Reflection on using AI for personal projects
I am someone who enjoys the craft of programming.
I do it for a living. I do it most days as a creative outlet. I spend a lot of my downtime thinking about it. I think computers are neat, y'know?
You might have heard about this new AI fad that's taking over the industry, I certainly have. My workplace bought everybody Cursor licenses and told us we need to use it so I gave it a shot and it's was fine. Sometimes it can really speed things up, sometimes it makes a dog's breakfast of everything. I'd heard that Claude Code was much better than Cursor but nobody wanted to buy me a license and that was that.
Anybody who has heard me ramble about my various side projects knows I start a lot of cool stuff and I finish well, a lot less stuff.
In late November, I decided screw it, let's buy a month of Claude Code and give this AI business a fair shake. First, I turned it loose on a total greenfields project, I wanted to mess about making a MUD but didn't foresee having the time to actually do it any time soon. Initial results were exciting. After a few days of running Claude on a raspberry pi over SSH on my phone, I had like 10,000 lines of Python implementing a MUD accessible over websockets or telnet with a bunch of rooms, quests, a scripting language, admin building commands, the works.
Holy shit, right?
Encouraged, I turned to some unfinished projects. Earlier in the year I had started building a little visualiser for .FIT files from my Garmin smartwatch but I'm not much of a frontend person so I just kind of ran out of steam. Maybe I could slam together a bunch of features and round that out to the point where I could start using it! Maybe even cancel my Strava subscription and save myself some money?
This time, I ran two worktrees and built out several features at once. Suddenly I had a map and graphs and all sorts of stuff, Claude even suggested a bunch of new features.
Great!
Then I turned to a game I was working on in Godot. The story was the same, I started bolting on features at breakneck pace.
Christmas rolls around, my Claude Pro subscription expires, and I take stock of what I've built and like... It's flashy but I don't know, I don't feel like I want to work on any of it any more? In pursuit of velocity, I feel like I've killed the fun in those projects.
I started to try and tidy up the Garmin project so I can deploy it and start using it and I kind of want to revert everything back to the pre-ai codebase. It all works but I don't really understand how and I think in the process of trying to finish stuff at all costs, I've sort of forgotten the point of a lot of these side projects, which is to learn new things and have fun.