From reading to practice
Interview Arena
For engineers preparing to explain systems design decisions, including what happens when the clean path fails.
ArchCrux essays are free to read. Interview Arena opens in a separate app. Start here with a design question grounded in the publication’s writing on agent recovery.
Open Interview ArenaIllustrative design exercise, not an Arena-generated report
The refund succeeded. The worker disappeared.
The payment provider completed a refund, but the worker crashed before recording the result. How should a replacement worker continue without issuing another refund?
- Identify what the workflow knows, what the provider knows, and which result is still uncertain.
- Define a stable operation identity. Explain how retries and concurrent replacement workers use it.
- Describe reconciliation when the provider is unavailable or its idempotency window has expired.
- Choose the evidence an operator needs before retrying, escalating, or declaring the operation complete.
Work through the assumptions and tradeoffs before drawing the components. This exercise is editorial material; it does not represent an available Arena scenario or generated feedback.
Read the underlying failure model
Designing Long-Running AI Agents That Survive FailuresContinue with Start Here for more production engineering essays.