Major

PagerDuty AI agents: API guide, apps and automations

A researched guide to PagerDuty's native AI, API permissions, and incident workflows on Major, with durable evidence, human approvals, and failure handling.

Ritvij GautamUpdated
Major x PagerDuty logo lockup

PagerDuty AI agents help responders investigate incidents and coordinate response. This guide covers PagerDuty's own AI, its API, and how to build governed incident work on Major with durable state, cross-system judgment, and human approvals.

For SRE and platform leaders, the useful starting point is a triage app that remembers evidence and review decisions, not another conversation that disappears after the incident.

What PagerDuty does for incident teams

PagerDuty manages the operational handoff from a detected problem to a responder. Its REST API overview distinguishes managing incident and configuration data from ingesting monitoring events, which belongs to the separate Events API.

SRE teams investigate service failures, on-call engineers acknowledge and coordinate response, and IT operations leads track unresolved work. A service represents an application, component, or team against which incidents can be opened, as the service reference explains.

The practical problem is often the work surrounding the page: matching a Datadog monitor to the affected service, checking whether a Jira issue already tracks the failure, and preserving what the responder approved. These are specific jobs with different access requirements, not a reason to give an agent unrestricted incident administration.

Major adds a deterministic app layer for that surrounding work. An agent can build the evidence viewer and action ledger once, then use the app on later incidents while reserving its reasoning for relevance, severity, and the proposed next step.

PagerDuty's built-in AI and where Major adds value

PagerDuty already has an incident-focused agent. Paige, its SRE Agent, ingests event data, runbooks, and diagnostics, surfaces likely causes, recommends troubleshooting and remediation steps, and recalls similar incidents. The documentation describes access through PagerDuty interfaces and Slack, with availability depending on the relevant entitlements.

PagerDuty also documents Intelligent Alert Grouping, which uses machine learning to consolidate related alerts. That is useful before a responder starts assembling a brief: fewer separate incidents can mean less duplicated investigation.

Do not position Major as filling an imaginary absence of AI or cross-tool investigation in PagerDuty. Paige itself can work with connected operational context. Major's role is the organization-specific software around that investigation: a durable evidence ledger, a custom review queue, and an approval path spanning security, engineering, and operations.

For example, an agent on Major can judge whether an Okta anomaly warrants an incident, while an app retains the evidence and a workflow asks the security responder for approval. Native incident intelligence remains valuable inside PagerDuty; the organization's cross-system operating process lives in software it can inspect and reuse.

How PagerDuty connects to Major

PagerDuty connects through Major's Pipedream connector. There is no native PagerDuty connector. IT sets up that connection once and verifies that the selected connector actions cover the intended reads and writes before enabling a workflow.

The underlying REST API uses https://api.pagerduty.com. PagerDuty supports account and user API tokens, but do not assume every authentication option described in its API is exposed by a particular connector action.

Set up the operating boundary explicitly:

  1. Connect a read-only credential for gathering incident context.
  2. Keep write-capable actions separate and available only to the approved execution path.
  3. Store credentials centrally in Major's connector vault, never in code, agent prompts, or incident text.
  4. Give responders the app views they need through SSO and RBAC, and retain audit logs for app, agent, and workflow activity.

Major orchestrates PagerDuty; it never replaces it. PagerDuty remains the incident system of record and paging authority. The Major app stores linked evidence, drafts, approvals, and execution history, not a competing source of incident status.

PagerDuty API guide for governed incident work

Authentication and permission boundaries

Account API tokens are account-wide and either read-only or full-access. User API tokens inherit the associated user's access and are available on accounts with Advanced Permissions. API-token requests use the Authorization header with the Token token= credential scheme.

OAuth functionality distinguishes Classic User OAuth from Scoped OAuth. Classic scopes are read or write; Scoped OAuth grants resource-specific access. The endpoint pages below specify incidents.read, incidents.write, and services.read. The webhook test requires webhook_subscriptions.write, an administrative capability that should stay outside a triage agent's routine permissions.

For incident writes, supply the valid account user's From header as required by the individual endpoint. Keep attribution truthful; a shared credential must not make an automated draft look like an engineer's independent action.

Pagination, limits, events, and testing

Classic pagination uses limit and offset, with more indicating another page. The default page size is 25, the maximum is 100, and offset plus limit cannot exceed 10,000. Split large history reads into bounded windows. Cursor-enabled endpoints instead return a next cursor for the subsequent cursor parameter; do not assume every collection uses the same scheme.

REST rate limits allow 960 requests per minute per account API key. User keys share 960 per user; app OAuth limits depend on whether the token acts as the app or user. Watch ratelimit-limit, ratelimit-remaining, and ratelimit-reset, and respect endpoint-specific limits rather than treating 960 as a universal allowance.

V3 webhooks include incident.triggered, incident.acknowledged, and incident.resolved events, filtered by service, team, or account. Verify the X-PagerDuty-Signature header using HMAC-SHA256 over the unaltered raw body and a constant-time comparison, following the signature guide.

The delivery rules expect a 2xx response within five seconds for Generic Webhooks. Temporary failures can be retried for up to 48 hours; most 4xx errors other than 429 are permanent. Persist verified events before acknowledging them, then process asynchronously.

PagerDuty documents trial onboarding, not a universal API test-mode switch in the pages reviewed here. Use controlled recipients and an isolated test service or account, then test the webhook ping and actual synthetic incident lifecycle separately.

GET /incidents

Lists incidents and requires incidents.read with Scoped OAuth. Use bounded reconciliation rather than repeated full-account scans.

  • App: retain incident IDs and reconciliation checkpoints.
  • Agent: compare recent incidents for potentially relevant recurrence.
  • Workflow: recover records missed during a webhook outage.

GET /incidents/{id}

Retrieves incident details by ID or incident number and requires incidents.read. This is the freshness check before acting.

  • App: display the latest status alongside the draft's observation time.
  • Agent: distinguish an active problem from a resolved incident.
  • Workflow: re-read the incident after approval before executing a change.

GET /services

Lists services and requires services.read. Maintain explicit cross-system mappings instead of matching names blindly.

  • App: store PagerDuty service IDs against Datadog monitor mappings.
  • Agent: flag ambiguous ownership rather than inventing a destination.
  • Workflow: refresh the service catalog and quarantine unmapped inputs.

POST /incidents

Creates an incident synchronously without a corresponding monitoring event and requires incidents.write. The reference flags operation-specific rate limits.

  • App: record the approved request and returned incident ID.
  • Agent: draft a security escalation with its evidence and severity rationale.
  • Workflow: create it only after approval and a duplicate check.

PUT /incidents/{id}

Can acknowledge, resolve, escalate, or reassign an incident and requires incidents.write. Treat these as consequential actions.

  • App: show the proposed change next to current incident state.
  • Agent: recommend reassignment when evidence points to another owner.
  • Workflow: enforce approval, revalidate state, and record the response.

POST /incidents/{id}/notes

Adds a note and requires incidents.write plus the documented From header. A brief belongs here only after its review policy passes.

  • App: retain draft versions and the returned note ID.
  • Agent: write a concise brief separating observations from hypotheses.
  • Workflow: post the approved version once and log the result.

POST /webhook_subscriptions/{id}/ping

Tests a subscription by delivering the pagey.ping event and requires webhook_subscriptions.write. Keep this setup permission with administrators.

  • App: record delivery-test status separately from incident processing.
  • Agent: summarize test failures for an administrator without changing configuration.
  • Workflow: run a controlled commissioning check before production activation.

PagerDuty automation use cases on Major

Datadog and PagerDuty triage briefs

Keep existing monitoring-to-paging behavior in place and add a source-linked brief for the responder.

  • App: hold incident-to-monitor mappings, evidence timestamps, and reviewed briefs.
  • Agent: judge which nearby alerts matter and distinguish observations from possible explanations.
  • Workflow: react to a new incident, gather context, route review, and attach the approved brief.

Use the incident detail, service list, and incident note operations above. Missing Datadog evidence should produce an explicit gap, not a confident guess.

Okta anomalies to approved security escalation

A security lead needs a reviewable record of why an unusual sign-in should interrupt an on-call responder. This is a proposed pattern, not a published customer deployment.

  • App: retain the event reference, affected identity, review outcome, and escalation status.
  • Agent: classify the anomaly and judge its significance using available identity context.
  • Workflow: collect new events, request approval, check for an existing incident, and create the escalation.

Use incident listing and incident creation. Account suspension is outside this workflow's authority; an incident recommendation does not authorize identity changes.

Jira, PagerDuty, and Confluence incident follow-up

Connect response to engineering follow-up without pretending a draft postmortem establishes root cause.

  • App: map the incident to its Jira issue, documentation draft, and review owner.
  • Agent: synthesize the timeline, identify unanswered questions, and draft corrective work.
  • Workflow: start follow-up after resolution, collect linked context, and route the draft to the incident owner.

Use incident detail for the current state and incident notes for an approved follow-up pointer. Broader patterns belong in the IT automation guide; tool-specific context is in the Datadog hub and Jira hub.

Build the incident architecture on Major

PagerDuty and Datadog context flows through a Major workflow, triage agent, approval policy, and stateful app

Figure 1. Major stores incident evidence and review state while the agent judges relevance.

Start with a small app schema: incidents, evidence snapshots, draft versions, approvals, and action attempts. Keep external IDs on every relationship so a responder can follow the chain back to the source system.

The workflow validates events, performs bounded retrieval, and resumes failed steps from stored checkpoints. The agent reads a defined evidence package, decides which facts matter, and proposes a brief or action. The app never decides severity or probable cause; it renders the agent's assessment alongside sources and uncertainty.

Build repeatable service mapping, field validation, and action-ledger checks into app code. The agent should not reason through those mechanics anew for every incident. This is Major's practical distinction: agents build the software they run on, and the resulting app becomes reusable operating infrastructure.

Preserve the evidence timestamp with each draft. A reviewer must be able to see that the incident was active when a recommendation was written but resolved before approval. Durable state makes that difference visible across runs and shift changes.

A worked example with explicit assumptions

For example, suppose an SRE team reviews 100 incidents in a month. Each takes eight minutes to collect monitor context, inspect related records, and write a handoff brief. That is 100 times eight, or 800 minutes of preparation.

In an illustrative Major rollout, assume 80 incidents arrive with enough evidence for a draft that takes two minutes to review. The remaining 20 have missing mappings or conflicting evidence and still take eight minutes. The arithmetic is 80 times two plus 20 times eight, or 320 minutes, a difference of 480 minutes, equivalent to eight hours.

Those are planning assumptions, not a customer result or a promise about resolution time. They exclude connector setup, ongoing maintenance, and the responder's actual investigation.

The end-to-end path is simple: receive the event, store it, fetch the incident and monitor, ask the agent for a brief, review it, and post the approved note. Measure preparation time, missing-evidence frequency, reviewer corrections, and duplicate writes before deciding whether the rollout helps. Faster draft generation alone is not a reliability outcome.

Governance and failure handling before external writes

A Major agent proposal goes directly to private draft storage or through human approval and fresh-state checks before PagerDuty changes

Figure 2. Private analysis proceeds automatically; PagerDuty writes wait for approval and revalidation.

Allow read-only collection and private draft storage automatically within the configured access boundary. In this starting design, incident creation, acknowledgement, resolution, reassignment, escalation, and note posting require an authorized person's approval. The agent must never interpret permission to draft as permission to page, silence, or remediate production.

Handle predictable failures in software:

  • Duplicate delivery: use PagerDuty's X-Webhook-Id for inbound deduplication, as documented in its delivery behavior, plus a separate action ledger for outbound writes.
  • Rate limiting: on 429, use the reset interval from the rate-limit headers; when absent, apply capped backoff with jitter.
  • Uncertain writes: after a timeout, reconcile the intended action before retrying. An inbound deduplication key does not make external writes idempotent.
  • Stale approval: re-fetch current incident state and invalidate a proposal that no longer applies.
  • Missing context: save a partial result with named gaps and alert the workflow owner.

Log evidence references, draft version, agent assessment, approver, credential identity, request outcome, and retry history. Set a retention policy for sensitive identity and diagnostic data rather than copying unrestricted payloads into every review screen.

PagerDuty AI Apps, Agents and Workflow Automations

Every published Major use case that works with PagerDuty, newest first. Each one names the PagerDuty API objects involved and shows the app, agent and workflow split.

  • Jira Automation for Engineering-Ops AI Agents: Turn Jira from a ticket tracker into the command center for an engineering-ops agent. Learn how agents triage issues, link incidents to postmortems, and draft sprint analytics by reasoning across Jira, GitHub, Slack, Confluence, and PagerDuty.

Frequently asked questions

What are PagerDuty AI agents?

PagerDuty offers Paige, its SRE Agent, for incident investigation and troubleshooting. Agents on Major can complement it by judging cross-system evidence and drafting actions, while deterministic apps hold state and workflows enforce approvals.

Does Major have a native PagerDuty connector?

No. PagerDuty connects through Major's Pipedream connector. IT configures the connection once, keeps credentials in the connector vault, and limits which actions each app, agent, and workflow can perform.

Which PagerDuty API permissions should an agent receive?

Start with read-only access for context gathering. Scoped OAuth uses incidents.read and services.read for the reads covered here, and incidents.write for incident creation, updates, and notes. Account API tokens instead offer read-only or full access; they are not resource-scoped OAuth tokens.

Can an agent automatically resolve PagerDuty incidents?

The incident update API supports resolution, but technical permission is not sufficient authorization. In the design described here, resolution, reassignment, escalation, and incident creation require a human approval followed by a fresh incident-state check.

How do you prevent duplicate incident notes?

Deduplicate incoming webhook deliveries using PagerDuty's X-Webhook-Id header, then track each intended note in an action ledger. If a write times out, reconcile the result before retrying instead of assuming the first attempt failed.

Can I test PagerDuty automation without paging production?

Use an isolated test service or trial account with controlled notification recipients. PagerDuty documents trial onboarding and a webhook subscription ping endpoint. A successful ping verifies delivery, not incident permissions, lifecycle handling, or production readiness.

Sources

Start building on Major

Build an incident-context app that retains evidence, lets an agent judge what matters, and keeps consequential PagerDuty actions behind review.

About the author

Ritvij Gautam is Head of Growth at Major.

Last reviewed November 20, 2026 by Ritvij Gautam.

Frequently asked questions

What are PagerDuty AI agents?
PagerDuty offers Paige, its SRE Agent, for incident investigation and troubleshooting. Agents on Major can complement it by judging cross-system evidence and drafting actions, while deterministic apps hold state and workflows enforce approvals.
Does Major have a native PagerDuty connector?
No. PagerDuty connects through Major's Pipedream connector. IT configures the connection once, keeps credentials in the connector vault, and limits which actions each app, agent, and workflow can perform.
Which PagerDuty API permissions should an agent receive?
Start with read-only access for context gathering. Scoped OAuth uses incidents.read and services.read for the reads covered here, and incidents.write for incident creation, updates, and notes. Account API tokens instead offer read-only or full access; they are not resource-scoped OAuth tokens.
Can an agent automatically resolve PagerDuty incidents?
The incident update API supports resolution, but technical permission is not sufficient authorization. In the design described here, resolution, reassignment, escalation, and incident creation require a human approval followed by a fresh incident-state check.
How do you prevent duplicate incident notes?
Deduplicate incoming webhook deliveries using PagerDuty's X-Webhook-Id header, then track each intended note in an action ledger. If a write times out, reconcile the result before retrying instead of assuming the first attempt failed.
Can I test PagerDuty automation without paging production?
Use an isolated test service or trial account with controlled notification recipients. PagerDuty documents trial onboarding and a webhook subscription ping endpoint. A successful ping verifies delivery, not incident permissions, lifecycle handling, or production readiness.