Ask HN: What problems do you have building / managing AI in production

I’m working on Satori (https://www.usesatori.sh), a memory management library for AI agents. The plan is to make a self-hosted, OSS version.

As I’ve been building it, I’m wondering if there are more pressing problems or product directions to focus on first — for example giving agents access to workspace data, or to production logs and internal knowledge bases.

Curious to hear people’s thoughts.