Improving my workflow

Author: John

I've been thinking a lot lately about how to improve how I work with agents. I'm trying to figure out exactly what the biggest problem is for me. I think I've finally narrowed down on it. Starting features that are too broadly scoped and adding additional work on top of them. This is two problems really, starting feature branches without a clear scope (typically because the nature of the work is exploratory), as well as keeping dependencies in the worktree / feature branch itself instead of making it separate, mergable work. This ends up making managing multiple tasks at once a nightmire. I have to track in my head what the depenedencies are, what level of readiness they're at, etc. On top of that I'm now having to deal with making sure that my co-workers can get their updates pulled in as well as get access to my latest. This is hard.

I think beads + GitHub issues are going to help me here. I think the dependency tracking alone should be a huge improvement. It will allow agents to see what's blocking me and this lives outside any one feature branch. That's what I've been missing!

I also struggle with remembering to implement ideas that I've had. They get lost in my notes and unimplemented. These are features that for example will help with debug efforts, but I never seem to get around to them even though they aren't complicated to implement. This is where GitHub issues should help me. Everytime I get an AI assigned to me, or have an idea for something I want to implement, I can open an issue. These will be visible in my worktrees and I can ask: "what's next" to an agent and away we go.

Anyway, still working on this flow. Maybe I'll test out beads here at home to get a bit more used to it in a simple project. That ususally works out well for me, starting with a toy model and building on the understanding of it.