PD Reliability Platform and Paige, PagerDuty's SRE Agent
This domain tests the platform-wide reliability model and Paige, PagerDuty's SRE agent. It covers how services, teams, incidents, analytics, postmortems, and status communication fit together, and how Paige uses incident context to help responders. The exam emphasizes human-in-the-loop use and data quality.
Core rules
- The Reliability Platform connects on-call, incident response, status communication, postmortems, and analytics around the service object. Analytics and postmortems depend on consistent service and incident data.
- SRE practices include defining SLOs and error budgets, reviewing incidents blamelessly, tracking action items, and using trends to prioritize reliability work.
- Paige is an AI SRE agent that works with incident context. In PagerDuty, Paige can summarize an incident, answer responder questions, surface related and past incidents, and suggest next steps or remediation. It is assistive: responders validate and execute actions.
- Paige's usefulness depends on permissions and data: integrations, runbooks, service ownership, and incident history must be accurate.
- Paige does not replace on-call ownership, escalation, or human judgment; it reduces investigation time.
Worked example
A checkout latency incident fires. The responder asks Paige, "What changed and what should I check?" Paige summarizes the incident timeline, lists recent deployments and related past incidents, and suggests checking the database connection pool and a rollback. The responder verifies the deployment and executes the rollback; Paige does not act on its own. Reasoning: Paige provides context and recommendations; the human confirms and acts.
Traps
- Treating Paige's suggestions as authoritative; always validate against telemetry.
- Expecting Paige to fix missing data; poor service ownership or incomplete runbooks limit answers.
- Confusing the Reliability Platform with a monitoring tool; it responds to and learns from incidents, not replacing observability.
- Forgetting postmortem action tracking; analytics without follow-through does not improve reliability.