AI Worker worker.md

AI Worker vs AI Agent

Workers and agents are often mixed together in conversation, but they solve different problems.

A worker is a bounded executor: it runs a capability with explicit inputs and returns explicit outputs. An agent is an autonomous loop: it sets subgoals, chooses actions, and keeps going until it decides to stop.

Both can use the same models and tools, but the operational risk profile is different.

Key ideas#

  • Workers optimize for predictability; agents optimize for autonomy.
  • Workers are easy to test with contracts; agents require behavioral evaluation.
  • Workers are safer by default because execution is bounded and permissioned.
  • Agents still benefit from workers: agents should delegate execution to workers.
  • If you can express the task as a contract, prefer a worker.

Diagram#

AI Worker (bounded)            AI Agent (autonomous loop)
-------------------            --------------------------
request -> execute -> response  goal -> plan -> act -> reflect -> ... -> stop

See also#

FAQ#

Can a worker contain multiple steps?

Yes, if the steps are still bounded and the contract stays clear. If it becomes open-ended planning, it is drifting toward an agent.

Do agents replace orchestration?

Not usually. In production, you typically still want explicit orchestration for observability, cost control, and failure handling.

Can an agent call workers?

That is the recommended shape: agents decide and delegate; workers execute bounded tasks with contracts.