And, in a few months, this will all be under the hood, with summary reports and checkins. We won't care how the swarms split up the work. We'll just watch the results come together and answer questions.
I went from copying and pasting with ChatGPT Canvas, to Claude Artifacts, to Cursor w Claude. I haven't explored Rules yet, but have been using the Notepads for much of this. Much of my time is spent managing git commits/reverts, and preparing to bring Claude up to speed after the next chat renewal.
AI coding is like having a partner who is both the smartest person in town, but also a functional alcoholic, who's really, really good at hiding it.
LLMs act like they are working in a dark warehouse with a flashlight, and Altzheimers.
They can get an amazing amount of functional code done in a few minutes, and then spend hours trying to fix one detail. My standard prompt begins with, "Don't guess, debug!" They have limited resources and will bs you if they can.
For longer projects, since every prompt is almost starting from scratch (they do have a limited buffer, which will make it easy to become complacent), if you get into repeated debugging sessions, it will start creating new functions instead making exisiting functions work, and code bloat is tremendous.
Perhaps Rules work, but I've given up trying to get it to code in my style. I'm trying to have AI do all the coding so I can just be the "idea" guy ("vibe" coding), so I'm learning to let go and let it code in ways the I would hate to maintain. It's working from code examples that don't use my style, so I'm not going to keep fighting it on style (with some execptions like variable naming conventions).
