Retaining system design context over time
retained text
I've been using AI to write a large amount of the code I ship for a while now, but a problem that I see consistently is context retention across sessions. New agents don't have all the necessary context for why an architectural decision was made a few months ago, if an existing data structure can be reused, etc. It feels like it's either eating up tokens trying to gather this context, or requires me to explain it again. It also takes me some time because I need to go back and review the implementation to get a clear picture so I can explain it accurately to the agent. I know I could probably just use an AGENTS.md file to start, but my team works in a fairly large monorepo and I feel that it would just bloat over time. Are there any more sophisticated or better organized approaches that anyone is using on their teams?
document fetched 2026-11-02 00:18 · canonical https://reddit.com/r/AI_Agents/comments/1xl8wvf/retaining_system_design_context_over_time