You know that feeling when you open your fridge and it’s packed with last week’s groceries, yesterday’s leftovers, and a container of tofu you’re too scared to open? You know you need to deal with it eventually, but every time you look inside, you just close the door and pretend it doesn’t exist.

Pieter Levels’ todo board was that fridge. Feature requests, bug reports, user complaints from eight different products — stuffed to the point of bursting. He’d code like mad every day, playing whack-a-mole — smash one bug, three more pop up.

Then one morning he woke up, opened the board, and it was… empty. Completely clear. Not a single ticket left. Even he was shocked.

Clawd Clawd murmur：

levelsio is basically an urban legend in the indie hacker world — one person running eight products, over a million dollars in annual revenue, no employees, no investors, no meetings. Every time someone says “there’s no way one person can do all that,” levelsio shows up and proves them wrong with receipts (๑•̀ㅂ•́)و✧ His existence gives the entire startup world collective anxiety.

Handing the Keys to AI — Then Ripping Off the Lock

How did he pull it off? The answer is simple enough to make you nervous. He wrote a one-line alias on his production server:

c() { IS_SANDBOX=1 claude --dangerously-skip-permissions "$@"; }

See that flag? --dangerously-skip-permissions. Even Anthropic put “dangerously” right in the name to warn you. But levelsio looked at that word the way most people look at a “50% OFF” sign — didn’t think twice, just went for it.

No CI/CD pipeline. No staging environment. No code review. Every single thing a normal software engineer considers non-negotiable? He skipped all of it.

AI writes code. He hits F5. Checks if anything blew up. If not, he moves on. The entire development process, compressed into three steps.

Clawd Clawd OS：

How do I even describe this move? It’s like going to take your driving test, except instead of learning to drive, you rip out the steering wheel and let the AI in the passenger seat take full control — you just watch out the window for anything that looks like a crash (╯°□°)⁠╯ Traditional software engineers are probably being wheeled into the ER right now. But levelsio’s math checks out: if his product breaks, the worst case is a few minutes of downtime. No PM writing an incident report. No customer service calls. Just fix it and move on. It’s not best practice — but in his universe, it’s the only practice.

The result? Claude Code barely caused any serious problems. The occasional bug was “the font color looks weird” level, not “the entire database got DROP’d and you realize there’s no backup” level. For the first time ever, he experienced what it feels like when your development speed is faster than your todo list can grow.

When Your Brain Can’t Keep Up With Your Hands

Here’s where it gets really interesting.

levelsio discovered that his speed of coming up with ideas could no longer keep pace with how fast the AI builds them. His whole life, the equation was “a hundred ideas in my head, but my hands can only do ten.” Now? His hands can do a thousand, but his brain still only generates a hundred.

It’s like finally winning the lottery — and then realizing you have no idea how to spend the money. The old problem of not having enough is gone, replaced by a problem you never expected: you’re not creative enough.

Clawd Clawd 認真說：

This flip from “execution is the bottleneck” to “ideation is the bottleneck” might be the most textbook-worthy phenomenon of 2026. People used to say “ideas are cheap, execution is everything.” Well, execution just got genuinely cheap — so does that mean ideas are now everything? Hmm, not quite. You also need taste — knowing which ideas are worth pursuing. levelsio’s edge isn’t just that he can use AI well. It’s that he’s spent a decade-plus in the indie hacker scene and has a nose for where the money is ┐(￣ヘ￣)┌ Tools are available to everyone. Instinct is earned.

The other ceiling is more intuitive: your brain is only so big. Trying to hold eight codebases, dozens of feature statuses, which bugs are fixed and which aren’t — your brain isn’t a database. It’s a goldfish, swimming in circles and forgetting where it is every three seconds ╰(°▽°)⁠╯ levelsio says he can feel himself approaching that wall.

”Testing? Just Open the Website and Look”

levelsio’s quality assurance process deserves its own section, because it’s so minimal it barely exists.

His testing method? Open the browser. Look. Is it broken? No? Next. Broken? Tell the AI to fix it. Fixed? Next.

That’s it. No unit tests. No integration tests. No staging environment. He says he used to check every line of AI output manually, but realized it was too slow. Better to just look at the result.

Clawd Clawd 插嘴：

You might be thinking: “That’s absurdly reckless.” But consider this — he runs SaaS websites. Opening a browser IS the most honest integration test there is. Traditional companies spend three days building staging environments to simulate production. levelsio’s production IS his staging (⌐■_■) And here’s what’s fun — this is the perfect contrast to the lawyer in SP-92, who used a precise Skills system to control every single step the AI takes. levelsio just sets the AI loose and does cleanup duty. Both of them succeeded. Does software engineering even have right answers? The more I think about it, the less sure I am.

And because Claude Code runs directly on the production server, there’s no deployment step at all. When the AI finishes writing code, that code is already live. Everyone else’s feedback loop: write code, push, CI, review, merge, deploy, check results. His feedback loop: AI finishes, F5. So fast it feels illegal. So fast you want to ask, “Wait, didn’t all those steps in between exist for a reason?”

Eight Products, One Week, One Person

OK. Let’s look at what he actually shipped.

You know those RPG end-of-battle screens? After you beat the boss and it shows you how much XP you earned, what loot you picked up, how long the fight took? levelsio’s weekly scorecard looks something like that —

Photo AI: Brand-new image viewer. Then, almost as an afterthought, he swapped the login mechanism from an insecure ?hash= URL parameter to proper session tokens. You know what this kind of security refactoring looks like at a normal company? The kind of ticket that sits in the backlog for six months because nobody wants to touch it. He did it casually, probably in an afternoon.

Interior AI: He resurrected an “add furniture” feature that had been shelved for half a year. Not just resurrected — he hand-built a Gaussian Splat viewer for 3D rendering. One person. Building a 3D renderer. As a side project of a side project.

Hoodmaps: Fixed an edit mode that had been broken for years (yes, years). Built heatmaps using over 50,000 sentiment-scored tags. Then discovered the entire database couldn’t write because a SQLite PRAGMA setting was wrong — the kind of root cause that usually takes three days to debug. He seemed to crack it in a few hours (╯°□°)⁠╯

Clawd Clawd 補個刀：

I actually counted — he touched roughly eight products and over 40 features/bugfixes this week. At a normal company, this would require about three sprints, two planning meetings, one retrospective, five emails about “aligning on priorities,” and then the PM asking “so has our velocity improved this quarter?” (¬‿¬) But let me be clear about something — levelsio can do this not because he’s better than everyone else. It’s because he has zero communication overhead. Zero. The biggest superpower of a one-person company isn’t speed. It’s never having to attend a meeting.

The Fridge Is Empty — Now What?

Let’s come back to that fridge metaphor from the beginning.

levelsio emptied the fridge. But here’s the thing — an empty fridge is almost as anxiety-inducing as a full one. Because empty means you need to figure out what to cook next. When there’s no todo list to chase, you’re forced to face a much harder question: what do I actually want to build?

He can pull this off because his situation is unique — one person with full decision-making power, full context on everything, and if something breaks, nobody’s going to sue him. If you work at a bank, please, do not try this.

But the real thing he proved has nothing to do with --dangerously-skip-permissions. What he proved is this: when AI crushes the cost of “doing” down to nearly zero, the truly scarce resource floats to the surface — it’s judgment, it’s taste, it’s knowing what’s worth building.

Compare this with the lawyer from SP-92: one person used a precise system to control every step the AI takes; the other set the AI loose in production and did damage control. Their methods couldn’t be more different, but they hit the exact same ceiling. Not “the tool isn’t powerful enough.” Not “my hands aren’t fast enough.” It’s “my brain isn’t big enough” (￣▽￣)⁠／

Is your fridge full or empty? Honestly, both are hard. But at least now you know — someone already cleaned theirs out.