On 1 March 2014, a Dutch developer named Pieter Levels published a post titled I'm Launching 12 Startups in 12 Months. The rule was simple and brutal: pick an idea, build it, launch it publicly, one per month, and judge whether it had a market within that month.
His stated reason was not ambition. It was a diagnosis: "We creatives have one common problem: finishing things."
Ten years later, Levels posts his revenue on his own site. In March 2024 he reported PhotoAI at "$105,000/mo revenue and 80,000/mo profit." In September 2024 he posted that he had "hit a new $420,000/mo revenue record thanks to the Lex Fridman podcast." He has no employees and no investors. As he told Lex Fridman: "I don't really raise money. I don't use VC funding, I do everything myself."
Before going further: every number in that paragraph is self-reported. There are no filings, no audit, and no third party confirming them. That caveat is not a footnote, it is the correct frame for reading the whole story.
What the twelve-month rule actually did
The products from that year, listed in his own post, are a mix of the trivial and the durable: an inbox music collector, a goal-accountability tool, a YouTube analytics product, Nomad List, a job board for nomads, a printed animated-GIF flipbook service, a chat community, and Remote OK.
Two of those, Nomad List and Remote OK, became businesses. The rest did not, and that is the point of the structure rather than a failure of it.
Three things about the constraint are worth copying.
The deadline was public. Levels describes on the Lex Fridman podcast posting each launch to Hacker News, which turned a private intention into a commitment other people could see him break.
The unit of work was a launch, not a product. A month is too short to build anything substantial, which forces the scope down to whatever can be finished. Most founders set the scope first and let the date float; this inverts it.
The verdict was fast. Assessing market fit within a month means a failed idea consumes a month, not three years. Twelve attempts at one month each is a fundamentally different risk profile from one attempt at twelve.
His line in the 2014 post is the whole argument compressed: "By doing nothing, you figure out exactly nothing."
The payment button as the test
The mechanism that makes rapid launching useful rather than merely busy is what Levels attaches to each one. In the Fridman interview he describes putting a Stripe payment button on the thing at launch to see whether anyone would actually pay.
This is the same discipline Buffer applied with a pricing page and Zapier applied by charging $100 for beta access, and it separates a real experiment from a vanity launch. Traffic, upvotes, and signups are all measures of politeness. A card charge is a measure of intent, and it arrives within hours rather than months.
The practical version for a reader: whatever you are about to launch, do not launch it without a way to take money, even if the product is embarrassing and the price is a guess. The information is worth more than the revenue.
Doing it manually first
The most recent chapter is the clearest illustration of a principle most builders skip.
Levels describes on the podcast how PhotoAI started. He posted AI-generated photos of himself, people sent him direct messages asking for the same thing, and he did it by hand: downloading their files, training a model, emailing the results back. Only after that did he automate it. His summary is a nod to the standard advice: "Do things, don't scale, Paul Graham says, right?"
The manual phase produces three things automatically. It proves demand, because people are paying before the software exists. It teaches you what the software has to do, in detail, from the requests themselves. And it produces the first customers, who arrived asking rather than being sold to. That last point is the one we keep finding in every documented account of where first customers come from.
He also reports being wrong in public. Of an earlier AI avatar trend he says: "This is hype. This is going to die down very fast, and it did, because it's too cheesy." He still reports making around $100,000 in the first week from Avatar AI. Correctly identifying a fad and monetising it anyway is a legitimate strategy, and an honest one to admit to.
The cost structure that makes solo possible
Levels publishes technical details that are unusual to see stated: he describes PhotoAI as a "40,870 line index.php" and reports serving "4 billion requests a year" for a hosting cost of "$244/mo."
Both are deliberately unfashionable. A single large PHP file is not how a team would build software, because a team needs boundaries between people. One person does not. Similarly, $244 a month for four billion requests is only possible because there is no infrastructure sold to him by a vendor whose pricing assumes a company.
This is the practical content of what he calls a setup with "100% automation and 99% profit margins" in a November 2024 post. Take it as directional rather than precise; a solo operator's margin calculation excludes his own salary. But the structural point holds, and it is what we described in the one-person software company is finally viable: the fixed costs that used to require a team have collapsed.
His own framing of the advantage is about speed, not cost: "Being alone by myself on my laptop...I can ship very fast and I don't need to ask legal."
The part that is not a playbook
Levels has an audience. He reported in December 2024 that his posts had drawn a billion views on X over twelve months, and he attributes a record revenue month directly to appearing on a podcast. That audience is the distribution engine underneath every launch, and it was built over a decade of publishing in public, including the failures.
So the honest reading of the twelve-startup year is not that shipping twelve products makes money. It is that shipping twelve products in public, with the results posted either way, built an audience that made the thirteenth product easy to launch. The compounding asset was the reputation, not the code, which is the argument we made in distribution is the moat.
What transfers: a public deadline; a scope defined by the date rather than the feature list; a payment button on day one; doing the service by hand before writing the software; and publishing what happened.
What does not: the audience, which took years, and the specific timing of the AI photo wave. And the numbers themselves come with no external verification. Read them as what a founder chose to publish about himself, which is more than most disclose and less than a filing.
Figures are from Pieter Levels' own posts on levels.io and his transcribed conversation on the Lex Fridman Podcast, checked July 25, 2026. All revenue figures are self-reported and unaudited.
Discussion
Sign in with Google or just a name. No email link, no password to remember.