I thought I’d try an experiment with letting Claude Code work on a fresh project. I’m not diving right into coding - I’m using Claude Code to write the specs first.

I’m blown away. It’s like having a short-range time machine. I got so many pages of user stories, tech requirements, roadmaps, mvp vs later versions, and all that stuff. Done in a few hours over two evenings. Yes I hit the limits way before the 5-hour window, but on the plus side I went to bed instead of sitting up half the night, so there’s that.

What would have taken me days of typing, Claude just magicked into existence with a snap of its virtual fingers. I review every line of it and still save oodles of time, plus I get to ping-pong about my ideas and refine them along the way.

Using Claude Code instead of just browser-Claude was the real boon. Working with markdown.md files is fast as hell. Running it on my Windows desktop using WSL to get a new one Linux session that maps to my home folder, and simultaneously using Obsidian in Windows to read and edit the output. That sounds a bit roundabout but it’s very efficient, and as a side effect I am beginning to grok Obsidian and loving it. A powerful combo, plus it syncs with my phone. Add git to the mix as a finishing touch.

Claude can execute git commands, it can spin up a Docker instance to run the code it will eventually write, and I get to see it my browser, all on localhost.

I won’t be surprised if the prototype it produces is shit. But I might be pleasantly surprised that maybe it isn’t.

(I wrote this text myself.)