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

Instruction: Describe how you decide where tool output should flow next.

Context: Checks whether the candidate can explain the core concept clearly and connect it to real production decisions. Describe how you decide where tool output should flow next.

Updated

Example Answer

I'd ask whether another model pass adds useful interpretation. A confirmed ticket update can be displayed directly with the stored status and operation reference. A search returning several relevant documents may need synthesis and citations.

Direct display still needs authorization, safe rendering, and appropriate redaction; raw tool output is not automatically user-ready. If I return data to the model, I'd bound it and treat embedded instructions as untrusted content. I'd keep deterministic facts, such as whether a write completed, anchored to the tool result rather than let the model embellish them. I'd compare latency and user understanding for both paths, including errors and partial outcomes, before choosing a default.

Why this works

Explains routing as a product decision while covering both rendering and model-context risks.

Practice follow-up

What should be shown directly when a write is only partially successful?

Your preparation path

Work through these questions in order. Read the answer aloud, then explain it in your own words.

1. Start with the foundations

Build the vocabulary and explain the core decisions.

  1. When should a model call a tool instead of answering from context? Free sample
  2. How would you explain MCP to a product engineer in one minute? Free sample
  3. What information should the orchestrator validate before executing a tool call? Free sample

2. Apply it to a real workflow

Practice diagnosis, validation, and everyday tradeoffs.

  1. Why do idempotency and retries matter more with AI-driven tool use? Free sample
  2. How do you keep permission boundaries clear when many tools share overlapping capabilities? Member answer
  3. A user revokes access to a connected app mid-session. How should the AI workflow behave? Member answer

3. Prepare for senior discussions

Explain failure boundaries, recovery, and production choices.

  1. Design a fallback strategy when tools are unavailable, degraded, or unauthorized. Member answer
  2. How do you implement MCP step-up authorization without losing scopes or retrying forever? Member answer
  3. An MCP request resumes after user confirmation. How would you validate its requestState? Member answer

Related Questions