An agent is a language model inside a controlled loop. The decision "agent or workflow?" was never a question about architecture — it was a question about trust in the runtime. With deterministic execution, it goes away.
