A really good post on the current AI landscape. Some thoughts:
Maybe that’s all any documentation is. Messages to future confused versions of ourselves.
I totally agree with this. I find that the more documentation you can feed AI, the better. I’m constantly noting “check off @TODOS.md, note this in @README.md”
If readme gets too big, use subagents to clean up the documentation and move it into a /legacy. File folders are free!
I think overall, there’s an uncertainity about how to build here. It’s a ton of getting vibes from other smart people on what model works, what’s the best way to prompt, set up documentation, really anything. Throwing Tokens at Problems is really the best way forward, which is crazy. But here we are.
Another freebee on adequate software. Inspirational to start building my own tools. The more technical, the more maintainable using base components gets (SQL, Python). I’m super wary of any direct manipulation (say, excel).