Péter István Fülöp’s Post

In my circles, surprisingly few people use git worktree — even though it’s been one of the most useful workflow changes I’ve made when working with coding agents. I use a simple hub + spokes setup: one central repo as the hub, with each branch checked out into its own “spoke” directory. Instead of constantly switching branches and stashing, you check out each branch into its own directory. They share the same Git history, but each has its own clean working copy. That means you can run a separate agent in each — one on an auth refactor, another on a new endpoint, a third writing tests — all in parallel, without collisions. Your main checkout stays untouched. A few things that surprised me: • In my experience, agents produce better output starting from a clean state • When one goes sideways, I just remove the worktree (after pushing anything useful) and start over • Same branch point, two worktrees, two agents — diff the results, keep the better one I wrote up the full workflow here: https://lnkd.in/dDD6V6w5 I’m curious: how are you handling parallel development with agents today? Do you use worktrees, separate clones, or something else? Feel free to share in the comments. #AIEngineering #AIAssistedDevelopment #EngineeringWorkflows

  • No alternative text description for this image

To view or add a comment, sign in

Explore content categories