to be than San Francisco doing this. So we really recommend anybody who's listening and is in San Francisco, hopefully on a Wednesday or Thursday night, we'd love to have you by the office and just reach out to us. And we're always publishing on Twitter and stuff like that whenever these events happen.

Yeah, One last thing, we actually help co host and sometimes host our own hackathons. And it's such a great dog fooding tool for agent ops and agents type and say, hey, tell us if the product sucks and we'll give you a prize to use it. Right. So it's actually like a developer feedback that we've been able to hack ourselves and say, hey, we get 200 developers in a room all using the same tool and give us brutal, honest feedback about what we can improve. And that is like one of the highest ROI investments that we've made as a team.

Yeah, that's cool. Then you could start like taking that around the major AI cities and you know, building communities across the nation and across the world. But, but why do you think it's the case that here in 2025, even though like the lean startup has been out for almost 15 years now, there's so much that's taught and like every single area, what you said is what the knowledge is there. Why do you think so many founders still don't do that? Still don't go Talk to customers, and Instead will spend six months, 12 months a year just building, building the solution and, and really focused on product, falling in love with the solution as opposed to falling in love with the customer and the customer's problem.

I think it's a handful of things. A, it's scary. B, some people get wrapped up in the tech. They think, I can build this cool thing that's difficult to make, and they want to build it and accomplish it. They don't think about people want it. And third, I think that sometimes people get so ingrained into what they're building, building that they're afraid to think that people don't want it. They think that, like, what they're building, they must know the customer. Especially in the dev tool space. If you're a developer or if you're building something for you, you think you know exactly what people want. So people, founders and engineers will say, like, I don't need to validate because I'm the validation, and that is catastrophic. The worst thing you can do is not talk to your customers and not validate as you're building. It happens all time. We see it constantly in the bay. It is vitally important to talk to your customers, regardless if you are the customer or not, or if you're afraid to talk to them or you're afraid to hear that it's not a good product. You have to.

Yeah, it's a good time to say, if you get an email from me at some point asking for feedback on our product, please respond. Like, we, we literally take all feedback to heart. Like, we have a big review almost every day where it's like, hey, this is what people are saying needs to be fixed. This is what, you know, most pressing issues are. And it's like the best way. Like I said, like, it's good having angry customers. Like, you know, I think some people are scared of that. Like, it's, you know, it's scary to fail. But, like, that's the only way you can learn in my eyes. And I think, like, our team is very, very okay with the fact that, like, sometimes you put stuff out and people are not going to like it. Now it's just a matter of, like, okay, how do we now make this experience as great as possible?

Yeah, I think if we had a magic formula to help technical founders get over the fear that you talked about, Braylon, and just realize that this is the most important thing. And, you know, like you said, Alex, it's free in the beginning. Like, you said, you would just spend the first Few hours of the hackathon just calling and talking to people and like, you know, it's something that you could just do. If we could change the culture. It's different if you're at the compute layer or at the, you know, at the LLM layer there. It's the technical risk and we talked about that, that's a different game. But if you're in the application layer and so like, if we could figure that out, I think that would be a massive help to the community.

It exists called YC People. But no, it's interesting. I think there's an example of, if I'm not mistaken, it was Chegg, where when they were launching they just put out all these different textbooks to rent and they didn't even own the textbooks. And they were doing price, they were testing different prices and stuff like that. And they were just doing analytics on how many clicks they were getting for renting textbooks. And like, you can fake it till you make it. Like there's tools like lovable v0, replied agent. All these things like we're very big on, like just prototype something, throw it out there and see what works. And I think that's like, you're going to save so much pain and like, you know, developing a whole product. Like I sometimes go to Alex and like, hey, we need to build this feature. It's like, let's just like see what the demand is first before we go and invest like our most precious resources, which are engineers, on building like a full fledged feature when we can just kind of ab test it very quickly in public.

We actually had probably three things that we wanted to build with AgentStack, three different features and put like a front end. Just like an example of what it looked like visually on Twitter. Twitter, all three of them, just to see which one got the most signal and that like easily tell us what our priorities were. People were extremely excited about one thing, a little less excited about another, and didn't care about the third. And so, okay, we'll build it in that order.

Yeah, I think one thing I'll add there for other founders listening and like frail and I don't know if you want to hear this but like the reality is that the, if you have your team who's pushing you to go build the product, that's one thing. But if you have 50,000 people who just see a tweet now you're accountable to 50,000 people to go to go build a product. And it's a bit of it. Like, I know Amazon does this thing with their product press releases and it's super interesting like you know they do press release before they launch a product but we basically do this in public all the time and I think it's good to get these cycles going seeing what works, seeing what doesn't and I think it's one reason like you know startups still have the competitive edge against like a, you know massive players is because we can just rapidly iterate and like that's such a, it's such a valuable thing that I think as when you're in a large corporation you can't make a press release about a feature that doesn't exist or something like that and kind of a B test it.

It's been such a great discussion. I really enjoyed it. Alex, Adam Braylon I want to thank you guys so much for sharing your journey and insights and your passion for agents with us today. You've given us a lot to think about especially as it relates to.