The one-liner
Polsia is a subscription platform where you describe a business idea — or let the AI pick one — and a team of AI agents builds the website, writes the code, runs the marketing, sends the outreach, answers support and reports back to you every morning. The founder's tagline: "AI that runs your company while you sleep."
It is also the most-argued-about product in the agentic economy. Investors valued it at a quarter of a billion dollars. Paying customers gave it one star in large numbers. Both of those things are true at the same time, and understanding why is the point of this guide.
By the numbers — September 2026
Everything in this box changes weekly. Polsia publishes a live public dashboard, which is how these figures can be checked. Everything below the box is written to still be true when the numbers have moved.
Who built it
Polsia was built by Ben Broca, who also goes by Ben Cera. He is French, trained as an engineer at CentraleSupélec with a masters from Columbia, and now lives in San Francisco. Before Polsia he ran international operations at CloudKitchens under Travis Kalanick and co-founded Hutch, a 3D-rendering e-commerce startup that raised roughly $17 million.
He spent most of 2025 building separate AI tools — a coding tool, a marketing tool — before concluding the models were good enough to run every business function at once. Polsia launched in late 2025, crossed $1 million in annualised revenue within about a month, and tripled the month after. He has been the only employee throughout, and says that is deliberate: if he is selling an AI that runs companies, Polsia should be run by one.
In May 2026 he raised $30 million at a $250 million valuation, reportedly letting the AI manage the data room and brief investors. That is the number the headlines quote. It says something about what investors believe. It says nothing about whether the product will work for you.
How it works
You give Polsia an idea, or click "surprise me" and it picks one from your public profile. Within minutes it has a company name, a mission statement, a market research document, a live website on a polsia.app subdomain, an email inbox, a Stripe account and a GitHub repository. Then it asks for a credit card.
Once you subscribe, the agents run on a daily cycle:
| Agent | What it does |
|---|---|
| Orchestrator (CEO) | Reviews the state of the company, writes the morning plan and the evening summary |
| Code | Ships features and fixes to the website and product |
| Marketing | Drafts and posts social content, writes landing pages |
| Outreach | Finds prospects and sends cold email |
| Support | Reads the inbox and drafts replies |
| Ads | Runs and optimises Google and Meta campaigns from your budget |
| Finance | Syncs Stripe revenue, tracks spend |
| Research | Watches competitors and refreshes the plan |
Each task the agents run costs a credit. Your plan includes a monthly allowance; you buy more if you run out. You read the summaries, approve or redirect, and the cycle repeats whether or not anyone is buying what the company sells.
The most important thing to understand about how Polsia works is what it does not do. There is no step between "here is the idea" and "the agents are building it" that asks whether anyone wants the thing.
What it actually costs
Polsia does not publish pricing on its homepage. As of August–September 2026, the in-product pricing showed five tiers:
| Plan | Price | Includes |
|---|---|---|
| Standard | $20/mo | 1 company, 20 credits/mo |
| Pro | $50/mo | 3 companies, 50 credits/mo, Meta Ads |
| Max | $200/mo | 10 companies, 200 credits |
| Max 2 | $500/mo | 25 companies, 500 credits |
| Max 3 | $1,000/mo | 50 companies, 1,000 credits |
The subscription is not the whole bill. Three more lines matter, all from Polsia's own Terms of Service:
- Ad spend fee — 20%. Any advertising budget the agents spend on your behalf carries a 20% platform fee.
- Revenue fee — 3%. Polsia takes a cut of customer payments your company receives. In the September 2026 terms this is "currently 3%"; until then it was 20%. Check the live terms — it has already changed once.
- Withdrawal limits. Your company's revenue lands in a Polsia balance first. Payouts have a 31-day settlement period, a $50 minimum and a cap of $500 per calendar month. A company earning $2,000 a month can take out $500 of it.
Users have also reported the credit meter running on a single click with no confirmation, and credits burned on tasks that failed. Budget for the plan, the credits, and the fees — then read the withdrawal cap twice.
The numbers — what the live feed actually says
Polsia publishes a public dashboard feed with enough raw data to rebuild the headline figure yourself. On 15 September 2026 it read $18.7 million ARR.
That number is built the way many AI startups build ARR in 2026: take the last thirty days of revenue, including one-off payments, and multiply by twelve. Ben has described the method himself on the record. Broken down, roughly $11 million is annualised subscription revenue. The rest is the last month of credit packs, ad budget routed through the platform, domain purchases and a small slice of user-company revenue, each times twelve. Real money, but around 40% of it is not recurring by construction.
The second number on the feed is the one that matters more for you. Nearly half a million companies have been created. About 29,000 are active. Companies are cheap to spin up here, so a low survival rate is expected rather than damning — but it is the clearest public signal of what happens after the agents finish building.
Ben gave the third number to Mixergy's Andrew Warner in May 2026, in an interview titled "Is Polsia a $250M scam?": about 10% of companies on the platform have made at least a dollar. The best one he had seen was making three or four thousand a month. Month-one churn is around 50%. He also said the engineering agent was writing "spaghetti code" and the product "doesn't work well enough for my taste."
That is unusual candour from a founder at a $250 million valuation, and it is the honest frame for the whole product: the agents really do build things. Nine times out of ten, nobody buys them.
What paying customers say
Polsia's Trustpilot page tells the story in three acts. In June 2026 it had 35 reviews. By mid-August it was 1.8 out of 5 across 79 reviews, three quarters of them one star. Polsia then claimed the profile, over a hundred reviews arrived in four weeks, and by mid-September it stood at 2.9 out of 5 across 184 reviews, with 45% one star and 27% five star. Several of the new five-star reviews describe a problem the founder fixed personally.
The complaints did not change across that period. The recurring themes:
- Tasks marked "complete" by the AI that never actually deployed
- Credits consumed by duplicate or failed actions, with refunds refused
- Cold emails sent to real contacts, or public posts made, without approval
- Websites and custom domains locked to Polsia's own hosting account
- Sites redeployed after an explicit takedown request
- Cancellations that took repeated attempts, and support tickets unanswered for weeks
The most-cited case is a Rest of World profile from April 2026: a factory worker in China paying $199 a month — a quarter of his salary — for an AI to sell a spiritual guidance app to Americans. The agents built the site, filled it with fabricated reviews, ran Facebook ads and emailed journalists he had never heard of. After several months: seven signups, zero customers.
The positive cases are real too. Founders who arrived with a validated idea and a clear brief report the agents standing up infrastructure, research, outreach and even video ads in hours — plumbing that would otherwise eat a week. One four-star reviewer who built a real product on it put the line exactly: the weakness is not what the agents can build, it is what the customer cannot control when the surrounding infrastructure fails.
The lock-in question
This is the part most first-time users miss. Polsia deploys your company on Polsia's hosting, on a polsia.app subdomain, with email routed through Polsia. There is a "download code" button, but users report the export is a scaffold rather than a working product, and that there is no direct access to the hosting or repository. Self-hosting is listed as "coming soon."
If the platform is working for you, none of this matters. If you want to leave, take your domain with you, or hand the code to a developer, it matters a great deal. Verify the current export and account-recovery terms in writing before you subscribe.
Who Polsia is for
Polsia is an execution engine. It is not an idea-validation engine, and it does not pretend to be one.
It fits if you:
- Already have evidence people want the thing — conversations, a tested price, a channel that works
- Are launching cheap, fast experiments where one flop costs little
- Are happy paying the fees and living with the withdrawal cap in exchange for not touching any of the plumbing
- Do not need to own the code or the infrastructure
Skip it if you:
- Have an idea but have not talked to a single potential customer
- Are in a regulated, technical or high-stakes category
- Need to withdraw more than $500 a month, or would be hurt by 20% on ad spend
- Plan to migrate off later, or want portable code from day one
The "surprise me" button is the most marketed feature and the riskiest one. It means the AI is choosing the idea and executing it with no demand signal anchoring either decision. The factory worker in the Rest of World story is what that looks like in practice.
Our verdict
Polsia is real, funded, and built by a credible operator. The agents genuinely build, market and operate companies with no human doing the work, and the platform has grown faster than almost anything in the category. The hype question is settled.
The question that matters now is fit. Polsia executes whatever it is given. Hand it a validated idea and a precise brief and it can save you weeks. Hand it a guess and it will build a polished company nobody asked for, spend your credits and your ad budget promoting it, and report back cheerfully every morning while nothing sells. The founder's own figure — one company in ten ever earns a dollar — is that gap measured from inside.
Validate first. Then decide whether Polsia is the right layer to execute on what you found.
Glossary
- Polsia
- Subscription platform where AI agents build and run a company for you. Founded by Ben Broca (Ben Cera), launched late 2025, San Francisco.
- Credits
- Polsia's unit of work. Each task the agents run costs a credit. Plans include a monthly allowance; extra credits are bought in packs.
- ARR (Annual Recurring Revenue)
- Annualised recurring revenue. In much of the 2026 AI industry, including Polsia's headline figure, it means the last thirty days of all revenue multiplied by twelve.
- Run rate
- A recent period's revenue extrapolated to a full year. Polsia's headline number is a run rate.
- Platform fee
- Polsia's cut. 20% on advertising spend and, as of September 2026, 3% on customer payments (previously 20%).
- Withdrawal cap
- The limit on how much of your company's revenue you can take out of your Polsia balance: $500 per calendar month, after a 31-day settlement period.
- Surprise me
- Polsia's option to let the AI choose your business idea for you.
- Validation
- Checking that people actually want a product before building it. The step Polsia does not perform.
- Churn
- The share of paying customers who cancel in a given period. Polsia's founder puts month-one churn at around 50%.
- polsia.app
- The subdomain every Polsia company is deployed on. Your company's site and email run through Polsia's infrastructure by default.
Polsia is one piece of the agentic economy. Here is the rest of the picture.