Run Ops / IT
Run Ops and IT keep everything running: infrastructure, internal systems, and in manufacturing the production lines themselves.
Key processes
- Monitoring
- Incident management
- Change management
- Capacity planning
- Maintenance
KPIs that matter
- Uptime
- Availability of critical systems.
- MTTR
- Mean time to recover from incidents.
What AI agents can do here
- Anomaly detection in logs and metrics
- Automated runbooks
- Self-healing infrastructure
- Predictive maintenance
- Lights-out operations with humans handling exceptions
What good looks like
- Monitor what customers experience, not just servers.
- Write runbooks for frequent incidents and rehearse recovery.
- Change management: risky changes are reviewed and scheduled outside peaks.
Common pitfalls
- Alert fatigue that hides the important signal
- Changes during peak hours
- No owner for recovery when something breaks
Agent tasks and the human control they need
| Task | Maturity | Human control |
|---|---|---|
| Detect anomalies in logs and metrics and open incidents | Today | On-call engineers confirm and lead response. |
| Execute approved runbooks automatically (restart, scale, roll back) | Emerging | Limits on what automation may change, with instant human override. |
| Lights-out operations with humans for exceptions | Future (3–5 years) | Humans own capacity plans and major-incident decisions. |
Today: in common production use. Emerging: piloted by early adopters. Future: plausible within 3–5 years, not yet dependable.
End-to-end processes this department is part of
Related processes
Incident management
Detect → Triage severity → Contain → Resolve → Communicate → Blameless post-mortem
Mature teams measure mean time to recovery (MTTR), not just how rarely things fail.
Sales & operations planning (S&OP)
Demand review → Supply review → Reconcile gaps → Executive decision → Publish plan
Without S&OP, Sales sells capacity that doesn't exist, and Operations builds stock nobody ordered.
Plan-to-produce
Demand plan → Material requirements → Capacity planning → Schedule → Produce → Quality check
Predictive maintenance agents can flag failing machines weeks before breakdown.