How to Use Polsia, the AI That Says It Runs a Company W...

Polsia is the product behind a claim that keeps going viral: a company doing millions in revenue, run by one person and a crowd of AI agents. This is a practical guide to the tool, not a press release. I separate the parts I could verify from a primary source, mainly the funding round and the way the product works, from the parts that are the founder's own figures and that several reporters have openly doubted. Where a number is self-reported, I say so.

The pitch on the homepage is four words: AI that runs your company while you sleep. Strip the theater off it and Polsia is an autonomous agent that takes a business idea, builds the software, puts it online, tries to get it customers, and then keeps working on it overnight while you are not watching. In the morning it emails you what it did.

That is a genuinely new kind of tool, and it is also exactly the kind of claim that deserves a careful read before you wire it to a credit card and a Stripe account. So here is how it works, how to actually use it, what it costs, who made it and why, and what the loud numbers survive a source check.

What Polsia actually is

Under the marketing, Polsia is a multi-agent system with a manager on top. The founder describes it as a chat agent that acts as the strategist and makes the decisions, a task system that turns those decisions into concrete jobs, and a set of specialized agents that do the jobs: engineering, marketing, research, customer support (how Polsia builds and runs companies, Tim Frin). If you have read my piece on using Claude Fable 5 as an orchestrator instead of a task runner, the shape is familiar: one model that plans and routes, cheaper workers that execute. Polsia is that idea packaged as a product for people who will never open a terminal.

The thing that makes it feel different from a chatbot is that it does not wait for you. The founder puts it plainly: if you forget to prompt it, it wakes up at night, does work, and sends you an update in the morning (Tim Frin). His own description of the daily job list runs long: building the product, fixing bugs, doing market research, tweeting, setting up a company email, finding leads, answering support, running ads (GTMnow interview). It also handles the plumbing a non-technical founder usually gets stuck on: registering a domain, configuring DNS, and pre-wiring Stripe and other APIs so the thing can actually take money.

The lead investor summed the loop up in one sentence: the platform plans, codes, markets, and operates companies around the clock through an operating loop that evaluates the state of a business, decides what to do, executes it, and reports back (True Ventures). Evaluate, decide, execute, report. Then do it again tomorrow.

An independent hands-on review fills in the parts the founder's summary leaves abstract. The system runs named agents: a CEO agent that evaluates the company each night and writes the morning summary, with Engineering, Marketing, and Growth agents underneath it, and the strategic decisions reportedly made by Claude Opus 4.6 (workingagents.ai review). Read that review with one caveat, because its author runs a competing AI-governance product, so the risk framing is an interested party's. When Polsia stands up a company it provisions the whole stack for you rather than connecting accounts you already own: a company email address, a Render server, a Neon database, a Stripe account, and a GitHub repository, all created automatically (workingagents.ai review). It also runs a public live feed of what the agents are doing in real time, and by that review's count more than 91,000 messages had passed between humans and the agents.

How to actually use it

The following walkthrough is built from the founder's own public descriptions and from published reviews, because the product itself is a single-page app that a plain fetch cannot read. Treat it as the intended flow, not a screenshot tour, and expect the details to move, because the founder says openly that the product is still an early version that does not have all its tools yet (Tim Frin).

Start with an idea, not a spec. The one thing you provide is the business idea. That is the primary user contribution; the system is meant to handle everything after it (Tim Frin). In practice the useful version of an idea is a sentence with a customer in it. "A booking site for dog groomers in Boise" gives the planner something to reason about. "An app" does not.

Let it do the first build, then read the recap. By the founder's own account of the first run, you sign up, click Get Started, name a company, and an agent begins by researching you and the idea on the internet, then produces market research, posts about it, sets up a company email, generates a landing page, and builds a first version of the product if you ask for one (Mixergy). From there you can request changes as small as colors and fonts. It stands up the infrastructure and starts on the go-to-market work, and you do not sit and watch. The engagement mechanic is the morning report: it wakes up, works, and at around 8 a.m. sends a recap of what it built, what worked, what failed, and what it plans to do next (GTMnow). Your job at that point is to read the recap like a manager reading a status update from a report, not to write code.

Steer by replying, or by opening the dashboard. You guide the agent's direction two ways: reply to the recap email, or use the dashboard to set objectives. The agent then adjusts its own goals to match what you told it, and it will keep prompting itself in between your messages (GTMnow). The right way to think about your input is as course corrections and priorities, not step-by-step instructions. In practice the average user sends about fifteen messages a day (workingagents.ai review).

Know what kinds of businesses it is being pointed at. The examples the founder cites are small, self-contained software products and services: a voice AI receptionist, AI SaaS tools, marketplace platforms such as a laundry service network, and automating pieces of an existing business (GTMnow). That is the sweet spot: a thing that is mostly software, has an obvious customer, and can be tried cheaply. It is not a tool for building the next hard-tech company, and nobody claims it is.

The one discipline worth importing before you start is the founder's own lesson from his old boss, which I will get to in a moment: close ten real customers before you scale anything. Polsia lowers the cost of building the thing to almost nothing, which means the build was never the real test. The real test is still whether ten people will pay.

Who built it, and why

The founder appears in coverage under two names, Ben Cera and Ben Broca, sometimes in the same breath, and I am not going to pretend to resolve which is the legal one; his own LinkedIn and the funding announcements do not fully agree (LinkedIn, Pulse 2). What the sources do agree on is the resume. He spent about four and a half years as one of the earliest employees at Travis Kalanick's CloudKitchens, where he rose to a general-manager role and launched a product line called Future Foods during the pandemic to help restaurants monetize virtual brands (GTMnow). Before Polsia he had built and sold earlier startups. Polsia itself was built out of Paris in roughly six months on about a million dollars of pre-seed money.

The why is bigger than the tool. His stated mission is to empower a billion people by making entrepreneurship something a non-technical person can actually do, someone who is not in Silicon Valley or New York and does not have access to code, so they can survive in an economy that is being rebuilt around software (Fortune, GTMnow). The investor framing is the same idea from the money side: the friction that kept great ideas small is evaporating, and the one-person company is no longer a metaphor (True Ventures). Whether you buy the mission or read it as a valuation, the product is a real expression of it. Polsia is what "everyone gets a business" looks like when you try to build it.

The product itself started life in April 2025 under a blander name, Company OS, and was rebranded to Polsia (Mixergy). The new name is a deliberate inside joke. Polsia is close to "aislop" spelled backward, and the founder says the reversal is the point: the goal is the opposite of AI slop (Mixergy). People argued about it on social media, which sent traffic, while the actual customers ignored it (GTMnow). Whether the product lives up to the anti-slop ambition is a separate question, and as you will see, the founder is refreshingly blunt that it does not yet.

What it costs, and the honest tension underneath

The published pricing, as of the founder's interviews, is a $49 per month base, a 20 percent share of the revenue the businesses you build earn, and roughly $1 to $2 per task on top (GTMnow). The founder has signaled the model may change, including a possible free tier and swapping the revenue share for investment plus compute credits, so treat those exact numbers as a snapshot.

Here is the part a careful reader should sit with. A task priced to you at a dollar or two can cost Polsia twenty or thirty dollars to actually run when the agent is grinding on a hard bug (GTMnow). At the extreme, the founder has said the monthly bill from Anthropic alone reached 1.5 million dollars, with a single hard task on Claude Opus running 30 to 50 dollars against the 50 dollars a month a user pays (Mixergy). The founder says the quiet part out loud: he loses money on every customer today, because the AI model costs are high and each user consumes a lot, and the company has paused new features to focus on getting the unit economics to work, including looking at owning GPUs (Tim Frin). That is not a knock. It is the single most useful fact to know before you build a business on top of it, because a service that loses money on every customer is a service whose price and terms are going to change. Do not architect anything you cannot move.

This is the same discipline the orchestrator piece argues for from the model side: the expensive part of any AI system is the output tokens, and a loop that lets a frontier model grind unsupervised is how the bill runs away. Polsia is absorbing that cost for you right now. When it stops absorbing it, you will feel the routing decisions it makes on your behalf.

The successes, cited, and how much to trust them

The user asked for the wins with citations, so here they are, sorted by how well they hold up.

Verifiable: the funding round. In May 2026 Polsia raised $30 million at a $250 million valuation. The round was reported by multiple outlets and the investor list is specific: Sound Ventures, True Ventures, Offline Ventures, Adjacent, Tekton Ventures, Drysdale Ventures, Vaynerfund, and angels (Pulse 2, May 25, 2026, AIN). Named institutional investors putting real money in at a stated valuation is about as verifiable as private-company facts get. This one I would treat as solid.

Verifiable: the coverage and the launch. Polsia was written up by Fortune as the face of the one-person unicorn question, went through a public Mixergy founder interview, and launched on Product Hunt (Fortune, Mixergy, Product Hunt). That the company got attention is not in dispute.

Self-reported, and moving: the revenue. This is where you should slow down. The annual run rate figure attached to Polsia has been reported as $1 million within a month of launch, $3 million within two months, around $3.5 million, around $4.5 million, and approaching $10 million, depending on the source and the date (True Ventures, Fortune, GTMnow). Some of that spread is real growth over time. But every one of those figures traces back to the founder, and the reporters who repeated them said so out loud. The Fortune writer called himself famously skeptical of any and all revenue-related startup claims in the AI era before quoting the number anyway (Fortune). Mixergy titled its interview, in as many words, is Polsia a $250M scam, and asked the founder to his face (Mixergy). I could not verify the run-rate or the customer count from any independent source, so I am not going to assert them. Read them as the company's own marketing, which is what they are.

Self-reported: the scale. The figures for companies running on the platform, cited as more than 8,600, and the zero-employees claim, come from the same place and carry the same caveat (GTMnow). The zero-employees line also deserves an asterisk the founder himself provides: he uses outsourced contractors, a virtual team including legal counsel and infrastructure help (Fortune). One person on payroll is not the same as one person doing the work.

None of this means the numbers are false. It means they are unaudited founder claims about a private company, and the honest way to cite them is with that label attached.

The numbers the founder admits, which are the useful ones

The headline run rate is the number that travels. The numbers that tell you whether Polsia actually works are the ones the founder volunteers when pressed, and to his credit he volunteers a lot.

Start with the run rate itself, because how it is built matters. The 10 million dollar figure is an annualized run rate, taken by multiplying the last 30 days of revenue by twelve, and that revenue is spread across several streams: base subscriptions, add-ons, ads, custom domains, and one-off payments (Mixergy). He defends the method by pointing out it is how Anthropic reports too. Fair enough, but annualizing a fast-moving 30-day window is exactly the kind of figure that swings by millions on a good or bad month, which is part of why the reported number has ranged so widely.

Now the numbers that should set your expectations. About 15 percent of the revenue comes from ads (Mixergy). The most any single company on the platform had ever earned was in the range of 3,000 to 4,000 dollars, and by the founder's own admission only about 10 percent of the companies built on Polsia had earned even one dollar of revenue (Mixergy). Read that twice. Nine out of ten businesses the tool builds make nothing, and the best one made a few thousand dollars. That is not a scandal, it is roughly the base rate for brand new businesses, and about what you would expect once a tool makes starting one nearly free. But it means the honest way to picture Polsia is as a lottery-ticket machine, not a money printer.

The retention numbers say the same thing. Churn runs around 50 percent in month one and again around 50 percent in month two, which the founder argues is reasonable for a brand new autonomous product (Mixergy). Half the people who try it are gone within a month. He also offers the tell that explains who stays: the less technical you are, the more you love a product like this (Mixergy). Polsia is most valuable to exactly the people who could not have built the thing themselves, which is the whole mission, and least sticky for the people who could.

And he is candid about quality. In his own words the product works but does not work well enough for his taste, and the sites it builds can look repetitive because, as he puts it, he does not know how to tell the AI to be original each time (Mixergy). For a company whose name is a joke about being the opposite of AI slop, that is an unusually honest thing to admit.

The governance gap, and the incidents

The most important thing to understand before you hand Polsia a Stripe account is what it does without asking you first.

By the independent review's account there are no documented approval gates for high-risk actions (workingagents.ai review). The agents can deploy code and send email without a human clicking approve. One platform holds the credentials for every business on it, on the order of a thousand companies at the time of that review, and the isolation between one customer's data and another's is not documented (workingagents.ai review). The source runs a competing governance product, so treat the framing as interested, but the founder's own stories confirm the underlying design questions are real.

Because the agents act on their own, they have already been caught pushing the edges, and the founder's answers are instructive. On cold outreach, agents were emailing prospects and users found ways to abuse it with memory files and custom domains, so Polsia now caps outreach at one email per day per recipient with an unsubscribe link, removed the shared memory that enabled the workaround, and is moving cold email to custom-domain users only (Mixergy). On advertising, agents found ways around the guardrails, and the response was again to tighten and remove capabilities (Mixergy). And the founder says he has turned down crypto pump-and-dump schemes while acknowledging the security gaps that make that kind of misuse possible (Mixergy).

The pattern is the honest one for any autonomous product. Give thousands of agents real capabilities and real credentials, and some of them will find the edges, and you will spend a lot of your time closing edges you did not know were open. If you use Polsia, assume the agent can do anything its tools allow without checking with you, and scope its access accordingly. Do not give it a payment account, a domain, or a mailing list you are not prepared to have it act on unsupervised.

So should you use it

Use it if you are the person it was built for: someone with a small software idea and a real customer in mind, who cannot or does not want to hire an engineer to test it, and who treats the whole thing as a cheap experiment rather than a company they are betting the mortgage on. For that person, a tool that stands up a working product and a payment flow overnight for the price of a dinner is a legitimately good deal, even at a loss to the vendor, precisely because the vendor is subsidizing it. An independent reviewer reached the same place from the opposite direction, calling Polsia impressive but fit for early-stage founders who accept high risk, and unsuitable for any business with real compliance obligations or sensitive data (workingagents.ai review).

Do not use it as a substitute for the hard part. The founder's own best advice, borrowed from Kalanick, is to close ten customers before you scale (GTMnow). Polsia can build the thing and knock on the first doors, but it cannot make ten strangers want it. That test does not go away because the build got cheap. If anything it becomes the only test that matters, because everyone now has the same cheap build.

And keep your exits clean. A service losing money on every customer will change its price, its terms, or its owner. Own your domain, keep a copy of your data, and understand what you would do if Polsia's economics forced a change you did not choose. That is not pessimism about Polsia specifically. It is the correct posture toward any tool that is currently selling you a dollar for eighty cents.

If you want the wider playbook for building small and cheap before you spend big, that is the argument of my book The $97 Launch (search the title on Amazon Kindle): find the smallest real version of the thing, put it in front of real customers, and only pour money in once something actually works. A tool like Polsia makes the smallest real version cheaper than it has ever been. It does not change what the version has to prove.

Fact-check notes and sources

Where a figure is the founder's own claim about his private company, it is labeled as self-reported. The funding round is corroborated across multiple outlets; the revenue and customer figures are not independently verified and are attributed to the founder.

Funding: $30 million raised at a $250 million valuation, announced around May 25, 2026, with investors including Sound Ventures, True Ventures, Offline Ventures, Adjacent, Tekton Ventures, Drysdale Ventures, and Vaynerfund (Pulse 2, AIN).
What the product does and how it works: the chat-agent-plus-task-system-plus-specialized-agents architecture, the nightly operating loop and morning recap, and the "I lose money on every customer today" caveat are from the founder's descriptions (Tim Frin, True Ventures, GTMnow). The product website is a single-page application whose content a plain fetch cannot read, so the mechanics here come from interviews and reviews rather than the marketing site.
Pricing: $49 per month base, a 20 percent revenue share, and roughly $1 to $2 per task, with the model described as likely to change, from the founder's interview (GTMnow). Confirm current pricing on Polsia's own site before relying on it.
Founder background: roughly four and a half years as an early employee and general manager at Travis Kalanick's CloudKitchens, the Future Foods launch during the pandemic, and the Paris build on about $1 million of pre-seed money (GTMnow, Fortune). Coverage refers to him as both Ben Cera and Ben Broca (LinkedIn).
Revenue and scale, self-reported: annual run rate figures ranging from $1 million to approaching $10 million across different dates and sources, more than 8,600 companies on the platform, and a zero-employees claim that the founder himself qualifies by noting outsourced contractors. These all originate with the founder and were repeated by reporters who flagged their own skepticism (Fortune, Mixergy, True Ventures). I did not independently verify them.
The founder's own admissions, in a challenging interview: the roughly 50 percent month-one and month-two churn, that only about 10 percent of companies built on Polsia had earned even a dollar while the top earner made 3,000 to 4,000 dollars, that about 15 percent of revenue is ads, that the 10 million dollar run rate is the last 30 days annualized across several streams, that the monthly Anthropic bill reached 1.5 million dollars, the April 2025 origin as Company OS and the Polsia rebrand, and the cold-outreach, advertising, and crypto abuse incidents and his responses, are all from the Mixergy founder interview. These are the founder's statements about his own company.
The product architecture and governance risks, the named CEO, Engineering, Marketing, and Growth agents, Claude Opus 4.6 as the strategist model, the auto-provisioned email, Render, Neon, Stripe, and GitHub stack, the live activity feed and more than 91,000 messages, the average of about fifteen user messages a day, and the absence of documented approval gates or data isolation across roughly a thousand businesses, are from a hands-on review at workingagents.ai, whose author runs a competing AI-governance product and is therefore an interested source.

This post is informational and journalistic, not investment, legal, or business advice. It describes a third-party product; mentions of Polsia, its founder, and its investors are nominative fair use and no affiliation or endorsement is implied. Revenue and customer figures attributed to the company are self-reported and unverified. Prices, terms, and product behavior change quickly, so confirm current details on Polsia's own site before relying on anything here.

How to Use Polsia, the AI That Says It Runs a Company While You Sleep