Scaling on-call from 8 to 150 engineers without burning anyone out

When you go from 8 engineers to 150 in 18 months, on-call breaks. Here's what we did.

The old way

Round-robin across the whole team. One engineer per week. It worked at 8. It collapsed at 40. People burned out, alerts got ignored, and our MTTR doubled.

The new way

Tiered response with AI as the first responder. Every page hits an AI runbook executor first. It correlates signals, runs diagnostics, and either auto-resolves (60% of cases) or pages a human with full context attached.