Curious what people think about enforcing policies at runtime for AI agents.

Feels like once agents start calling tools or APIs, there’s no clear control layer. Wondering if others have run into this or are solving it differently.

undefined

[dead]