Tell me about a prototype that ended up in production. What did you have to fix first?

How to answer

In FDE work the prototype can become the production system. Sierra’s April 2026 post on its AI-native interview for engineering candidates says the Review stage includes a discussion of the path to production. ^[1]Source 1The AI-native interview (Vijay Iyengar, Arya Asemanfar, Angie Wang)PublisherSierraSource typecompany blog The question asks for your order of repairs, and why.

1. How it crossed over. Who started relying on it, and what it touched by then: real data, a real process. If nobody decided it, say so plainly, and say when you told the customer it was now production, with the support you could promise meanwhile.
2. The inventory. List the shortcuts you knew about, then rank them by what failure would cost. A defensible order: anything that can silently produce wrong data, then whether you would know it broke, then access and secrets, then anything you cannot recover, then whether someone other than you can run it, then performance and code quality. Do the fixes that take an hour first, wherever they rank: a nightly backup and an alert buy time for the rest.
3. What you fixed first. The top one or two, with the reason: “It wrote to their ledger with no idempotency key, so a retry double-posted.”
4. How you fixed it live. A feature freeze, one change at a time, backups before each, a test around the dangerous path before touching it.
5. The handover and the habit. Who runs it now, and what you record differently when you build the next prototype.

Say the ranked list out loud before you describe any fix; it shows you knew the shortcuts rather than found them by accident.

The trap is the rewrite: “we threw it away and rebuilt it properly” skips the triage the question asks about, and leaves users on the risky version while you rebuild.