AI-built startups: what the evidence really shows
Solo founders and small teams have built real products with AI coding tools, and a few have sold or scaled them. Ten of those stories are on our Cases pages. Put side by side, they show which numbers can be trusted, which are only claimed, and why the winners are a small, loud minority.
The short version
- Two of ten cases are rated High evidence. Both are things a third party could check: a sale price confirmed by the buyer (Base44) and a security finding (Moltbook). Five are Medium and three are Low.
- Revenue numbers are almost always the founder's own. They are relayed by press and not audited.
- The big wins are not tiny teams any more. Base44 had employees when Wix bought it, and Lovable has raised hundreds of millions.
- These ten are exceptions. Broader data suggests most AI-built apps earn little or nothing.
The ten cases at a glance
Each row links to the full case with its timeline and a list of what the sources say.
| Case | The headline claim | Evidence |
|---|---|---|
| fly.pieter.com | Free multiplayer flight sim, billboards sold to sponsors. $72K a month reported within two weeks. | Medium |
| TrustMRR | Database of Stripe-verified revenue, built and launched in 24 hours, with sidebar ads from day one. | Low |
| LaunchFast | 48-hour demo for a coaching company. About $20K MRR after 90 days, as reported. | Medium |
| DualShot Recorder | Paid iOS app. #1 on the paid chart within 12 hours, according to its creator. | Low |
| OpenClaw | Open-source agent. 145K+ GitHub stars by early February 2026. | Medium |
| Moltbook | Social network for AI agents. The creator says he wrote no code, and a security firm found its database exposed. | High |
| Lovable | Vibe-coding platform. $50M ARR in six months, company-reported. Later: $600M run-rate (Sep 2026). | Medium |
| Vibecode | Build apps by describing them on a phone. $9M to $9.4M seed round. | Low |
| Claude Cowork | Desktop agent built mostly by Claude Code in about a week and a half, per Anthropic staff. | Medium |
| Base44 | Sold to Wix for $80M in cash six months in. | High |
The ratings come from our case pages. High means independent reporting confirms the key facts. Medium means reputable outlets relay what the founder says. Low means the figures trace back to a single secondary source.
What the stories share
In the three cases we read most closely (fly.pieter.com, Base44 and Lovable), distribution mattered as much as code.
- Building in public. Pieter Levels turned each iteration of his flight sim into content, and Maor Shlomo grew Base44 on LinkedIn and X.
- A product you understand in seconds. The flight sim needs no signup. Lovable turns a text prompt into a working app.
- An audience or a channel already in place. Our fly.pieter.com case notes that the founder's existing following did much of the work, and that most builders do not start there.
Why these are the exceptions
Survivorship bias is the main thing to keep in mind. One vendor study, BigIdeasDB, reports that only 7.6% of apps still hosted on Lovable, Replit, Vercel or Netlify subdomains show any monthly revenue, and none of them clears $1K. That is vendor research, so read it as a signal rather than a census. 404 Media framed the fly.pieter.com story in March 2025 in the same spirit: its headline pairs the game's success with the odds that others will not repeat it.
Claims worth checking before you repeat them
These come up often in roundups. Here is what each one rests on.
| Who | Claim | Source |
|---|---|---|
| Marc Lou (ShipFast and related products) | $1.5M+ ARR | Reported in a May 2026 roundup, Ascero AI News. Founder-reported. |
| Danny Postma | $1M+ ARR on a Lovable and Cursor stack | Same roundup, which does not name the product. |
| Cameron Trew (Kleo and Mentions) | $62K MRR within about three months | Reported in an Indie Hackers post. |
| Habit Pixel (solo developer) | $1K MRR after eight months | Reported in an Indie Hackers post. A smaller outcome, and closer to typical. |
How to check a success story yourself
- Find who says it. A price confirmed by the buyer is stronger than a post by the founder. A figure repeated by ten blogs is still one claim if they all trace back to one post.
- Ask what the number is. Revenue, profit, ARR, MRR and run-rate are different. Ask over what period it was measured.
- Check the team. "Solo" often describes the first months. Look for headcount at the time of the sale or funding round.
- Separate the platform from the apps. Lovable's $600M run-rate is the platform's revenue. It says nothing about how much the apps built on it earn.
- Look at what is public today. Ads in the Meta, Google and LinkedIn libraries, pricing pages and how a site changes over time show what a company is doing now, not what a headline said once.
- Look for the counter-case. Moltbook is rated High because a security firm found its database exposed, not because of a revenue claim.
Sources
This page draws on the fly.pieter.com post by Pieter Levels, 404 Media on the same game, Globes on the Base44 sale and TechCrunch on Lovable's revenue. The BigIdeasDB figure and the roundup claims are cited as published. Everything here is as of November 3, 2026. For the full detail on each product, open its case.
Questions
Is ARR the same as revenue?
No. ARR is a monthly or short-period number multiplied up to a year. fly.pieter.com's "$1M ARR in 17 days" headline annualises a short window, so it is not money received. Check whether a figure is revenue, profit or run-rate, and over what period.
Were these products built by one person?
Often not. Base44 is the best-documented case: reports differ on its size, with TechCrunch (citing Wix) saying eight employees and Globes saying six, but neither calls it a one-person company at sale. Lovable has raised over $700M. "Solo" usually describes the start, not the finished business.
Why do figures for the same product differ between sources?
Many figures come from the founder's own posts. Outlets read them differently, some as monthly revenue, some as a total over a period and some as profit. When two sources disagree, we say so on the case page and do not pick one.
Does a successful AI-built product mean AI-built products usually succeed?
No. These ten are the visible exceptions. The wider data we found points the other way (see the section on survivorship above).