I'm not a professional engineer. I shipped a real production app — Next.js, Supabase, Prisma, live with real users — almost entirely by directing Claude Code instead of writing the code myself.
The origin story
I was logging push-ups and pull-ups in my phone's Notes app because every calisthenics tracker I tried was either bloated with coaching programs I didn't want, or built for barbell lifting with bodyweight reps bolted on as an afterthought. So I built Nickels & Dimes: log reps or timed holds, auto-tracked PRs, and clubs where you compete with friends on a monthly leaderboard.
What the work actually felt like
Not "coding," and not quite "managing" either. It was product decisions all day — what to build, what to cut, how something should behave — and almost no time spent on how to implement it.
The skill that mattered wasn't managing the model's morale, it doesn't have any. It was being unusually clear and decisive about what I actually wanted, faster than I've ever had to be with a human collaborator. There's no async buffer where someone quietly sits with a vague ask for two days before coming back with clarifying questions. You get an attempt back in minutes, so vagueness gets punished immediately instead of eventually.
What I'd tell someone trying this for the first time
- Vague requests produce vague results fast, so you find out you weren't clear in minutes, not days. Use that.
- The hard part isn't the syntax, it's still product judgment: what should this actually do, and why. That part doesn't get automated.
- Reviewing what got built matters more than reviewing how it got built. I couldn't audit the code line by line, so I audited outcomes obsessively instead: did the feature behave the way a real user would expect.
It's live and free at nickelsanddimes.app. If you've built something this way, or tried and it didn't click, I'd like to hear what was different about your experience.
Top comments (0)