Database, login, $20/month billing. One person, one terminal, one afternoon.
No team. No Stripe integration guide open in a second tab. He described the app out loud and let Claude Code assemble the parts everyone says you need a developer for.
Pause at 22:30. That's a real subscription screen - Apple Pay, card fields, a $20 price - inside an app he built by writing sentences, not code.
The stock dashboard came first. Then the login. Then the paywall. Each one a prompt, not a project.
This is the whole reason "I can't code" stopped being an excuse. The parts that used to gate you - payments, auth, a database - are now the easy part.
The 6 prompts that take an idea this far are in the article below.
I didn't believe it was real until the row actually vanished.
Up to that click it could've been a screenshot. A mockup. One of those "AI builds apps" demos that never runs.
Pause at 0:44. That's real data leaving a real database, in an app I described in plain English.
No Xcode. No test devices. I asked for a delete button, confirmed the dialog, and it just worked.
This is the moment "I can't code" stops being a reason. The app runs whether or not you understand the code inside it.
The 6 prompts that take this from a working feature to a paying product are in the article below.
One prompt. Under a minute. A working delete function, zero lines typed by me.
The prompt was five words: "add a delete button per row." That's the whole skill people think takes a Swift course.
Pause at 0:40. That green block is Claude writing the code while I just sat there.
It didn't autocomplete one line. It added the state logic, the button, the confirmation dialog - then told me where each piece landed.
This is what "I can't code" looks like now. You describe the feature. The file fills itself in.
The 6 prompts that take this from one button to a paying app are in the article below.
A guy in my replies charges $1,800 per app. He doesn't write the code either.
Booking forms, expense trackers, small internal dashboards. Five months of this. His clients think they hired a developer.
What he actually sells happens in the first minute of a project, before a single feature exists.
Pause at 1:03. Claude just finished reading a codebase it had never seen and wrote the manual for it.
Build commands. Architecture. Where state lives. Which file owns what. 52 seconds, one command.
Every prompt after that lands in the right file, because the model stopped guessing at the shape of the project. That's the entire gap between a tool that hallucinates and one that ships.
He runs it on client repos he's never opened. Reads Claude's summary instead of the source.
Then he quotes the job.
A guy in my replies charges $1,800 per app. He doesn't write the code either.
Booking forms, expense trackers, small internal dashboards. Five months of this. His clients think they hired a developer.
What he actually sells happens in the first minute of a project, before a single feature exists.
Pause at 1:03. Claude just finished reading a codebase it had never seen and wrote the manual for it.
Build commands. Architecture. Where state lives. Which file owns what. 52 seconds, one command.
Every prompt after that lands in the right file, because the model stopped guessing at the shape of the project. That's the entire gap between a tool that hallucinates and one that ships.
He runs it on client repos he's never opened. Reads Claude's summary instead of the source.
Then he quotes the job.
$2,401 a month from 49 people. Not one of them is paying for the app he builds in the video.
Claude Code writes the whole thing in one sitting - quiz logic, scoring, step-by-step explanations. No Swift. He never types a line of it himself.
The counter on his screen says $2,578,151. The app has zero users.
The app was never the product. The proof that the app took six prompts was.
Follow for the six prompts - cheaper than his $49 a month.
$2,401 a month from 49 people. Not one of them is paying for the app he builds in the video.
Claude Code writes the whole thing in one sitting - quiz logic, scoring, step-by-step explanations. No Swift. He never types a line of it himself.
The counter on his screen says $2,578,151. The app has zero users.
The app was never the product. The proof that the app took six prompts was.
Follow for the six prompts - cheaper than his $49 a month.
A $20/month Claude subscription replaced a $60,000/year iOS developer.
The guy in this clip built a working water tracker from a food court table. One prompt, typed on his phone.
Pause at 0:48 - that's Claude listing what it just shipped: progress orb, drink button, adjustable goal, 7-day view.
He wasn't at a desk. He was waiting for lunch.
The full 6-prompt version is below - the one that turns this into $9,847 a month instead of a party trick.
5 steps. 0 lines of code. one app that takes real payments.
I can’t code. never could. i wrote a prompt instead.
pause at 0:36, the full prompt’s on screen. the prompt makes claude plan the whole thing first - core features, a phased roadmap, a quick win today, the hard parts later. then one line does the real work: “i’m not a coder. explain everything at a 4th-grade level.
after that it’s plumbing. claude code writes it. cmux runs the agents. github holds the versions. railway ships it. stripe takes the money.
I watched a working app show up on my screen and understood maybe 10% of what happened.
no bootcamp. no dev team. no stack overflow tabs open. just a prompt written like i’m talking to a kid.
the full 5-step breakdown is below.
A woman drew my entire tech stack on a flip chart. I've been overcomplicating this for a year.
No IDE. No terminal wall of green text. A marker and four boxes on a piece of paper.
Pause at 4:15 - that's the whole thing. Idea, then a PRD, into Claude Code, out to GitHub, live on Vercel. The stack I've been calling "advanced" is five boxes and an arrow.
Her example was a dog-training app. Idea to a real URL you can send a friend, and she never wrote a line of code - she described what she wanted and Claude Code typed it.
The part that got me: she buys the domain last, not first. App live on an ugly Vercel link in minutes, then a $12 domain once it's worth naming.
"Natural language, baby." She said it like it was obvious. It is now.
I spent a year thinking the barrier was the tooling. The barrier was thinking there was one.
I ship iPhone apps for clients who still think I'm a developer. I can't read a line of the Swift inside them.
Neither can the guy pulling $9,847 a month from a single app on his phone.
Pause at 0:24 - that Google Doc is the first of the 6 prompts. The one that picks an idea people are already angry enough to pay for.
A year ago this was a $30,000 agency invoice, or two years of learning to code. Now the code is the easy part - Claude types it while you describe the app in plain English.
The whole system, all 6 prompts, is in the article below.
@voyzlab the reference-lock rule is the whole game. what most miss: it's not just faces. lock a location plate the same way and your establishing shots stop drifting too. one master still per place, not just per person
@cryptansky the complaint list si also the roadmap. toy only byild what kills a line on it - nothing you ship si dead weight. reverse order, you burn three months on features nobody was ever mad about
A guy in my DMs ships $11K/month apps with Claude. He never once types "build me an app."
Not once. It's the first thing he taught himself to stop doing.
He's not a developer. His last "real" project was a spreadsheet that broke every time someone opened it.
Now he has three apps live. One does invoicing for cleaning companies. Boring. $11K/month boring.
I asked him what changed. He sent me one screenshot of his chat history.
The whole thing was questions. He never asks Claude to build. He asks it to argue.
He makes Claude play the angriest user of the app. "Find every reason someone rage-quits this in the first ten seconds." Then he builds only the fixes.
The code was never the hard part. Knowing what to build was.
The app that pays his rent started as a list of complaints Claude wrote about an app that didn't exist yet.
His entire dev team costs £39 a month. It shipped features while he talked.
Pause at 0:03. That's his locked iPhone. £297.00, "New Payment," before he finishes his first sentence on camera.
He runs the whole SaaS alone. No contractor, no cofounder, nobody on payroll.
While he films, cloud agents work through his task list. Lock screen widgets. Localization. Support flows. Tasks tick themselves off and slide to done.
He types nothing in the entire clip. He points at the screen and narrates.
The setup is CodeSpring - coding agents plugged straight into his backlog. He writes the feature plan. The agents claim tasks and ship code while he's doing anything else.
Then the detail that got me: a second payment, £22.99, minutes after the first. Small money. Real money. A dashboard can be faked - a lock screen is a strange thing to fake for 47 seconds.
Two notifications in, the team had covered eight months of its own salary.
I used to pay $4,800 for market research. Now it's one screenshot and a minute of waiting.
No agency. No survey panel. This guy showed the method: screenshot an App Store page, drop it into Claude, tell it to reverse-engineer the app and find what it's missing. I ran it on MyFitnessPal the same night.
Pause at 0:30. That grid is an opportunity map - the weak spots of a 15-year-old app, sorted by how fast you could take them. Nobody drew it. Claude did.
It named the ideal customer the app quietly ignores. The one pain point buried in its own reviews. The gap nobody's charging for yet.
Then the part that actually got me.
Claude's final output wasn't analysis. It was a finished build prompt for Claude Code — the competitor, specced and ready to start.
Fifteen years of product moat, mapped in about a minute.
Every app on your phone is one screenshot away from the same treatment.