How to Use PagerDuty Paige AI SRE Agent to Diagnose Incidents

Article

How to Use PagerDuty Paige AI SRE Agent to Diagnose Incidents

PagerDuty announced Paige, its AI SRE Agent, and the PD Reliability Platform as generally available on November 3, 2026. Paige examines incident events, runbooks, standard operating procedures, diagnostic data, logs, and related incident history. It then presents likely causes, incident context, and suggested investigation or remediation steps. The launch is described in PagerDuty’s November 3 announcement and the company’s Paige product overview.

This guide explains how to use PagerDuty Paige AI SRE agent to diagnose incidents while keeping human responders responsible for evidence review and approval. The workflow suits engineers learning incident response, teams improving runbooks, and professionals evaluating AI-assisted operations. Readers building wider AI-enabled software can also review these AI development services for related project context.

What Paige brings to incident diagnosis

AreaDocumented capabilityPractical meaning

EvidenceAnalyzes incident events, runbooks, SOPs, logs, diagnostic data, and related incidents.Diagnosis can start from connected operational context instead of one alert alone.
InvestigationSurfaces likely root causes, recalls related incidents, and recommends diagnostic steps.Responders receive a structured starting point for questioning and testing hypotheses.
AccessAvailable through the Operations Console, PD Reliability Console, incident details, Slack, and Microsoft Teams.Teams can investigate from their normal incident-response surface, subject to configuration.
ControlsOperates within configured permissions, with responders expected to review AI output.Use Paige as an evidence assistant, not as an unchecked incident fixer.

The key value is evidence collection and reasoning support. Results depend on the operational information available to Paige, so the quality of runbooks, logs, ownership data, and incident history matters during every investigation.

How to use PagerDuty Paige AI SRE agent to diagnose incidents

Start by opening an active incident in the Operations Console or PD Reliability Console. Select the Paige tab, then wait for its incident summary and suggested next steps. You can ask follow-up questions, provide missing information, use available investigation nudges, and attach supporting files. This workflow is documented in PagerDuty’s Paige documentation.

Web and incident-page workflow

Give Paige a focused starting point. Ask it to identify the leading hypothesis, show the evidence supporting that hypothesis, and recommend the next diagnostic check. You can also ask it to compare the current event with related incidents or explain which runbook step applies. These questions turn Paige’s summary into an investigation path that a responder can inspect.

Access depends on account configuration. PagerDuty states that Operations Console access requires AIOps and PagerDuty AI, while access through the incident details page requires PagerDuty AI. Administrators configure PagerDuty AI features and team access through the documented settings. Teams planning broader incident-response tooling can also review project consultation services for implementation context.

Slack workflow

Slack users must configure the Slack integration and PagerDuty AI first. During an incident, select the Paige Triage button or ask an incident-related question with @pagerduty. Paige can then use the connected incident context to summarize the situation and suggest investigation steps. The same documentation lists Microsoft Teams access, but identifies it as Early Access.

Use a safe test incident when learning the workflow. Attach a relevant runbook or diagnostic file, ask Paige for its top hypothesis, and compare the answer with the incident timeline. This practice helps beginners see whether connected operational data gives Paige enough context to produce a useful diagnosis.

More details

Prepare Paige for a first investigation

Before opening an incident, confirm that your PagerDuty account includes the required AI features. PagerDuty documents Paige access in the Operations Console for accounts with AIOps and PagerDuty AI. Access from an incident details page requires PagerDuty AI. PagerDuty AI is included with PD Reliability Platform plans and is also offered as PagerDuty Advance on Professional, Business, and Enterprise plans. Check the PagerDuty AI documentation for current administrator and team-access requirements.

Check access before opening an incident

An administrator may need to enable PagerDuty AI features, configure team access, and connect the relevant incident-response integrations. This matters because a responder can see Paige in one PagerDuty surface but lack access in another. Verify the intended workspace, team permissions, and incident visibility before using Paige during a live event.

Slack requires extra configuration. The PagerDuty Slack integration and PagerDuty AI must be set up before responders can use the Paige Triage button or ask an incident question with @pagerduty. Microsoft Teams access is documented as Early Access, so availability may depend on the account and feature state. The Paige documentation lists the supported entry points and setup flow.

Choose evidence to attach

Prepare information that narrows the investigation: the affected service, approximate start time, recent deployment or configuration change, relevant runbook, and diagnostic output. Paige can reason over connected incident context and supplied material, but it cannot infer missing operational facts reliably. A focused evidence bundle helps it compare the current event with related incidents and identify a testable next step.

Teams building internal tools around incident data may also review application development services for related implementation context. Keep the first test bounded: use a non-critical incident or controlled exercise, then compare Paige’s proposed investigation path with the timeline and available runbook.

Turn Paige’s summary into a testable diagnosis

After Paige summarizes an incident, treat its first explanation as a hypothesis. Ask which service appears affected, when the problem began, what evidence supports the leading cause, and which alternative explanations remain possible. PagerDuty describes Paige as able to connect incident context with runbooks, standard operating procedures, logs, diagnostic data, and related incidents. Its product explanation also highlights hypothesis testing and visible reasoning. Read the PagerDuty explanation of Paige’s reasoning workflow for the product’s documented model.

Ask for scope, evidence and alternatives

Use questions that produce checks, not vague certainty. For example, ask: “What is the leading hypothesis, and what evidence supports it?” Then ask: “Which timeline event would confirm or weaken that hypothesis?” A third question can request the next diagnostic step and the expected result. This turns Paige into an investigation guide that helps a beginner follow the same logic as an experienced responder.

Ask Paige to compare the current incident with related historical incidents when the service, alert pattern, or failure message appears familiar. Request the matching symptoms and the important differences. Similar incidents can suggest useful paths, but the current timeline must decide whether the comparison applies.

Supply missing context

If Paige cannot separate competing causes, add the missing operational detail. Provide the affected service, recent deployment, configuration change, error sample, time window, and relevant runbook step. State what has already been checked. A compact evidence bundle reduces ambiguity and makes its recommended next action easier to evaluate.

Keep prompts tied to one incident question at a time. Ask for a diagnosis, evidence, and next check before asking for remediation. This sequence helps teams build a repeatable training exercise and exposes gaps in runbooks, ownership metadata, or observability data. Learners studying practical project consultation workflows can use the same structure when documenting an incident investigation for a technical project.

Further reading

Review Paige’s evidence before approving remediation

A useful diagnosis must show more than a likely cause. It must connect that cause to evidence, explain competing possibilities, and identify a bounded next check. PagerDuty’s SRE Workbench documentation describes structured findings that include hypotheses, confidence descriptions, supporting evidence, and alternatives considered. Use this evidence-first pattern when reviewing Paige’s incident summary.

Check the evidence chain

Start with the incident timeline. Confirm that Paige’s proposed cause matches the alert start time, affected service, recent changes, and observed symptoms. Open the linked logs, diagnostic results, runbook steps, or related incidents that support its reasoning. If evidence comes from an older incident, compare the service version, deployment context, and failure pattern before treating it as relevant. The SRE Workbench User Guide explains how investigations can retain evidence and present alternatives for responder review.

Ask Paige to identify the strongest evidence and the evidence that would weaken its leading hypothesis. Then perform one low-risk check. For example, compare error rates before and after a deployment, inspect the named dependency, or verify whether the suggested runbook condition is present. Record the result in the incident timeline.

Approve only bounded actions

Separate diagnosis from remediation. A recommendation to inspect a configuration is different from a recommendation to restart a service or change production traffic. Require a human owner, a defined scope, a rollback path, and an observable success condition before approving a consequential action. PagerDuty states that Paige operates within configured permissions, but its AI guidance also tells responders to fact-check generative-AI output. Review the PagerDuty AI documentation before establishing team controls.

For learners building incident-response tools, this review method creates a practical project: store the timeline, evidence links, hypothesis, test result, and approval decision in a small application. Relevant application development guidance can help frame that implementation without treating Paige’s answer as an automatic production decision.

The right rule for using PagerDuty Paige AI SRE agent to diagnose incidents

Use PagerDuty Paige AI SRE agent to diagnose incidents when the incident has enough connected evidence for a responder to test its reasoning. Do not treat a plausible explanation as a production decision. The practical decision rule is clear: use Paige to narrow the investigation and recommend a bounded next check, but require a human responder to verify evidence before approving remediation.

This rule keeps Paige in its strongest role: an investigation assistant that helps organize incident context, identify likely causes, and suggest what to examine next. It also prevents a weak runbook, incomplete log coverage, or missing deployment context from becoming an unchecked operational action. The responder remains responsible for deciding whether the evidence supports the proposed change and whether the action has a defined scope and recovery path.

Your next step is to run one controlled practice incident. Choose a non-production service, connect one relevant runbook, and provide a limited time window of logs or diagnostic data. Ask Paige for one leading hypothesis, the evidence supporting it, one alternative explanation, and one low-risk check. Record the response beside the incident timeline, perform the check, and note whether the result strengthened or weakened the hypothesis.

If the exercise produces vague answers, improve the runbook or observability context before expanding usage. If it produces an evidence-backed investigation that a responder can verify, repeat the exercise with another incident pattern. This gives your team a measurable adoption test without assuming that AI assistance alone will resolve every incident.

Additional image