1 min lesson
If you've ever built an interface for verifying machine output
Use "A moderation queue, a diff viewer, an eval dashboard" to say what you would do next.
Step 1 of 2
When you discuss your background, pick stories that rhyme with these three shapes. A 0-to-1 thing you scoped and shipped fast beats a multi-year maintenance project, even if the maintenance project was bigger. If you've ever built an interface for verifying machine output - a moderation queue, a diff viewer, an eval dashboard - lead with it, because it's the closest analog to the PR-review invention they prize.
Don't describe the A/B example as just "I'd ship it behind a flag and watch the numbers." At millions of users on an agent product, the hard part is choosing a metric that doesn't reward a worse experience - accept-rate can climb while quality drops if suggestions get shorter and safer. Name that failure mode and you sound like someone who has actually run experiments on a model product.