My current AI assistance strategy

So basically since cursor came out I've tested many different ways of augmenting my workflow with the help of AI. I've tried "AI-native" editors, CLIs, rolled my own plugins etc - but for now my setup has set on a quite simple configuration.

My main focus is simply to offload boring or mundane tasks and adding an additional layer of "self review" before taking other developers time.

The tools I use

Since Omarchy included try in an update (version 3.something IIRC?) I fell in love with it. I use it (and git worktrees) to handle several branches of a repository in some "WIP" stage at the same time. Many of the services I work on are structured as monorepos, and you can probably ignore this if you usually deal with polyrepos.

Opencode is currently my TUI of choice. I started out kind of disliking the TUI approach to agentic development because I feel that my state of mind is different in editor vs in a UI like this, but I've changed my opinion here.. I think it helps me to keep on track regarding the fact that the output of the agents is still just a blueprint or suggestion for me to refactor or fix in my editor - where my "developer mind" is activated.

Avante.nvim for in-editor assistance. I find myself using this less and less as I further adopt a TUI -> Editor workflow where my editor step is much more like a review or refactoring stage, and therefore usually require my manual labor and some brainpower.

This allows me to produce a lot more than before.. Since so much code that we write on a day to day basis is "basic", or plumbing, it really allows the most of my mental focus to go towards higher value work instead!

Out.