Your goal: show that you treat a prompt as something you design, run, read and test, not a
paragraph of wishes.
How this works (about 30 minutes)
- Felix asks which role you are interviewing for (product, engineering or other), then one
question about your approach to prompt engineering.
- Then you write a prompt on the notepad. Say "run it" and Felix plays back what a model
returned, one line at a time. Ask for the next line when you want it. The outputs are made up
for the exercise.
- Think out loud: name what is wrong with the output and which part of the prompt caused it.
- A curveball, then a trade-off: ship, hold or change?
- No hints and no scores during the call. Your report comes after.
What scores well
- A plain account of what a prompt is and why a missing fact turns into a confident guess
- A prompt with parts you can name: the goal, the facts, rules (including what to do when
unsure), an example or two, the format, the input marked as data
- Reading each output, naming the failure and the cause, and changing one thing
- A test set with awkward cases and a bar set in advance
- Saying where prompting stops, and what could go wrong with untrusted text
Tips
- Take Gen AI Coach · Prompt Engineering first.
- Ask for the test email when you want to see one.