What does testing in production mean to you, and how do you do it without breaking things?

It means validating behavior against real traffic and real data that no staging environment can reproduce, using controls that limit the blast radius. The techniques are feature flags with progressive rollout, canary releases, synthetic monitoring running real journeys continuously, shadow traffic, and automated rollback on metrics defined in advance. It supplements pre release testing rather than replacing it.

It means accepting that staging will never have production's data volume, traffic mix, third party behavior, or the twelve years of strange records real customers have created, and deliberately using production to learn what only production can teach. What makes it responsible rather than reckless is blast radius control. New behavior goes behind a flag, off by default, enabled for internal users first, then one percent, then progressively wider, with the metrics that trigger an automatic rollback defined before the release rather than argued about during it. Alongside that I run synthetic monitoring: the critical journeys, signup, login, checkout, executed against production continuously so we find out before customers do, and so a broken third party is visible immediately. I also like shadow traffic for backend changes, where the new path processes real requests and results get compared without anything being served. What I always say clearly is that this supplements the earlier layers rather than replacing them. Unit and integration tests still gate the merge. Using production validation as a reason to skip them is how you ship broken software very confidently.