Pieter Levels (levels.io) needs little introduction to most indie hackers: Nomad List, Remote OK, Photo AI, and a $420,000/month peak in 2024, all run solo. If you don't recognize the name, you may still remember the 2025 story about an AI-coded flight simulator that hit $1M ARR in 17 days while he built it live on stream. That was him.
Worth stating up front: this is one person, not a company. No co-founders, no employees, since 2014. The only things he outsources are customer support and server monitoring — and by his own account, that's a choice, not a funding constraint. $420,000/month annualizes to roughly $5M a year, the kind of revenue that usually comes with a headcount in the dozens. He's also one of the more prominent "build in public" figures, posting most of his numbers on X.
I saved all 751 of his blog posts (2008–2026) from RSS and worked through them with AI assistance — a close read of 41 posts, with the rest processed into a searchable reference. Every number below was human-checked against the original posts. That check wasn't a formality: it caught the AI inventing a plausible-sounding maxim and dressing up a paraphrase as a verbatim quote, and those spots were fixed against the sources. So this is not an article written by AI — it's one made with AI and inspected by a human.
Full disclosure: going in, I barely knew who he was — vaguely aware of the flight simulator story, nothing more. So I read with no prior narrative, working only from what's publicly posted. What stuck with me by the end wasn't "wow, $420K/month" — it was the failure count. This is a report on both numbers: the successes he's public about, and the failures.
The success numbers
Starting with the well-known side. Here's what he's posted publicly, in chronological order:
| Period | Number |
|---|---|
| 2008–2013 | First income source: a YouTube music channel. Peaked around $8K/month, then declined to $500–900/month as copyright claim disputes piled up |
| 2014 | Started "12 Startups in 12 Months" while income was falling and he was in a depressive stretch (Nomad List was one of the 12) |
| 2015 | $202,785 in annual revenue (one server, one MacBook, no office, $0 raised) |
| Feb 2018 | $52,843/month (Nomad List + Remote OK + a book) |
| May 2019 | Crossed $1M ARR — 4 years in. 11,996 customers, $115 average payment |
| Nov 2022 | AvatarAI hit $100,000 in its first 10 days |
| Sep 2024 | $420,000/month across all products (all-time high). ~80% margin; GPU costs alone were $60K/month |
| Mar 2025 | fly.pieter.com, an AI-coded flight simulator built live on stream, reached $1M ARR in 17 days |
| Mar 2026 | Photo AI hit $105,000/month, running on a single 40,870-line index.php file |
I included the first two rows deliberately, so the $202,785 in row three doesn't read as a cold start. 2015 was year 7 of the project, coming right after two years of declining income. This isn't a story about someone who hit it big immediately — it's seven years in, one rough patch, then a jump the following year, once you read it in order.
Two footnotes. "17 days to $1M" refers to ARR (the run-rate implied by that month's revenue), not $1M actually collected. And the 40,870-line index.php isn't a joke — he states it himself, the result of over a decade sticking to plain PHP with no framework.
One infrastructure number worth including: as of 2026, he's serving 4 billion requests/year across all his sites on self-managed VPS infrastructure for $2,932/year. By his own estimate, the same setup on managed cloud services would run $338,913/year — a 115x gap.
The failure numbers
This is the part that matters. In a post titled "List of all my projects ever," he self-categorizes every project he's ever built — 113 total. Of those, 19 are tagged "Failed" (didn't take off / never made money). That's about 17%.
Reading this as "so the other 83% succeeded" would be the wrong takeaway. Of the 113, at most 9 could reasonably be called revenue pillars. Most of the rest carry no success/failure tag at all — they're experiments that quietly died. The 19 "Failed" tags are just the projects he explicitly admitted failed; as an actual failure count, it's closer to a floor than a ceiling. By hit rate, we're talking under 10%.
In other words, behind the handful of projects that get named as wins, there are close to a hundred projects nobody's heard of. A few have documented reasons for failing:
- Tubelytics (2013–2014) — a YouTube analytics dashboard. About a year of development; the interface was polished, but in his words, "nobody paid for it." He's said he's "so traumatized by this" that it became a standing rule: "I'm not going to build anything until there's customers."
- Uber Clone for Netherlands (2010) — he and a friend pitched wealthy investors for funding. His own summary: "It turned out to be a massive time waste and they never invested in anyone and it was quite humiliating." This appears to be the origin of his distrust of VC funding.
- Panda Mix Show (2008–2014) — his first success, a YouTube music channel. Peaked around $8K/month, declined to $5K as copyright claim handling got harder, then shut down. He's written that this income decline, combined with a depressive period, was the direct trigger for the well-known "12 Startups in 12 Months" project. That project wasn't a relaxed side experiment — by his own account it came out of a rough stretch, as a way to force himself to do something.
- Hoodmaps (2017–) — self-tagged "Failed," yet still running as of 2026. A live example of a stated principle: if upkeep cost is near zero, leave a failed project running rather than shutting it down.
The remaining roughly 15 are dismissed with a line like "never took off," with no real postmortem. There's an asymmetry worth naming here: the more successful a product, the more detail he gives in hindsight; failures get a sentence. Any read of self-reported data needs to account for that skew.
Seven preconditions that don't transfer
Reading 751 posts in order also surfaces the conditions under which this approach worked — the things to subtract before trying to copy it:
- 300,000+ X followers, a distribution channel built up over roughly a decade since 2013. The leverage of "launch across multiple channels at once" depends on already having this.
- 2013-era market conditions. He's said, in effect, that back then it was a nearly uncontested space; now everyone, including large companies, is competing on the same turf.
- Low-cost living in Southeast Asia underpinning his runway math (Bangkok at roughly $240/month, $3,000 in savings lasting a year).
- A decade of honing plain PHP. "You don't need a framework" isn't beginner advice — it's the endpoint of someone who repeated the same pattern for ten years.
- Low stakes from running solo. Running AI agents with bypassed permissions on production servers is viable partly because, from his position, a mistake isn't catastrophic.
- His VC distrust is a generalization from one bad experience, not an industry-wide analysis.
- AI product margins track GPU pricing directly — he's mentioned suppliers raising prices on him after seeing his revenue numbers posted publicly.
What's still worth taking
Even after subtracting all of the above, a few things seem genuinely portable for an indie developer anywhere:
Don't build until you know there's a customer. This carries weight specifically because it comes from a wasted year on Tubelytics. In practice: put up a payment link and a signup form before writing any code, and confirm someone will pay first.
If upkeep costs next to nothing, don't kill a failed project. The Hoodmaps pattern. Redefining "failure" as "does it cost money to keep running" rather than "did you shut it down" lowers the cost of staying in the game.
Make your failures countable, and publish them. Someone self-reporting 19 "Failed" tags out of 113 projects is rare, even among people who've succeeded publicly. Plenty of people write success stories. Far fewer publish the failure denominator — and that's what it takes to actually talk about odds.
One last thing. In systematic trading, there's a well-worn discipline: don't trust a good track record until you've verified whether it's skill or luck. I think public revenue numbers deserve the same treatment. $420,000/month is self-reported, not an audited statement. I'm not doubting the number — but I think the right way to read this kind of material is with the premise built in that public numbers are a survivor's self-report. Reading all 751 posts was, in part, a way to check that premise for myself.
All figures are sourced from his own blog at levels.io (specific post titles are noted in the text above). Numbers reflect the value at the time each post was written, not current figures.
Originally published in Japanese on Zenn: https://zenn.dev/idonthaveapen/articles/levelsio-751posts-19-failures
Top comments (0)