Customer profile: Andres Zubillaga, Audos, on supporting a…

Audos wants to mint one million $1M companies in the next ten years. Audos co-founder Henrik Werdelin calls them donkeycorns: they party like unicorns but grind like mules. Audos gives everyday entrepreneurs everything they need to start a business: a landing page, an MVP of their app, marketing, analytics, a brand, and a community of other founders. Otto, its flagship agent, acts as each entrepreneur's AI co-founder.

Andres Zubillaga is on Audos's founding team. We sat down to talk about the Audos mission, and about why Audos chose Replay QA to power QA Guru, the agentic QA agent that Audos entrepreneurs use to test and improve their apps.

At a glance

The problem: Audos entrepreneurs love building, but most don't have the vocabulary to describe a bug to an LLM, or to read an error message. Asking them to test endlessly took time away from their customers.
Why Replay QA: It was thorough. It captured screenshots, read the logs, and explained what actually went wrong, so a non-technical founder could act on it.
How it works: QA Guru is a "tenant agent" in the Audos agent store. It takes context from Otto, calls the Replay API, and returns the results to Otto inside the entrepreneur's workspace.
The result: "It's saving them time. It's catching bugs they didn't realize."

Key moments

[01:05] The mission: one million $1M companies
[01:55] Donkeycorns
[02:30] The best place to build a business, not just a product
[04:20] Screen share: the Mike Tyson approach to entrepreneurship
[07:00] Portfolio entrepreneurship
[08:20] Relationship capital
[10:30] Life before Replay QA
[11:35] People don't like testing as much as building
[12:55] Finding Replay QA
[14:00] From one agent to an agent store
[16:10] QA Guru, built on Replay
[17:30] Tenant agents
[18:20] Early results
[20:20] A revenue share, not a subscription
[21:50] Human insights powered by AI

Transcript

Lightly edited for readability.

Meet Andres and Audos

Thomas: [00:00] Today I'm with Andres Zubillaga, part of the founding team at Audos.com. Not to be confused with autos [dot] com, the car site, because every time my transcript runs, it writes A-U-T-O-S. So welcome.

Andres: Thanks for having me, Tom. Luckily, with my own transcription tool, I've been very clear that "Auto" generally means Otto, our flagship agent, and Audos is A-U-D-O-S. But it's taken some trial and error.

Thomas: Happy to have you today to talk about Audos, the vision and the mission behind it, and how Audos and Replay found each other and built something pretty awesome together. I'll hand it [01:00] off to you, Andres.

One million $1M companies (aka Donkeycorns)

Andres: Our goal, as ambitious as it might sound, is to mint one million $1M companies in the next ten years. We believe we're headed into a world where entrepreneurship is not only going to be more important, but a necessity for a lot of people. We're trying to make it as easy as possible for people to be in business for themselves and be their own boss.

Our goal, as ambitious as it might sound, is to mint one million $1M companies in the next ten years. We believe we're headed into a world where entrepreneurship is not only going to be more important, but a necessity for a lot of people.

Andres Zubillaga, Audos

There's a lot of attention right now on how AI can help entrepreneurship, but a lot of it is an obsession with the first one-person unicorn. What's going to be the first one-person billion-dollar, ten-billion-dollar company? Sam Altman talks about it a lot. We're much more interested in the long tail. For every one-person billion-dollar company, there are probably a hundred one-person $10M companies, a thousand one-person $1M companies, and so on.

We call them [02:00] donkeycorns. They party like unicorns but grind like mules.

"We call them donkeycorns. They party like unicorns but grind like mules." [01:55]

Thomas: I love that.

Andres: That one never misses. We used to have an animated donkeycorn on the Audos website, but it got a little distracting. We have some merch, so I walk around New York City sometimes in a hat with a donkey in a rainbow party hat, and it always gets a second look.

Thomas: You're going to need to send me one of those.

Andres: Absolutely. Actually, we're due for more merch.

A place to build a business, not just a product

Andres: To do that, we've tried to build the entire operating platform you need to get started. Everything from getting a landing page and a V0 built very quickly. There are a lot of vibe coding platforms out there, and we strive to be the best place not to build a product, but to build a business.

"There are a lot of vibe coding platforms out there. We strive to be the best place not to build a product, but to build a business." [02:30]

So we have everything you need for your early marketing: it's fully integrated with Meta ads, you can run analytics on your website, get merch, define your brand. And what we've been focusing a lot of time on recently is creating a super vibrant community of entrepreneurs.

[03:00] Everyday entrepreneurs, as we call them: people who aspire to build donkeycorns rather than the funded Silicon Valley startup. They're in a Slack community sharing ideas with each other, and they have access to us. We do a lot of programming where they're matched with one another. We even run weekly seminars and a curriculum to make sure these everyday entrepreneurs are armed with everything they need to be in business for themselves.

Small team, outsized impact

Thomas: Amazing. How big is the team? That's a lot of activity.

Andres: Full-time, we're about six people, and everyone wears a bunch of different hats. I don't always know what to tell people my job title is, because it can be six different things on a given day.

We try to practice what we preach: AI should unlock a lot of efficiency in entrepreneurship. As much as entrepreneurship gets glamorized, there are a lot of tedious tasks, [04:00] whether it's running analytics, your email campaigns, or working on messaging. A lot can be automated now, and that lets people spend more time with their customers. We feel that way as a small team too. We automate what we can so we can spend as much of our time as possible with our customers, learning from them.

The “Mike Tyson” approach

Thomas: Is there anything you want to show us?

Andres: Sure. I'll show you how easy it is to get started on Audos. Nick, our CEO and one of our co-founders, coined a phrase that I've stolen and use a lot: we take a Mike Tyson approach to entrepreneurship. If you're not familiar with the quote, everyone has a plan until they get punched in the face.

Everyone's got a product plan or a business idea, but in reality, you don't know if it's going to work, or who's going to want it, until you put it in front of people.

Everyone's got a product plan or a business idea, but in reality, you don't know if it's going to work, or who's going to want it, until you put it in front of people. One of the businesses I run on Audos is [05:00] a wine education business, helping people learn about wine.

The idea is that, very quickly, we want to get you something to react to. We talk a lot about putting entrepreneurs in edit mode, rather than making them do all the hard work of creating things. People tend to do their best work when they're reacting to something.

We don't run away from the hard work of being an entrepreneur, but we want to reserve it for the most differentiated tasks, where the marginal utility is greatest. We think that's understanding your customer and spending time with them. So we try to speedrun the rest and get you a V0 as soon as possible. We think the real product development starts after you've talked to customers, not before.

"We think the real product development starts after you've talked to customers, not before." [05:45]

Thomas: Love it.

Andres: This is what our onboarding looks like. It takes you [06:00] through getting set up: your brand, what you might want, what customer you want to serve. It won't ask you much about capabilities yet. It's focused on: do we understand your customer?

Thomas: The first time I came across Audos was after the TechCrunch article. I immediately fired it up and was blown away by the ambition behind it. I had a lot of fun and built three ideas in an afternoon. I sent you some thoughts and feedback, which kicked off this collaboration between our two companies. It's impressive to see how much it's evolved in the three and a half months since we first started talking. The pace.

Portfolio entrepreneurship

Andres: I love that you mentioned trying out multiple ideas. That's something we believe in too. [07:00] Henrik, another of our co-founders, calls it portfolio entrepreneurship. We're now in a world where you can de-risk entrepreneurship in a way you couldn't before. You can get started quickly and cheaply, get things in front of customers, get feedback, and see what's sticking. In the same way you might make ten investments and a couple of them hit, you can almost start ten companies and see which one resonates with customers.

We're now in a world where you can de-risk entrepreneurship in a way you couldn't before. You can get started quickly and cheaply, get things in front of customers, get feedback, and see what's sticking.

A lot of people run multiple ideas on the platform and, over time, zero in on what works. Sometimes zeroing in means two or three, and we have a handful of people who actively manage a few companies.

When you get started, we get you to react to a landing page: the branding and the copy. Then you get set up with a V0 of your app. It's all based on the idea that Otto is your co-founder. It will even propose meetings for you to have and ways to [08:00] make progress.

Thomas: That's awesome. There are plenty of app builders out there, like you mentioned, and that's not the piece you need to be the best at. You need to be good at it, obviously, but the whole framework around it, making sure it's the right idea at the right time, is really smart.

Relationship capital

Andres: The way we think about that was the main thesis of a book we put out almost two years ago, called Me, My Customer, and AI. The thesis is that as the capability of AI increases, existing technical competitive advantages trend down. What remains as the main competitive advantage is what we call relationship capital. How well do you know your customer? How well do they feel you understand them? Do you speak their language? Do you even share their sense of humor?

"As the capability of AI increases, existing technical competitive advantages trend down. What remains is relationship capital." [08:20]

[09:00] We believe in empowering people to build for the customer they understand better than anyone else. If we put the right support system in place, technically and operationally, people can win with a competitive advantage based on a customer they understand better than anyone else. And I'm a big believer that everyone has a customer out there they understand on a unique level.

Life before Replay QA

Thomas: So while building an app is one piece of the overall value Audos brings to entrepreneurs, it's still a very important piece. If you think about the artifacts that come out of starting a business, it's your website and your app. A lot of the rest the public never gets to see: the planning, the rigorous customer interviews, turning insights into changes in direction, deciding when to pivot, when to zoom in, when to zoom [10:00] out.

This is where we came into the picture. When I started using the early version of Audos, I recognized the pains we'd experienced ourselves. For a minute, we tried to build our own app builder, and decided very quickly there was no way we could compete on that alone. We realized our secret sauce wasn't the app building. It was giving the agents that do the app building the context and evidence to fix their own mistakes, and doing that in a loop that produces a surprisingly better end product.

We realized our secret sauce wasn't the app building. It was giving the agents that do the app building the context and evidence to fix their own mistakes, and doing that in a loop that produces a surprisingly better end product.

— Thomas Daly, Head of Product, Replay

So I'd love to hear from you, Andres: before Audos and Replay started talking, what was your life like with these platform-generated apps, making sure the quality was up to your standards?

Andres: Of the many things we've [11:00] learned over the last two or three years, one is that people really love to build product. Not many people knew how to code before. Now the whole world can do something that seemed reserved for the top one percent: think of something and see it come to life.

That's a drug to a lot of people. They have all these ideas, they see them come to life, and they want to build. That's awesome, and those are the kind of people we like to work with: ambitious, with ideas. But as with any piece of software, whether it's built by expert software engineers or the latest models, you're going to run into bugs, or things that could be better. And people don't like testing their stuff as much as they like building it.

"People don't like testing their stuff as much as they like building it." [11:35]

Thomas: Yeah.

Andres: One thing that was certainly true a year ago: it was much easier to describe a feature you wanted in plain English [12:00] than to describe how something needed to be fixed. That often required more precise language for an LLM. A lot of the people we work with don't have the vocabulary to be precise about why the UI is off-center, or to read an error message and think, "Oh, this is probably what happened." They can say, "I'd like it to look like this." But being able to decipher what was wrong and communicate it to an LLM felt like it was lagging behind.

"A lot of the people we work with don't have the vocabulary to be precise about why the UI is off-center, or to read an error message and think, 'Oh, this is probably what happened.'" [12:05]

We want entrepreneurs to spend their time where it's most valuable: with customers, learning from them and talking to them. Telling people, "You have to test your thing endlessly, make sure it's perfect, and decipher error messages," felt misaligned with how we want to help people.

"Telling people, 'You have to test your thing endlessly, make sure it's perfect, and decipher error messages,' felt misaligned with how we want to help people." [12:40]

Finding Replay QA

Andres: So, lucky enough, just as we were starting to think about a more effective [13:00] and more independent way to help entrepreneurs QA and improve their apps, I got a LinkedIn message or email from Tom (I can't remember which) introducing me to Replay.

We'd done some early surveying of agentic QA solutions, and we were really impressed by Replay's thoroughness. Things that might seem trivial now, but at the time not everyone was doing: taking screenshots of the website and putting them in the report, understanding what was actually going on in the logs. It was clear there was an attention to detail that we needed to help the people who don't know what a console log is.

"It was clear there was an attention to detail that we needed to help the people who don't know what a console log is." [13:25]

Thomas: Right.

Andres: So we made the partnership, and we've been working together since.

From one agent to an agent store

Thomas: It's funny: when two companies get together, they start with initial ideas of how the pieces fit, but the end result is usually different. Talk to me about the evolution from knowing you needed to plug this in, to how it [14:00] shows up in the Audos platform today.

Andres: When we first started chatting, a lot of the Audos capability set was built natively into our flagship agent, Otto, which performed tasks itself through tools. If you know anything about the way LLMs work, as you give them more to do and ask them to remember more, performance degrades as the context window fills up.

Thomas: Oh yeah.

Andres: We were getting to the point where we were asking far too much of Otto. We knew we had to turn Otto into an orchestrator and delegator, rather than the agent that performed every action.

When we first engaged with Replay, we thought Otto would delegate tasks directly to Replay. But as we thought about the most effective way for Otto to delegate and work with other services, we realized that even if Otto [15:00] delegated to something like Replay, it would still need all the context: the Replay documentation, what kind of prompt to pass in. What Otto should actually do is delegate to specialized agents, built on the Audos platform, that each perform a specific function. Those agents engage with the third party.

So we started working on our agent store, which is basically an app store within Audos. But instead of traditional apps, it's things that work agentically, with any capability set. They provide capabilities you might expect to be native to Audos, but they're actually built by other entrepreneurs on the platform.

So we have entrepreneurs serving each other. It means actual experts on the subject build businesses for themselves, instead of someone on our team wearing many hats. And it creates a lot of good mojo. We have a very strong network and community, and people love to help one [16:00] another. We often find they're more motivated when they know their service is being bought by another Audos entrepreneur.

QA Guru, built on Replay

Andres: Long story short, Replay became the foundational back end of an agent in our agent store called QA Guru, which offers agentic QA to our entrepreneurs. They can describe an issue they're having, or QA Guru can just do a normal everyday check: "Let me make sure everything's working." Otto delegates the task to QA Guru, which has its own specialized prompting and has learned the best way to work with Replay. It calls the Replay API, gets the results back, and Otto can read all of them.

"Replay became the foundational back end of an agent in our agent store called QA Guru, which offers agentic QA to our entrepreneurs." [16:10]

Thomas: That's a real evolution, and it's fascinating: you build the platform, you realize the limitations of what you built, and from our perspective, you found new footing really quickly and turned Otto into the [17:00] orchestrator. It's been great to watch, from our side, how the Audos platform has evolved and how Replay has been able to enable it.

Even though QA Guru lives as an app in the agent store, my experience is that it feels like a first-party tenant. It has awareness of the underlying platform. It's not just firing off to some third party and hoping you get something back, then copying and pasting the output. It feels pretty seamless, which is great.

Tenant agents

Andres: That's well described, and it's awesome you used the word "tenant," because that's actually our internal name for these agents. We call them tenant agents.

The idea is to give each agent a very specific task and the right context. It's taken work to make sure QA Guru asks Otto for the right context, Otto passes the right context to QA Guru, and QA Guru passes the right context to Replay. But as long as you give each one a well-defined task, it works. [18:00] I'm glad to hear that, even though it's calling an external service, it feels very integrated into your Audos workspace, and that Otto is just talking to an agent in its toolkit. In the same way that delegating to any sub-agent in an LLM feels seamless.

"Even though it's calling an external service, it feels very integrated into your Audos workspace. Otto is just talking to an agent in its toolkit." [18:05]

Early results

Thomas: I know you're still in the early phases of rolling this out to the entire Audos community of builders. How's it been going so far?

Andres: It's been great. We've been really happy with the early reactions.

I wouldn't say I'm surprised, because it's everything we've seen from our network and community, but it's always a worry. People really jump at helping each other. If people come across a bug in any normal product, they're not going to be very kind in how they describe it. In our community, people are very willing to help, or to share [19:00] ideas and product suggestions.

And people seem to be unlocking a lot of value from these agents. QA Guru is saving them time. It's catching bugs they didn't realize they had.

"QA Guru is saving them time. It's catching bugs they didn't realize they had." [19:05]

We also have an agent that helps you do PR outreach. It frames that outreach for you and helps you get in front of journalists. We have an agent that helps you with SEO and AEO.

It feels like a very sustainable way to expand our capability set and continue the culture we want on Audos: a community and a network that serves each other. Not just a place to build product, but a support system, both technical and personal, for building a real business.

Layers of trust

Thomas: I love it. It seems like at the core of what you're doing are layers of trust that compound over time. I come up with an idea, you walk me through building it, I validate it with customers, and you help me QA it and [20:00] make sure it's the best it can be, instead of forcing me to go watch YouTube videos and run all kinds of third-party tools on it. That's a trust that's really hard to get otherwise. Almost anything you roll out, you've already established that handshake with the community.

A revenue share, not a subscription

Andres: It's the way our business model works. We take a revenue share instead of a subscription fee, and I think that builds a certain amount of trust. We only succeed if they do, and we're doing everything in our power to help them succeed. We definitely don't want them just churning tokens and renewing subscription after subscription. We feel very mission-aligned with our entrepreneurs.

"We take a revenue share instead of a subscription fee. We only succeed if they do." [20:20]

Thomas: That's a really ethical setup, and a win-win. It's funny, I'd almost forgotten the old days at the beginning of 2026, when it felt like all you'd hear about on social media was people feeling [21:00] like these app builders were internet slot machines designed to make them lose money and burn tokens. The fact that you took a different approach is awesome, and as a company doing business with you, it feels good to be aligned with that. We're really proud of this partnership.

Human insights powered by AI

Andres: We feel very lucky that it attracts a certain kind of person: the everyday entrepreneur who really wants to build in a community and be around other entrepreneurs. Every day, I feel lucky about the quality of the people we interact with, and I think that extends to partners we work with, like Replay.

Ultimately, to be in business with us, you have to believe in the mission of helping everyday entrepreneurs and helping human stories. We believe that human insights powered by AI is how the next generation of companies will be built. If you don't believe that and don't want to help [22:00] humans, it's hard to hang around with us.

"We believe that human insights powered by AI is how the next generation of companies will be built." [21:50]

Thomas: I think that's a great place to end. Andres, thank you so much for spending time with me today and talking about Audos and how Replay QA enabled this built-in QA experience. Good luck. Our luck is now intertwined, so let's go create our own.

Andres: Absolutely. A rising tide, man. You've been incredibly responsive and helpful, and we really appreciate having partners like you.

"You've been incredibly responsive and helpful, and we really appreciate having partners like you." [22:20]

Thomas: This has been great. Have a good one.

Bring Replay QA to your app builder platform

Want to explore how Replay QA can integrate into your app builder platform? Let's talk.

Have an startup idea you’ve always wanted to pursue? Build it with Otto on Audos.com