Agent System Design Interviews

The signature round for agent-engineering roles. You'll be asked to design an agentic system end to end and defend the trade-offs.

The framework (use this every time)

1. Clarify the task envelope — what does the agent do, how autonomous, what's the blast radius of a mistake?
2. The loop — model → tool calls → observations → repeat. Define the stop conditions first.
3. Tools — which tools, schemas, error handling, sandboxing.
4. Context & memory — what goes in the window each turn; summarization/compaction strategy; persistent memory.
5. Orchestration — single agent vs sub-agents; when to fan out.
6. Evals & observability — how you know it works; traces, success metrics, regression suite.
7. Safety & cost — permissioning, human-in-the-loop gates, token budgets.

Practice prompts

• Design a coding agent that fixes failing CI on a large monorepo.
• Design a customer-support agent with access to refunds (irreversible actions!).
• Design a research agent that produces a cited report from the live web.
• Design the eval harness for any of the above.

Deep material in the handbook

• The Agent Loop
• Memory Systems
• Tools & MCP
• Orchestration
• Agent Evals

Work in progress

This page is a scaffold — worked examples and depth are being added. Want to help? Grab a good first issue or read the contribution guide.