AI-103 Practice Question 84: Agent Run Requires Action

A run's documented lifecycle is queued → in_progress → (optionally) requires_action → completed, and Microsoft is explicit that a language model never executes a function itself: "When the agent determines a function is needed, the run status indicates requires_action with function call details. Your application code must detect this, execute the appropriate function, and submit the results back to the agent via the API." A loop that only waits during requires_action leaves the run permanently parked in that state (nothing ever supplies the output the model is waiting for) until the documented 10-minute run expiration fires and the run becomes expired, matching exactly what the team observed.

- B is wrong because a longer polling interval only changes how often the loop *checks* the status; it never executes the function or submits an output, so the run still has nothing to advance past requires_action and still expires at the same ten-minute mark. - C is wrong because streaming changes how status updates are delivered to the client, not the run's expiration window or the requirement to submit tool outputs for a pending function call. - D is wrong because a new run would ask the same question again, generate the same function call, and land back in requires_action immediately; it never supplies the missing output either, it just repeats the failure.

Reference: Use Foundry Agent Service with function calling (classic)