At 27, Pieter Levels Stared at a Hotel Ceiling. Then He Shipped Twelve Startups in a Year

Pieter Levels's method is the inverse of the usual startup path. Most founders raise, hire, then hunt for something that pays; he designs the logo, writes the code, and launches a tiny product in about two weeks to see whether anyone will actually put a credit card down. Most ideas fail. The ones that work do so because people pay, not because they sign up.

He compares that speed to Google. The company invented transformers and, he has heard, could have shipped a ChatGPT-like product around 2019, but bureaucracy held it. Alone on a laptop in a hotel room, often in shorts because he overheats, he does not ask legal for permission. He just ships.

Building, for him, is closer to his father's renovation hobby than to a liquidity event. His dad has been circling the same house for forty years, breaking through one room and starting the next; the lesson for depression was to get a pile of sand and a shovel and move it. Levels treats startups the same way: something to wake up for, a bug to fix, a reason to make coffee. He was already making drum-and-bass and websites when none of it paid, so he does not pretend the work is mainly about money. Founder friends with huge raises tell him, privately, that next time they want to do it his way, because becoming a CEO is a job a lot of builders secretly hate.

After university in Holland he refused a normal job. A YouTube channel, Panda Mix Show, was paying two to three thousand dollars a month in AdSense for dubstep, drum and bass, and techno mixes. Friends took offices, weekday hanging-out ended, and the suburban house-and-job script looked like a trap. He flew to Asia to become a digital nomad, then felt worse: alone in hotel rooms, income shrinking toward five hundred dollars a month, friends at home doing "well." At 27 he called it Saturn return and decided he was a loser.

His father's advice was not talk therapy. If you are depressed, shovel sand; do not sit still. Levels announced "12 Startups in 12 Months" as a blog post on Hacker News so the public could hold him to it. Each month: a product, a launch, a Stripe checkout. He is not sponsored by Stripe, he says, but a checkout button is still the easiest way to get paid — a long way from faxing Worldpay merchant forms from his father's machine when he was nine.

The first product was Play My Inbox. Before Spotify, friends mailed YouTube links around giant Gmail threads; he used POP or IMAP to pull those links into a gallery. Tens of thousands of people used it and tech press wrote it up. There was no pay button, so it made nothing, but it proved strangers would use something he built. The constraint of thirty days forced him to skip a real login system — Photo AI later launched with a hashed static link after Stripe, and Google login took about a year.

Lex asks him to go back to the hotel ceiling, because a lot of listeners are there now. Levels says it is normal, especially around 27, and then says the important part: do not kill yourself, get through it. He cites Jordan Peterson's line that people are depressed because they have nothing — no partner, no job, no family. Alone in an Asian hostel with no income, depression is logic; in the moment it feels like a defect in the self. Travel, he adds, can also make you a little crazy. He has a hunch that early digital nomads had a high suicide rate, though he has not checked it against a baseline.

The lifestyle found him by accident. After a 2009 exchange in Korea, Thailand looked cheap and easy; Chiang Mai rent was about $150 a month. Other people on laptops in Colombia, Vietnam, and Bali looked adventurous, and leaving Dutch social pressure felt like freedom. Lex quotes Kafka: "I'm free, therefore I'm lost." Levels agrees. Freedom is no constraints; everyone assumes that is happiness. He learned the opposite. The nomads he met then were often in sleazy e-commerce — diet pills, gray-area shipping — and he did not want that subculture. He wanted Hacker News startups.

The happy memories are still vivid. In Bali he and his best friend Andre would wake at five or six in the evening, buy thirty hot lattes in a plastic bag, ride motorbikes to a silent coworking space, and code industrial techno until yoga people said good morning at six. It was lock-in mode, and it was fun. Thailand still wins for kindness; Brazil is amazing in the right neighborhoods and dangerous in the wrong ones. He later built Nomad List because he was lonely, then started meetups and made friends. Waking up for half a minute unsure which country you are in is romantic in an Anthony Bourdain way. It did not make him happy. It made him anxious.

His book *Make: Bootstrapper's Handbook* walks idea, build, launch, grow, monetize, automate, exit. Ideas start as irritation: list what is broken in daily life and which fixes you can actually ship. Travel is arbitrage — Southeast Asia already had super-apps for food, groceries, massage, and mechanics, years before the West. When he had no money he kept a statistical list of things to try, assuming one would hit. With money he can start from technology, which he admits is the classic hammer looking for a nail.

Stable Diffusion on a MacBook was bad at people and good at houses, so he shipped thishousedoesnotexist.org: generated houses, descriptions, votes, a leaderboard. It was popular, then everyone including him got bored. Interior AI came next — upload a room, restyle it. It did $10k to $20k a month in the first week and still does about $40k to $50k two years later. Fine-tuning interiors worked so well that, for fun, he trained on photos of himself. The model learned him as a concept. Avatar AI made about $150k in a week, the most he had ever made. Then Lensa, a retouching app with real iOS talent and an existing user base, pushed the same feature and he thinks made around $30 million and hit top-grossing. He was not crushed. He was impressed. He also realized cheesy Picasso-Barbie avatars hide model errors as art; photorealism has nowhere to hide. He pivoted to Photo AI.

The whole stack is still vanilla HTML, jQuery, PHP, and SQLite. He calls that an accident: it is what he knew, "learn Node.js" has sat on the to-do list for years, and the products grew too fast to switch. His more controversial claim is economic. Framework companies raise tens of millions, hook 18-year-olds, and steer them onto paid platforms that cost a thousand times a VPS. Influencers get paid to demo the stack. When he says PHP and jQuery, an army asks why he is not using the new thing. Lex wants data: shipping speed, bugs, and holes for one developer, five, ten, twenty. Levels's test for the framework army is Elon's: what did you build this week, and did anyone pay?

He did not write an app. He registered a domain, put up HTML, stuck a Stripe payment link on it, and sent payers to Typeform to upload photos. Then he downloaded zip files, trained by hand, generated from a text file of prompts, and mailed results from his personal address. Paul Graham's "do things that don't scale" was literal. Within 24 hours demand was huge; a few hundred customers later he had no choice but to automate. Famous people signed up, including, he will not name under GDPR, tech billionaires. The hype was cheesy and it died down. The lesson stayed: people were paying, so the need was real.

The first GPU vendor watched him tweet revenue and raised training from $3 to $20, wiping a $30 product's margin. In Slack they said they might lower it later. He DMed Replicate's CEO, Ben Firshman, and begged for DreamBooth even after a no — the job took half an hour and they lacked GPUs. A week later Replicate launched it. Avatar apps flooded in on the same API; Replicate went from an ML-native console he did not understand to a one-click playground. They never price-gouged him. He has now trained on the order of 36,000 user models there.

The product taught him things about faces that are not in a manual. People say a photo "doesn't look like me" when the training set clearly does; he calls it face dysmorphia and tells users to let someone else pick the shots. A YouTube comment once said he looked like a hammerhead shark because of wide-set eyes; the trait he thought was ugly is the one that makes him specific. Training photos must vary clothes, light, and setting so the model learns the face, not the green shirt. He mixed full-body shots back in after face crops erased body type. At a million-plus images a month he A/B tested samplers and step counts on 10 percent of users, using favorites and downloads as the metric, and told them at signup. Most of the scene, he says, still sits on Stable Diffusion 1.5. Newer, "neutered" versions were worse. Stability AI raised hundreds of millions, made little money, and the indie apps kept the model.

On the "build" step he is blunt: use the technology you already know. Nomad List launched as a public Google Sheet anyone could edit — internet speeds, cost of living, cities — tweeted to a few thousand followers and filled in from around the world. If you cannot code, use a spreadsheet; if you cannot spreadsheet, use Wix or Squarespace. You can rewrite later. The first job is to see whether anyone will pay.

Hoodmaps came from a traveler's problem. In Amsterdam the center is tourists; the interesting bits are Jordaan or De Pijp. He wanted a color overlay on Google Maps: red tourist, green rich, yellow hipster, gray residential. He did not know HTML5 canvas when he started. He documented the embarrassment of learning it, laid a 50 percent opaque canvas over a Maps iframe, and let people paint. A daily cron majority-votes each little area. GPS decimals got truncated into a grid. Offensive doodles exist, but they do not cluster, so the vote washes them out.

Tags turned the map into a meme machine. Austin sprouted "Drunk frat boys," "AirBroNBros," "Hipster girls who do cocaine," and "Saw a guy in a fish costume get beat up here" — lines he could not have written, because he does not know Austin. It hit Reddit's front page and almost took the server down. It has never earned a dollar. Google Maps blew the free quota; Mapbox later sent a $20,000 bill; he is on an open-source map now. Ads fail because 90 percent of the UI is the map. Charging would kill the crowd and then the accuracy. Some products, he says, are just cool, and he wants Hoodmaps to exist. When 4chan arrived with slurs, he treated them as free pen testers: a dirty-word list redirected them to relaxing ASMR — glowing jelly, cheese melting on toast.

Nomad List started in Chiang Mai because he needed fast internet and $200 rent on a $500 month. Much of the world was at 0.5 Mbps, too slow for YouTube. The sheet became a site that is almost ten years old. Crowdsourcing warped: Indian users would max every score for India, not one person but a crowd. He moved most metrics to World Bank and UN datasets, which are usually national, not city-level, and kept the unique signal — members logging where they are and where they have been — to see which cities are heating up.

Membership is on the order of 30,000 to 45,000. Austin might show 1,800 remote workers and a handful of members checked in, plus people arriving soon. About thirty meetups a month organize themselves: someone sets a time, the system checks how many members will be in town, then tweets and DMs them. He does not run the events. "Why would you hire someone to organize this?" he asks. Let the code do it.

Remote OK spun out of a Nomad List jobs page when companies started paying. It is now a top remote job board, automated, with profiles that feel a bit like LinkedIn for remote work. In 2021, when the Fed was printing, he piled on upsells — rainbow posts, custom backgrounds, pin, rush — and a single listing could run $3,000 to $4,000. The site hit about $140k a month. When the printer stopped it fell to about $10k. It has come back to around $40k. "Those were good times," he says. The same conversation turns to why airline and hotel sites stay terrible. His theory: you have to book the Qatar flight anyway, so they have no incentive to make the flow nice. Booking.com, a Dutch company, runs teams on a 10-by-10-pixel patch, A/B testing "only one room left" in red, and took an EU fine for it. It is a monster, and it is number one — which tells him UX is not the ranking function, revenue is. Lex's theory is organizational: great engineers cannot ship inside the bureaucracy, so they leave. Levels adds the cycle: Booking was the 1997 disruptor; now it is the old thing. Uber in many cities already feels like a taxi again.

In 2016, when the HTC Vive appeared, he gave himself thirty days to learn 3D. He installed Unity and Blender knowing nothing, live-streamed the whole awkwardness, and treated it like AI later: a new wave you should touch. People ask how to learn to code; he is still Googling or asking ChatGPT every day. Start with zero knowledge. Ask how to make a website, paste the file, open it in a browser, then add a button. Action beats books. He only looks at the next problem. At the end an app is sitting there. Most people look too far ahead.

Haters, he says, often have a point once you strip the insult. Thick skin on X is required; he has muted about 15,000 accounts in ten years, one by one, because a block screenshot starts a fight. Organic still beats ads: a stranger's TikTok once added about $20,000 a month to Photo AI, then the influencer asked for $4,000 to do it again. Algorithms now test a clip on a few hundred people first, so follower count matters less than whether the thing is good.

Free users never converted well for him. With venture money you can buy a funnel; as an indie he wants a card from day one, and more than Netflix's roughly $10. An individual needs about $30 per user per month to make the math work — a thousand people is $30k, a livable indie business if costs stay tiny. Paying users also filter spam. He emails vendors and asks for 25 or 50 percent off, especially in a soft economy; the website price is not the real price. Being public helps. Lex jokes that the secret is to be a good person. Levels says it depends how big the discount is, so maybe shitpost more.

Anonymous city reviews on Nomad List can run thousands of words. He pipes them through GPT-4: is this a good review, offensive, racist? Rejects land in Telegram for a human look. It is uncannily good at double meanings and jokes. A 10,000-person chat used to have human mods; users fought them for bias. GPT-4 ended the argument. Subjective judgment, he says, can now be an API call instead of an if-statement.

The rest of the company is crontab. He SSH-es in, `sudo crontab -e`, hourly PHP, done. Healthcheck.php prints green ticks and red crosses: Amsterdam internet speed a plausible number, signups in the last 24 hours, database queries. UptimeRobot hits it every minute looking for a red X and Telegram-alerts him. JavaScript and PHP errors from every site go to Telegram too. Early on he drowned; now errors are near zero because each one got fixed. Last week he jokes about 100.000 percent uptime. New Relic exists; he would rather build the boring version himself.

He has never sold a company. Acquisition theater is a six-to-twelve-month due-diligence grind through Stripe and source, then the price drops because you are invested. At 90 percent margins a three-to-five-times multiple is three to five years of keeping the asset. Six, seven, eight times would be different. Nomad List is his baby — people should work anywhere, meet other cultures, go to China and learn what is propaganda and what is not — so he does not want to sell it. Photo AI he could. Founders who sell often get depressed, ask who they are, and start the next one. He thinks people should work on cool things until they die; retirement is probably bad. He has almost never coded with another person. Philip did Python for Photo AI for ten months because NumPy and packaging were a wall. One other developer lasted a week: the first demand was rewrite everything from jQuery to Vue. Lex's rule is to assume the other person is the smartest alive and find the genius in the code first. Levels calls that code empathy, and guesses it is the top 5 percent. Most people in most jobs are average, including doctors. Hiring more would cost him energy. Things kind of run fine.

There is no staging environment. He edits in VS Code, commits, GitHub webhooks the server, it pulls, and production updates in one to two seconds. About 37,000 commits in the last twelve months. PHP lint and djLint catch some mistakes; the rest is the discipline of not taking the site down. Someone tweets a bug, he starts a stopwatch, and it is often fixed in two minutes. He knows this is idiotic in a big company. For him it is the whole advantage. Bugs at a newly unconstrained Twitter, he and Lex agree, can be a sign that a team is actually shipping.

The IDE barely matters. He loved Sublime's multi-cursor for mass-editing World Bank dumps, then Twitter called him a boomer for screenshotting it, and a clean MacBook install left him on VS Code. Whatever you know well, you go fast. In jQuery he accidentally rebuilds React-shaped problems and ends up with a tiny framework he understands. "No time" is always a lie; he does not prioritize it. Running the businesses and learning AI beat a frontend rewrite. Maybe a little Next, a little Laravel, a little Tailwind — concepts, not a migration. Lex proposes rebuilding one page with one API call in five stacks, stopwatch plus happiness. Nobody benchmarks developer joy or shipping time. If you want to learn AI, imagine Karpathy next to you building an LLM from scratch. He is probably anti-framework too.

He switched to Mac in 2014 because his brother said it was the travel standard, opened the aluminum box, and did not touch the machine for three days. All the businesses started working after that, which he knows sounds dogmatic. Finder is horrible compared with Windows Explorer; Unix underneath is the reason it is a good dev box. His favorite desk is a hotel sofa, laptop on a pillow, trackpad only, 16-inch single screen. He once had three monitors, a standing desk, and a Logitech MX, then at 6 a.m. in a Chiang Mai coworking space he turned his head and his neck locked. He thought he might die and did not want to die in a coworking space, packed the bag, rode home, legs up. Lifting — overhead press at 60 kilos is his victory pose, plus bench, squat, deadlift — fixed RSI and tingling. Friends with beautiful battle stations are not more productive. He is suspicious of more. *The Mythical Man-Month* for tools: extra screens are extra things to maintain. Gear-acquisition syndrome is a hobby that looks like work.

He sleeps around 2 a.m. as a compromise with a girlfriend who goes at midnight, wakes around 10, showers, and makes V60 drip with 20 grams of coffee — friends call it nuclear. Then bugs, then code. He used to need all-nighters; living with someone trained him to lock in and out in twenty-minute windows, the same muscle parents describe. When she leaves for a week he wakes at 1 or 2 p.m. because he coded until 6, then goes nonstop. Andre comes over and they sit on the sofa, mostly silent, Berlin techno from a YouTube channel called HOR filmed in a white-tiled toilet. Four nuclear coffees put him on the edge of anxiety without going over — the state for making new things, not for maintenance. Girlfriends, he says, should let their guys disappear for two weeks a year and make something.

Around 2012 he did the hundred-things challenge: spreadsheet every object, sell down toward 100. An MDMA trip that year left him sure he had to throw things away and leave. eBay was a tour of Dutch society, rich to poor, at the door. The hard goodbye was a Canon 5D he and friends had bought for about $6,000 to shoot music videos. Selling it meant the videos were over, Holland was over, that life was cut. What remained was a backpack, a MacBook, a toothbrush, clothes, maybe fifty things — swim shorts and flip-flops in Asia, laundromat, done. Minimalism shifts cost onto services: Uber instead of a car, restaurants instead of a kitchen. Freedom is real. So is the bill. Hedonic adaptation is fast for him, about two weeks; studies say a new car or house fades in six months. Skip the purchase. Keep people, food, experiences, work, exercise.

He wrote "Why I'm unreachable and maybe you should be too" after DMs became a queue of life stories asking for advice he is not qualified to give. Andre walked up in a coworking space, asked how to learn, installed Nginx, and went. People who ask for advice and then debate it rarely ship. Sending those messages, he thinks, is often procrastination: if you want a podcast, start one and copy the black background until you learn it is not about the background. As he has worked on himself he has become more "normie," calmer, happier — and, he worries, less interesting. Crazy people are the interesting ones. Lex's reply: you do not have to be broken to be interesting.

Nvidia, a three-trillion-dollar company, started at a Denny's table. People need a third space that is not home and not the office. Four hundred years ago, he says, the scientific revolution and the Enlightenment happened in European coffee houses — coffee, writing, debate, sea routes, in France, Austria, Britain, Holland. In American cafes he saw laptops everywhere. Large parts of Europe now ban them: we are here to enjoy life, not to work. He understands the Anthony Bourdain cigarette-and-coffee aesthetic. He also thinks a culture that mocks "startup bros" and forbids dreaming in public slowly stops producing companies.

He ran the numbers. Eighty percent of top EU companies were founded before 1950, versus 36 percent in the US. Median founding year among the top firms: 1913 in Europe, 1963 in America — fifty years. Europe used to be entrepreneurial; around 1950 the lead flipped. Old firms lobby, regulation thickens, and the result is regulatory capture: incumbents can afford compliance, a newcomer cannot. Starting an AI company in Europe, as he tells it, means notaries, certificates, licenses. In America you open a laptop. He likes Beff Jezos and e/acc as a needed optimistic counter to climate-and-collapse pessimism: spend more energy, including clean energy, build, go to space. China, for all its problems, is building the way America did around 1900. Comfortable regions go complacent. He made EUACC merch and a Make Europe Great Again hat, first red, then European blue, because red looked too Trump.

Advice for young people is the most direct stretch of the interview. Listen to nobody. Do what you think is good and follow your heart. Parents, family, and society will pressure you; try your own thing, because it might work. It will probably go badly at first — it did for him, relatively. If you are smart about it you can make a small life that is yours, like a village where he sells bread and you sell meat. You do not have to be a normie. You do have to go all out. Half-ass cannot succeed; lean into the outcast and the difference. Lex: whole-ass it. If he were 18 now he might skip university, travel, build internet businesses, and go back at 25 with money and a reason to study. Lex notes campus is forced networking. Travel, Levels says, gives you weirder people. Lex closes with Drew Houston: don't worry about failure; you only have to be right once.