Agent Architecture

writing

An agent needs enough context to reason and enough control to act safely. Treating them as one system hides different failure modes.