SLO Definition
What does ‘working’ mean for an agent? Typical SLOs: resolution rate > 70%, p95 latency < 3 seconds, error rate < 2%, cost per conversation < $0.30. Define; measure; manage.
Error Budgets
SLO of 99% available = 1% error budget per month. When budget is healthy, ship aggressively. When budget is burning, freeze non-critical changes until recovered. Classic SRE discipline applied.
Chaos Testing
Inject failures to verify resilience. LLM provider unavailable? Agent degrades gracefully. Retrieval system slow? Response times extend but don’t fail. Test these explicitly; don’t discover in production.
On-Call Rotation
Ops engineers on rotation. Incident response playbook specific to agent failures. Document what good looks like. Tools and access pre-provisioned, not assembled during incidents.