AI that runs your company raised $30M. Here's what its own numbers say.
Polsia's founder says about 10% of the companies on it have made a dollar. What that tells you about autonomous business in 2026.
An AI startup that promises to run your company while you sleep reportedly raised $30 million at a $250 million valuation this spring. Polsia says its agents plan, code and market a business every night, then send the founder a summary in the morning. The pitch is simple: type an idea, or press "Surprise me", and wake up to a company.
The interesting part isn't the round. It's the numbers the founder has shared about what happens to the businesses on the platform, because they say a lot about where "autonomous business" actually stands in 2026.
What the headline says
According to coverage of the raise, Polsia went from zero to $1 million in annual recurring revenue within 30 days of launch, reached close to $10 million, hosts more than 7,600 businesses, and is run by a single employee. Investors in the round reportedly include Sound Ventures and True Ventures.
That is a real business. People are paying for the idea of a company that runs itself.
What the businesses earn
The revenue figure is mostly subscriptions: what founders pay Polsia, not what their AI-run companies earn. When asked, the founder has reportedly said that about 10% of companies on the platform had made at least one dollar, that the best result he had seen was in the low thousands, and that roughly half of users leave in the first month. He has also said the product "doesn't work well enough for my taste."
Those figures come from interviews quoted in a review by a competitor, so treat them as indicative rather than audited. But they line up with what you'd expect, and with the platform's own advice: one setup guide describes it as about 80% autonomous, with strategy and judgment still left to the founder.
Why most AI-run companies don't make money
The pattern isn't really about AI. It's about where the business starts.
- An idea isn't demand. An agent can build a site, write emails and buy ads for an idea that nobody wants. Doing it faster doesn't change the outcome.
- Autonomy without limits is a risk, not a feature. The most common complaints in public reviews are tasks reported as done when they weren't, credits that run out faster than expected, and actions taken without asking.
- Nobody is measuring the right number. Platform revenue is easy to show. What each customer's business earned is harder, and much more useful.
What would make it work
If you're evaluating any "AI runs it for you" tool, three questions cut through the marketing:
- Does it start from something people already buy? Products with existing sales and real reviews, or a service with proven demand, give automation something to amplify.
- Can you cap what it spends? Daily limits, approval before ads go live, and a clear log of what it did.
- Will it show you outcomes, not activity? Orders, refunds and profit, not "tasks completed".
We're testing this ourselves. Arbi, the commerce engine we build, starts from products that already sell, keeps ad spend paused until you approve it, and its own store publishes a live order count on arbi.creai.dev, including the zeros. We'll report back here on what the numbers do.
Sources: Noqta on the funding round, Crevio's review quoting the founder (a competitor), Mika Reyes' setup guide.
