After using coding agents heavily for the last few months and I've noticed something I guess most of us find frustrating: the agent, the model, the harness aren't necessarily the problem. Problems show up when a complex task gradually loses its original priority after several rounds of steering, compaction, or get handed over into new sessions.
I'm curious whether other people experience this too. How do you currently keep a long-running coding task coherent, even across multiple agent sessions?
In my own workflow, I now use a tool that I built for myself to solve this issue and it works surprisingly well! It's now at a stage where I am ready to accept private alpha testing. Feel free to DM me to get an invitation code!
Top comments (0)