A practice prompt we wrote. No company or candidate report names it, so it carries no company tag.
How to answer
Some postings put incidents in the job. As of September 2026, OpenAI’s Financial Services FDE posting in New York asks the FDE to own reliability, observability and on-call readiness for the production AI systems it ships, and Baseten’s Forward Deployed Engineer (Training) posting asks for willingness to join an on-call rotation. Source 1Forward Deployed Engineer (FDE), Financial Services- NYCPublisherOpenAI (Ashby)Source typecompany job postingSource 2Forward Deployed Engineer (Training) @ BasetenPublisherBaseten (Ashby)Source typecompany job posting
The question names its own center: what you did first. A strong first move stops the damage before explaining it. Five moves, in order:
- What broke and who felt it. Impact in the customer’s terms, and how you found out.
- The first move, and why that one. Rollback, a flag off, failover, shedding load, or declaring the incident and handing out roles. Say what you chose not to do: “I didn’t open the diff.” Say why it was safe even if your guess about the cause was wrong.
- The first message. Who you told, how soon, what you said before you knew the cause, and the time you promised for the next update.
- Recovery and cause, briefly. Cleaning up the damage already done, then the cause in one or two sentences.
- The follow-up that prevents recurrence. A change to the system, not a reminder to be careful, and evidence that it works: it has fired since, or a test now fails on the old behavior.
Give intervals, not adjectives: minutes from alert to mitigation, and to the first customer message.
Hold back the debugging detail until you are asked; it is the part you will most want to tell.
The trap is the hero-debugger story, where the incident is a puzzle you solved and the customer appears only at the end. If you rolled forward with a fix when a rollback was available, say why.
Follow-ups
What the interviewer may ask next, once your first answer is on the table.
- What did you decide not to do in the first few minutes, and what would it have cost to do it?
- Who else did you pull in, and what did you hand to them instead of doing it yourself?
- Has the follow-up you added fired since, and what did it catch?
Where answers go wrong
- Leads with the root-cause hunt and the clever debugging, and mentions mitigation and customer communication in passing, after the cause was found.
- Ends with a postmortem action such as “we added monitoring”, with no sign it was built, what it watches, or whether it has caught anything since.
Answer this in two minutes
Write the answer you would say out loud. The clock starts with your first word.
Illustrative answer about a fictional project
I was the lead engineer on order ingestion for a medical-supply distributor: electronic orders from about 600 hospitals flowed through our pipeline into their warehouse system. At 6:40 on a Monday, the east warehouse’s floor lead called me: the morning pick lists were about a third shorter than a normal Monday, while their customer service line was hearing from hospitals about orders placed over the weekend.