62% of on-call engineers burn out within 12 months (PagerDuty 2026). That is not a people problem. It is an engineering design problem. Three root causes every burnt-out rotation has: 1. Alerting on symptoms instead of causes CPU spikes, queue depth, request rate - none of those are incidents. All of them wake people at 3AM. 2. Stale runbooks Page fires. Runbook points to a renamed service. Engineer reverse-engineers the system on adrenaline and no sleep. 3. Symmetric shifts for asymmetric load Weekends are the Super Bowl. Weekdays are a sprint. Same rotation treats both identically. Four fixes the best teams ship: 1. Error budget Cap unplanned on-call work at 25% of the week. Above that, feature work stops. Google SRE wrote the playbook. 2. Alerts linked to SLOs If it does not correspond to a user-observable SLO breach, it is noise. 23% of on-call time is false positives (Blameless 2026). 3. Runbook-or-delete Every production alert ships with a current runbook, or the alert is deleted. You will lose half your alerts. That is the point. 4. Load-aware rotations Peak windows need heavier coverage or shorter shifts. One-size-fits-all punishes whoever draws the peak. If you lead engineering, on-call design is your job. Not HR. Not your EM. You. This week: open the dashboard, count the pages, ask how many should have been handled by the platform, not a human. #EngineeringLeadership #OnCall #SRE