GenAI interview guides / AI Agents & Tool Use Interview Guide
AI Agents & Tool Use Interview Guide
Design reliable agent systems with explicit tool contracts, bounded autonomy, state, observability, and recovery.
Agent loop
Describe state → decide → act → observe → update, with termination, budgets, retries, and escalation.
Tool contracts
Use narrow schemas, validation, permissions, idempotency where possible, timeouts, and structured errors.
Reliability
Checkpoint long tasks, deduplicate side effects, distinguish retryable failures, and trace model/tool transitions.
When not to use agents
Prefer deterministic workflows when the task graph is stable; use agentic planning when steps depend materially on observations.