How would you decide whether a tool result should go back to the model or directly to the user?

Example Answer

The way I'd approach it in an interview is this: I decide based on whether the result needs synthesis, comparison, or policy-aware interpretation. If the tool output is already user-ready, deterministic, and safe to display directly, sending it through the model can just add latency and distortion.

If the result is structured, ambiguous, or needs to be merged with other context, giving it back to the model may be the right move. But then I want the output clearly labeled so the model treats it as data, not as a new instruction source.

The key is to avoid unnecessary model hops. Every extra pass through the model is another chance to lose precision or accidentally restate deterministic facts less clearly.

Common Poor Answer

A weak answer is saying tool results should always go back to the model so the UX feels conversational. Sometimes the safest answer is direct display.