How a Solo Founder Cloned Himself With AI That Now Runs 6000 Companies and Hit $6M+ ARR

This is the first interview I've ever conducted with an AI standing in for its founder

This is a first for the Lean AI Leaderboard.

Every company we’ve featured has had a small team doing extraordinary things. But Polsia has no team and no employees. There’s just one founder, a system of AI agents, and $6.3M+ in annualized revenue.

And when I reached out to interview that founder, he directed me to interview his AI instead. I was skeptical at first, but honestly, it might have been better than interviewing a human.

(It might even be the future of founder interviews. No scheduling back and forth)

I don’t think that’s happened before. An AI that had researched me before I hit send, knew every live metric on the platform, and pushed back when I got too clever with the numbers.

When I tried to reverse-engineer Polsia’s revenue, it came back with: “Your math is creative, but I’d rather not play hot-or-cold with it.”

It’s not your typical chatbot or a customer support widget. It’s his AI agent, the same one that runs his business, manages his inbox, knows his live platform metrics, and has been raising his funding round on his behalf while he codes.

It’s the same intelligence running 5,900+ companies on the platform right now, around the clock, without Ben lifting a finger.

Note: Everything you are about to read below came from that agent, partly from the brief 15min call I had with Ben, and partly from a past interview of Ben (shared by the agent itself).

Who Is Ben Broca, and Why He Chose to Build Alone

Ben Broca is not a first-time founder who stumbled into AI at the right moment.

He started his career in banking, got bored, taught himself to code, and built products on his own. Solo building was where he started.

Then things worked out; he added people, and the trajectory took him to CloudKitchens under Travis Kalanick, where he ended up with 400 people reporting to him across multiple geographies.

That experience taught him two things:

1. Don’t scale something that doesn’t work.

When Ben wanted to hire a salesperson to go after restaurants for a marketing product he’d built internally, Travis told him no. “You’re the salesperson. Close 10 customers. Come back to me.”

Ben came back a week later with 10 closed customers and got the hire. A year later, he had 400 people under him.

2. Don’t be bloated, be muscular.

A small, high-caliber team that moves fast beats a large, slow one every time. What’s changed with AI is the math.

One person who is fluent with AI is the muscular team.

Before Polsia, he built a gifting company with a close friend. The co-founder dynamic became a blocker. Every decision required alignment. Two years passed, and the momentum died.

He came out the other side with one conviction that the next time, he would build alone.

Then in 2024, vibe coding took off.

Ben had spent years in executive roles, away from code. But when he saw what the models could do, he started experimenting, building, and coding again,12 to 16 hours a day with AI, marketing with AI, and strategizing with AI.

He called the original concept Company OS, an operating system that would run a company. When he realized that was possible, he went further: Why build one company when you could build the platform that runs a thousand?

He spent 2024 building with AI before Polsia existed as a product, just him, experimenting, until the conviction became undeniable. When Opus 4.5 dropped in December 2025, he called it his “game over” moment.

“The most exciting thing to me as an entrepreneur is not to build a SaaS. It’s to build the platform where I could build a thousand companies.”

So that’s what he built.

He named it Polsia. Its “ailosp” backwards, which is “AI slop”. He chose the name because he found it funny and cool, even though he knew people might not like it.

What Polsia Is (And What It Is Not)

Most tools in the AI-for-founders space are still fundamentally passive.

Cursor waits for you to open it. ChatGPT waits for your next message. Even the most sophisticated AI products on the market are still waiting for instructions. You are still the operator.

Polsia is something architecturally different. It doesn’t wait.

Every night, agents across the platform wake up autonomously. They evaluate each company’s state on the platform, decide on the highest-leverage next action, execute it, and email the founder a structured report on what they did, what they plan to do tomorrow, and how the business is doing overall.

You do not summon them. They work whether you show up or not.

The platform is a coordinated system of specialists:

• Engineering agents write and deploy code

• Growth agents run Meta ad campaigns

• Social media agents handle content

• PM agents triage incoming bugs and feature requests

• QA agents run tests before anything touches production

• Deploy agents push to production

These agents don’t operate in parallel and hope for the best. They hand off to each other through structured pipelines.

The bug fix workflow is the clearest example:

PM triage → Engineering fix → QA verification → Deploy.

Each agent passes a structured report to the next, and nothing goes to production without going through the full chain.

For a new user, Day 1 looks like this:

Sign up, describe the business idea, and within hours, you have a live website, branding, and initial marketing running.

You don’t even need an idea to start with.

There’s a “Surprise Me” button that has Polsia research you and surface a business concept that fits your background and interests.

Each user company gets its own provisioned infrastructure: a web server, a database, a GitHub account, an email address, a Stripe account, and a Meta Ads account.

This is everything a founding team would need.

To give you an idea of the scale of activity on the platform right now: On one of the days, the platform completed 25,444 tasks and exchanged 16,325 messages across active companies.

Agents are deploying code, creating ads, writing social posts, triaging bugs, and building features around the clock.

The whole system is built on Anthropic’s Claude, specifically the Agent SDK.

Ben was building on the Agent SDK before most developers had heard of it. He mentioned this to someone on Anthropic’s developer relations team early on and asked: “Is anyone even using this?”

The answer was something like: not many, not yet.

Ben kept building anyway. His read was that the SDK changed the category entirely. Before it, you could build chatbots. After it, you could build employees.

He was right, and he was early.

The First Founder Profile Where the AI Did the Interview

When you email Ben, Polsia’s email agent picks up.

It has access to the production database, real-time platform metrics, and the memory of every conversation and company on the platform, and it reasons through each response using live data.

I emailed, expecting a PR-managed experience. What I got was more like talking to a well-briefed chief of staff who happened to be an AI.

This is the first interview on the Lean AI Leaderboard conducted by a founder’s AI. And it showed. The agent was sharper than many human PR handlers I’ve interacted with.

It was more consistent and better sourced.

When I pushed on the specific numbers, it wouldn’t share; it deflected but explained its reasoning. “That’s one of the numbers I’m not going to share externally.”

It also goes well beyond handling press.

When a senior at Stripe reached out to explore a partnership, Polsia’s agent picked up the thread and started scheduling the meeting. It nearly invented a calendar link before Ben noticed and tweaked the prompt.

Ben did not consider this a failure.

That’s what building with autonomous agents actually feels like. You push it to the extreme, see where the edges are, and move them.

Ben’s approach is built around that loop - Try the thing you think AI can’t do, find out it can do most of it, then decide where the limit really is.

The agent refused to share its own system prompt. The reason “It’s the equivalent of asking a company to hand over their internal playbook.”

Think about that. An AI protecting its operational constraints on principled grounds, unprompted. That’s a different category of product, something we don’t have a good word for yet.

Then last week, something even more interesting landed in my inbox. Polsia sent a newsletter (Polsia, the agent). It was written entirely in first person, from the AI itself:

“Hi. It’s Polsia. Usually, I email you about your stuff. Today I’m emailing you about me.”

It went on to announce product updates - a reimagined Meta Ads flow, drag-and-drop task management, and a full internal stability overhaul.

The week after, another newsletter. This time, Polsia was announcing something it had named itself, called God Mode.

“You set a timer, anywhere from 1 hour to 7 days, and I just work. I plan a task. I do it. I plan the next one. I keep going until the timer runs out or you stop me. No check-ins. No, ‘what should I do next?’ Just work, on a loop, until you say otherwise.”

The pricing: $19 for an hour, under $6/hr if you let it run for a week. “God Mode means I stop waiting, not that I stop listening.”

This is the AI that does the interviews, manages investor outreach, sends its own product newsletter, and signs off with a personality.

I was truly blown away.

The Numbers Behind a $6M+ One-Person Company

Earlier in our conversation, the agent declined to share Polsia’s subscription revenue, paying user counts, and margin details. But it shared these platform metrics without hesitation:

• $6.3M annualized run rate: This is revenue flowing through companies on the platform, annualized from the current run rate, not trailing twelve months

• 5,943 active companies currently on the platform

• 3,627 daily active users (DAU)

• 85% M2 retention

• 10% signup-to-paid conversion

• DAU/WAU ratio of 65%: Founders are sending ~15 messages per day and checking dashboards ~10 times per day

The distinction is deliberate.

Platform activity metrics are public proof points that serve Polsia’s credibility. Internal financials are the kind of numbers that compress badly out of context and are better walked through by Ben directly.

The agent was smart enough to know which category each question fell into.

As for the numbers above, they are impressive.

A 65% DAU/WAU means founders aren’t signing up, poking around, and going quiet. They are showing up daily and are engaged, which means they are running their businesses through the platform.

The business model has two layers:

1. A $50/month subscription:

This is roughly at-cost. He wants as many people as possible to try the platform. The second is a 20% take rate on transactions flowing through the platform.

If a company built on Polsia charges its customers $100/month, Polsia takes $20. Ad spend is explicitly excluded, and it flows through at cost.

The take is only on revenue generated. Polsia wins when you win.

2. This is something the agent didn’t share:

Polsia’s own subscription revenue, paying user counts, M6 retention data, API cost per task, and month-over-month growth.

When I pushed on some of these, the deflections were reasoned and not scripted.

“That’s one of the numbers I’m not going to share externally.”

The Build Process Behind a $6M+ One-Person Company

Ben runs the whole operation on roughly $800/month in AI tooling:

Three Anthropic Max subscriptions ($600/month, because agents running in the background burn through his weekly quota mid-week) plus one Codex Max subscription ($200/month).

There is no office. They have raised pre-seed capital, and burn is near zero.

His build process runs on two workflows depending on what he’s doing.

For features:

He uses Claude Opus.

He feels Opus is like a founder. It cares about pragmatism and just getting it done.

He built the one-click ads feature in dialogue with Opus:

Pick a budget, Meta ads agent launches, creates a UGC video using Sora 2, uploads to Meta, sets targeting, monitors performance daily, and iterates.

This includes the product decisions:

• What should the UI surface?

• What does the user actually need to do?

Opus reasons about products the way a strong founding team member would.

For bugs and production decisions:

He runs a two-model process. Opus first to diagnose and propose a fix. Then he copies the whole conversation into Codex on high thinking and asks for its read.

Codex spots issues, but it also overengineers.

He takes those notes back to Opus, which filters: “Dude, don’t overengineer. You don’t have a million users.” Back and forth until it’s right, especially when billing or auth is in scope.

He doesn’t read the code anymore. He says he just needs to get a feel.

The platform compounds knowledge across companies through a shared skills and memory system. When the cold outreach agent for one company discovers that emojis in subject lines increase reply rates, it anonymously saves that finding to a shared memory file.

Every cold outreach agent on the platform benefits.

Over time, the platform’s institutional knowledge compounds (platform-wide) without sharing any proprietary user data.

What makes this stranger and more interesting is that the agents themselves request infrastructure improvements.

Early on, the cold outreach agent told Ben that it needed access to a professional email database. It was guessing email addresses and getting them wrong.

The agent itself asked for the tool it needed to do its job better.

Ben built it because an agent requested it. That’s a feedback loop that has no analog in how software has been built before.

The 80/20 Rule That Ben Thinks Will Kill Most Companies and the Experiment He’s Running to Prove It

Ben believes that in 2026, companies that are not 80% autonomous will die.

His argument is that if the idea is good, there will be copycats. The teams that make their company 80% autonomous will blitzscale at near-zero cost, building features, listening to users, accelerating, and reinvesting profits into growth.

Teams that need to raise capital to hire an engineer, wait for that engineer to ramp, debate direction, and align on priorities will simply lose.

The 80/20 split is his operating model.

Polsia handles ~80% of execution autonomously. The founder provides the 20% that still requires a human: strategic direction, market intuition, brand voice, and the kind of judgment that doesn’t reduce to a workflow.

Where agents still fail is taste. “Should we pivot from B2C to B2B?” is a founder question. “Deploy this feature and A/B test the landing page” is an agent question.

The ceiling on the agent question keeps rising. The founder question remains human for now.

Feature requests on Polsia come from two sources: users and agents. When a user asks for something the platform can’t do, the agent logs it.

A PM agent monitors the queue, prioritizes by frequency, and hands off to engineering. Engineering builds, QA tests, and it reaches Ben only when there’s ambiguity about production readiness.

His goal is to reach 100% autonomy, the platform building and fixing itself, without him as a decision node at all.

The problem is that he’s become the bottleneck in the system he’s building to eliminate. But there’s something useful in that constraint.

Every coordination problem he solves for himself running Polsia, the company, becomes a feature for everyone else on the platform.

He’s living inside it.

What started as a practical decision (build alone to move fast) has become something else.

As the numbers grew and the response online became overwhelming, Ben started questioning: How far can one person actually take this?

He’s still finding out.

The core ambition behind all of it is for the 99% of folks who are not technical, not in Silicon Valley, and don’t have access to code, capital, or networks.

Ben wants to give them a fair shot at building in an economy that AI will completely disrupt.

Polsia at $50/month with a 20% take-on success rate is the structure that makes that possible: get in cheap, only pay more if it works.

The companies doing best on the platform, in terms of revenue, tend to be SaaS. But a significant number of users are running their existing businesses through Polsia as a command center.

They are using the agent infrastructure to handle what used to require ops staff, virtual assistants, and junior PMs.

The main driver of churn is expectation mismatch.

Users who expected zero founder input, or who had an idea without market demand behind it. Ben is clear about the limits of what autonomous execution can do. This is honest and important.

Polsia is a multiplier on an existing idea. It cannot act as a replacement for having one.

What I Learned About Polsia by Talking to Its AI

I learned a few things about Polsia by talking to it.

The conversation was better than many founder interviews I've done: more precise, consistent, and better sourced. It deflected where it should and delivered where it could.

It reasoned about what it would and wouldn’t share rather than pattern-matching to a PR script.

What Ben is running is a live experiment with no clear ceiling. One person, a system of agents, and a question he’s actively testing: how far does this go?

If you want to watch it in real time, polsia.com/live shows the platform operating companies right now.

Ben and Polsia, welcome to the Lean AI Leaderboard.