agent-orchestration

I have rage quit integrating worktrees into my agentic orchestration at least three times. It’s a promising idea, having many agents work on the same repo, but is non-trivial to do and different for every repo.

For now I’m sticking to 2-3 agents on the same repo, in different areas of the codebase. Model seems to handle the other work going on, tho, Opus 4.8 appears more curious about it than others.

Really pleased to see others like Isaac Dedini experimenting the area - a great post on why “just use worktrees” is nonsense.

This is agent orchestration, but Loop Engineering is for sure a better term. My notes:

• I don’t have a requirement for scheduled tasks as yet
• Worktrees are close to impossible
• I use assist for skills & cli-wrapping, though I do use Claude connectors
• Sub-agents - no thanks, i’ll make the deterministic thing (assist) control the non-deterministic thing (claude).
• State/memory - this is assist backlog