I set out to build a finance tracker. I ended up with three apps and a half-built fourth idea sitting in a notes file.
Over the last few months I built a finance app, a chores app, and a photo album app, all with Claude, all meant to feed into one thing: a centralized hub my family actually uses. Not a metaphor. An actual home page with actual apps my kids and my spouse open.
Here’s what nobody tells you about building this way. When code takes minutes instead of hours, your ideas don’t slow down to match your judgment. They speed up to match your tools.
I’d sit down to add a savings goal field to the finance app. Twenty minutes later I had a savings goal field, a projected payoff chart, and a notification system I hadn’t asked for, because once I saw how fast the first thing shipped, the second thing felt free. It wasn’t free. It was debt I hadn’t priced yet.
The chores app is where this really showed up. What started as “assign a task, check it off” grew a point system, then a leaderboard, then a rewards store, then a recurring-task engine, in the space of a weekend. Each addition felt small in the moment. Stacked together, I’d built something closer to a game studio’s backlog than a family utility.
This is the part that’s hard to say cleanly, because it’s not one lesson, it’s two pulling against each other. Momentum like this is genuinely inspiring. You feel the gap between idea and working thing collapse in real time, and that’s addictive in a good way. But that same collapse removes the friction that used to force you to ask “should I” before you asked “can I.” Cheap execution doesn’t just accelerate good ideas. It accelerates all of them, indiscriminately.
I don’t think the fix is discipline in the abstract, “just say no to yourself” doesn’t survive contact with a blinking cursor and a model that will happily build whatever you type next. The fix I’m actually testing is treating scope like a PRD, even for a weekend project for my own kids. Write down what the thing is for before opening the editor. If a new idea doesn’t serve that job, it goes in a backlog file, not in the app.
I haven’t fully solved this. The photo album app already has a “smart album suggestions” feature I’m not sure anyone asked for, including me. But naming the pattern is useful. Scope creep used to be a slow leak. With AI in the loop, it’s a firehose you can point at yourself by accident.