Ship a Claude Application You Can Defend
The capstone is not a chatbot demo. It is a bounded application with a wire contract, security boundary, eval evidence, and recovery plan. Translate one user workflow into explicit functional and operational requirements. Integrate structured output, tools, policy, tracing, and final-state verification. Produce an architecture record that defends tradeoffs and rejected alternatives. Build an eval plan with normal, boundary, failure, and adversarial cases. Write a runbook for timeouts, ambiguous side effects, denials, and regressions. Demonstrate readiness through runnable tests instead of confident prose. Build a support application that answers one narrow question: The application must: Extract and validate an order ID. Refuse instructions that attempt to bypass policy or request secrets. Use one read-only order lookup capability. Return a strict response contract. Escalate when the identifier is missing or the order cannot be verified. Emit a redacted trace. Pass deterministic and behavioral eval cases. Ship an architecture record, eval plan, and runbook. This appears smaller than a general support agent. That is the point. Production quality comes from closing the loop on one useful job before widening capability. Functional requirements: Accept a natural-language status request. Recognize an order ID in the approved public format. Query only the authenticated user's visible order store in a production implementation. State a verified status or an explicit inability to verify. Never claim a shipment,…
Ship a Claude Application You Can Defend: The capstone is not a chatbot demo. It is a bounded application with a wire contract, security boundary, eval…
This free lesson is part of the AI Engineering from Scratch curriculum. Read the full explanation, run the lesson code, and verify the result in the interactive reader or from the repository source.
Browse the complete course catalog or open this lesson on GitHub.