Core answer
Start here. This is the direct spoken answer to practice first.
Why this question matters
The first minutes of an incident shape both user impact and the quality of the investigation. Calm ownership, a shared timeline, and controlled mitigation prevent random debugging from making the failure larger.
I would first confirm the incident and define user impact: what is broken, who is affected, when it started, and how severe it is. Then I would assign an owner, open a shared channel, and start a timeline. In parallel, I would look for safe mitigation such as rollback, disabling a feature flag, scaling a bottleneck, pausing a job, or routing away from a bad instance.