← Glossary
Glossary · AI and agent workflows
Action and observation loop
The repeated cycle in which the model proposes an operation, the harness executes it, the result returns as an observation, and the model decides again.
The loop ends when the model produces a final message or when the harness stops it because of a limit, an error or a person. Neither ending shows that the request was met; that is checked afterwards against the record and the completion evidence. Descriptions of this cycle as continuing until the objective is fulfilled describe an intended design, not a guarantee.