Human in the Loop

Separating planning and execution in agent systems

Operable agents need a reviewable plan, effect-bound approvals, durable execution state, checkpoint-safe retries, and independent evidence of completion.