Labs · Ongoing · month 7
Can a beginner really build an app with AI?
I'm a thirty-year technologist who had never shipped an app. I decided to find out in public whether the promise is real or a demo trick.
The question
Thirty years in technology and I had never shipped an application. That sounds like a contradiction and it isn't — I ran systems, I administered platforms, I fixed things other people built. There's a whole generation of us who are deeply technical and have never written the thing.
So when people started saying that AI meant anyone could build software, I was the ideal test case: enough context to not be lost, not enough experience to fake it. If the promise is real it should work for me. If it only works for people who could already code, that's a very different product than the one being advertised.
What I set out to build
Something real, with a user who wasn't me. I picked a small tool for tracking class registrations and attendance across recurring session dates — a genuine problem I had, small enough to finish, and unglamorous enough that nobody would confuse the demo for the difficulty.
Seven months in: it's live, it's used every week, and the thing works. I want that on the record before the complaining starts, because the complaining is the useful part and it can be read as a verdict if you skim.
The first six weeks were genuinely astonishing
I'm not going to be cool about this. Describing a screen in a sentence and watching it appear is remarkable, and if you're a beginner it will make you feel like you've been let into a room you were told you couldn't enter. That feeling is real and it is worth something — it is exactly the feeling that gets a fifty-eight-year-old to keep going past week two.
In six weeks I had working screens, a database, and login. Roughly a hundred hours of tutorials that I did not have to do first. That's not a small thing. That's the whole reason I now teach this.
Then I lost two weeks
A bug where the attendance count was wrong for exactly one class — the one with two sessions on the same day. Two weeks. A junior developer would have found it in an hour.
Here's the anatomy, because it's the most useful thing in this write-up. I couldn't describe the problem precisely, so I couldn't ask for the fix precisely, so I got confident wrong answers. Each fix changed something, and because I couldn't read the code well enough to know what, I couldn't tell a real improvement from a coincidence. I broke two working things chasing it. At one point I had four versions of the same function and no idea which was running.
What eventually solved it was not a better prompt. It was sitting down with the printed code and a pen and reading it line by line until I understood what it actually did. Forty minutes, once I stopped asking and started reading.
That's the trap, and it's specific to how these tools work: they will always give you another answer. There is no point where it says "you don't understand this yet." A tutorial would have stopped me. The AI cheerfully helped me stay confused for two weeks.
What actually got me unstuck, in order
Learning to read code, which I'd been avoiding on the theory that I no longer needed to. Wrong — and this is the single highest-value hour I've spent on any of this. You don't need to write it. You absolutely need to read it.
Version control, about four months later than I should have. Every beginner horror story I've heard since is the same story: a working thing, destroyed by a fix, unrecoverable. Learn this in week one, not month five.
And describing problems in terms of behavior rather than guesses. "The count is 2 when it should be 1, only when two sessions share a date" gets a useful answer. "The attendance is broken, fix the counter" does not.
So — can a beginner really build an app
Yes. Genuinely, and I'd have said no two years ago. I built a thing that people use and I could not have done it without these tools.
And it is harder than advertised, in a specific way worth naming: the difficulty has moved, not disappeared. It used to be front-loaded — the wall was at the beginning, and most people quit there. Now the beginning is easy and the wall is at eighty percent, where the demos end and where you find out whether you're willing to understand what you built.
Most people who quit now quit at month three, not week one. They quit further in, having built more, which honestly feels worse. If you know the wall is coming, you can plan for it. That's the entire reason this became a course instead of a blog post.