The one-liner
Any of the autonomous-company platforms will build you a live business in twenty minutes. None of them will find out whether anyone wants it. This guide puts the finding-out first, then the building, so that by Sunday night you have a company that is either selling or is a cheap, fast, honest failure.
The platforms' own numbers explain why the order matters. Polsia's founder says one company in ten on his platform has earned a dollar. NanoCorp publishes its leaderboard and the top company is in the low thousands. The agents build fine. The gap is demand, and demand is the part you do on Saturday.
By the numbers — September 2026
Saturday — validate before you build
The whole weekend hinges on Saturday. Skip it and you are the factory worker in the Rest of World story: a polished company, fabricated reviews, ad spend, zero customers.
Morning — pick one problem, not one idea
Write down a problem a specific kind of person has, that they already spend money or real time on. Not "an app for dog owners." Instead: "small dog groomers lose bookings because they answer the phone between clients." The narrower the person, the easier the next two steps.
Two tests before you continue. Can you name five real people who have it? Would they pay $20 a month, or $100 once, to make it go away? If either answer is no, pick another problem. This step should take an hour and it is where most weekends should end.
Midday — talk to five people
Message five people who have the problem. Not friends being kind — people in the group. A forum, a Facebook group, a subreddit, a WhatsApp community, a trade association, a LinkedIn search. Ask three things: how do you handle this now, what does it cost you, what have you tried. Do not pitch. Listen for the words they use; you will give those words to the agents on Sunday.
Three of five say it is a real pain and describe what they do about it: continue. Fewer: change the problem, not the plan.
Afternoon — write the one-page brief
Everything the agents will need, in plain English, in the customer's own words:
- Who — the person, precisely. "Independent dog groomers in the US with one to three staff."
- Problem — in their words, from the calls.
- Offer — the smallest thing that solves it. One feature, not a platform.
- Price — the number you tested. If you did not test one, $29 a month or $49 once are safe starting points for a small-business tool.
- Proof — one line about why you, or why now.
- Not this — three things the agent must not do: no fabricated testimonials, no cold email to people who did not opt in, no spend above the cap.
Evening — the pre-sale test
Before any platform, put up the cheapest possible test: a single page with the offer, the price and a button. A free page builder, a Stripe payment link, or even a Google Form that says "reserve your spot." Send it to the five people and to the group you found them in. If two people click through to pay or reserve by Sunday morning, you have a business worth building. If nobody does, you have saved a month.
Sunday — choose the platform and launch
Which platform
Three real choices, and the honest differences:
| Polsia | NanoCorp | Paperclip | |
|---|---|---|---|
| Setup | None | None | Real: a server, a database |
| Cost to launch | $20–50/mo + credits | Free trial, then $30/mo | $0 + model usage |
| Fees on sales | 3% + 20% on ad spend; $500/mo withdrawal cap | 20% on withdrawals | None |
| Control | Low; their infra, their agents | Low; full autonomy, no gates | Total; your agents, your rules |
| Transparency | Live platform stats | Publishes every company's revenue | It is your data |
| Best for | Fastest hands-off launch | Honest fast experiments | Anyone who wants to own it |
For a first weekend: Polsia or NanoCorp. You want speed and you want to spend your attention on customers, not servers. Paperclip is the right second company, once you know the idea works and want to keep the margin and the control. The full trade-offs are in each platform's own guide.
Morning — launch
Paste the brief. Do not press "surprise me"; that button is the platform choosing an idea with no demand signal and it is where the failed companies come from. Let the agents build the site, the product and the payment flow. Read what they produce against the brief. Fix the copy where it stopped sounding like your customers.
Set the caps before anything else: ad budget at zero for the first week, credits at the plan minimum, outreach off unless you have a list of people who asked to hear from you.
Afternoon — the first ten customers, by hand
The agents will offer to run ads and send cold email. Say no for now. Your first ten customers come from the five conversations and the group you found them in. Send the live link personally, with the same three questions you asked yesterday. This is the one part of an AI-run company that a human still does better, and it is the part that decides whether the company lives.
Evening — set the report
Every platform sends a daily summary. Tell the agent what you want in it: visitors, signups, payments, replies, and one question it needs you to answer. Then close the laptop. The point of the weekend was to reach this moment: a real company, a real offer, real people who have seen it, and a machine that will run it while you sleep.
The week after — decide honestly
Give it seven days and one rule: the company earns money or a clear "no" from real people, and either is a good outcome.
- Two or more paying customers: double down. Turn on a small ad budget with a hard cap, let the outreach agent contact people who match your first customers, and start a second week.
- Interest but no payments: the offer or the price is wrong, not the company. Change one thing, message the interested people, give it seven more days.
- Nothing: kill it. Cancel the plan, export what you can, and run the next problem through Saturday. A dead company after one week and $30 is exactly what this process is for.
Most people who "fail" at AI-run companies never reach this decision. They launch on Sunday, watch the agents work for a month, pay for ads that reach nobody, and quit at $200. Deciding on day seven is the discipline the platforms cannot give you.
The mistakes that kill weekend companies
Surprise me. The AI picks the idea. Nobody validated it. It builds beautifully and sells nothing.
Ads before customers. Ad spend is where the money goes on every platform and the fee structures reward it. Ten hand-won customers first; ads to find more people like them second.
Trusting the daily report. "Task complete" means the agent thinks it finished. Click the link. Send the test email. Buy your own product once.
Not reading the fees. A $500 monthly withdrawal cap or a 20% cut changes what "working" means. Know the number before the first sale.
Building the platform, not the feature. Agents will happily add a dashboard, an onboarding flow and an integrations page. Your customer asked for one thing. Ship that.
Waiting. A month of watching is not patience; it is avoidance. Seven days, then a decision.
Glossary
- Validation
- Confirming real people want a thing before building it. The Saturday of this guide, and the step every platform skips.
- Pre-sale test
- Offering a product for sale before it exists, to measure demand. A page, a price, a button.
- Brief
- The one-page document the agents build from: who, problem, offer, price, proof, and what not to do.
- Surprise me
- The platform option where the AI chooses your business idea. Avoid.
- Cap
- A hard limit on what an agent may spend: ad budget, credits, outreach volume. Set before launch.
- Withdrawal cap / fee
- The limit or cut a platform applies when you take out revenue your company earned. Polsia: $500 a month; NanoCorp: 20%.
- Daily report
- The summary every platform's agents send. Read it; do not trust it.
- Kill
- Ending a company that did not sell. Cheap, fast, and the normal outcome of a good process.
This is the how-to. Here are the platforms and the thinking behind them.