1 post
Weighing approaches to memory extraction for a coding agent, per-turn hooks, a background API worker, and manual human-in-the-loop, and why the last one is the one that actually held up.