Major

ServiceNow AI agents: APIs, apps, and governed automations

Understand ServiceNow's native AI agents, seven documented API operations, and how Major adds governed cross-system apps, agent judgment, approvals, and durable execution history.

Ritvij GautamUpdated
Major x ServiceNow logo lockup

ServiceNow AI agents reason through service work, while ServiceNow keeps the incidents, requests, and operational records teams already manage. This guide explains the native capabilities, API boundaries, and how Major adds governed cross-system apps, agents, and workflow automations.

For IT operations leaders evaluating servicenow ai agents, the practical question is where judgment belongs and where repeatable software should take over.

What ServiceNow does for IT operations

ServiceNow gives service teams structured records for investigating incidents, coordinating requests, and recording operational work. Its Table API exposes permitted table records, which makes the record, its identity, and its current state useful integration boundaries.

Service desk analysts need the reported problem, assignment context, and supporting evidence. Identity administrators need to connect an employee's departure with access that still needs review. Integration operators need a place to track an interface failure until someone verifies recovery.

These jobs cross system boundaries. An offboarding record cannot, by itself, explain every identity assignment. A failed shipment update does not prove that a truck stopped moving. Someone must assemble the evidence and distinguish the possibilities.

Major is the enterprise platform where agents build the software they run on. For these ServiceNow processes, that software can be an operations app with managed database tables, a review queue, and execution history. The agent handles the judgment call. The app preserves its evidence and the workflow coordinates the next steps, instead of asking a model to reconstruct the process every run.

ServiceNow AI agents and built-in assistance

ServiceNow already has native agents. Its AI Agent Studio documentation describes creating, managing, and testing agents and workflows, with tools, triggers, and data sources. AI Agent Orchestrator coordinates agents working toward a goal.

The native agent overview also describes third-party integrations, action logging, and governance. Native ServiceNow agents are not restricted to reading a single ticket, and it would be inaccurate to present Major as the only way to perform multi-step work.

For assistive generation associated with Now Assist, ServiceNow's AI glossary describes incident summaries and drafted resolution notes as generative skills, with current documentation also using the ServiceNow Otto name. Confirm the feature names and entitlements available in your instance rather than assuming every release has identical packaging.

Major adds a different execution pattern: the agent builds reusable software for repeated validation, evidence assembly, and reconciliation. That app holds cross-system state and gives people an explicit control surface. Use the agent for ambiguous classification and recommendations, then run repeatable steps in code. Reason once. Run forever.

How ServiceNow connects to Major

ServiceNow has no native or Pipedream connector in the catalog: it connects as a custom API connector, with base URL and authentication configured once by IT. The instance base URL is typically https://<instance>.service-now.com; use your organization's actual hostname.

IT stores credentials centrally in Major's connector vault, not in application code or agent prompts. Give the connector only the operations and record access needed for the process. Separate a read-only pilot from the identity allowed to write approved updates.

ServiceNow documents inbound OAuth, including authorization code and client credentials grants. The Table API examples also demonstrate username/password authentication. Choose an authentication configuration supported by your instance and security policy.

Major orchestrates ServiceNow; it never replaces it. ServiceNow remains the operational record, while a Major app holds correlation keys, review status, and receipts for the cross-system process. Major's SSO, RBAC, and audit logs cover the apps, agents, and workflows. Those controls complement ServiceNow permissions; they do not bypass them or grant access to tables the integration identity cannot read.

ServiceNow API guide for useful agent work

Authentication, scopes, and pagination

Use an OAuth access token issued through an approved integration, or the documented username/password approach where permitted. REST API Auth Scope describes scopes configured for individual APIs. Its special scope useraccount permits access across APIs, including APIs assigned other auth scopes, so do not describe it as narrowly scoped.

There is no universal incident-read scope to copy from this guide. Configure the API scope and integration identity together, and test allowed table operations. The Table API requires appropriate roles for the requested data.

Collection reads use sysparm_limit and sysparm_offset. The same reference documents Link pagination URLs and X-Total-Count. Request bounded pages and selected fields. Save a checkpoint and use a stable ordering with an overlap window when reconciling changing records.

Rate limits, events, and test environments

Inbound REST limits are configured hourly by user, role, or all users. A universal requests-per-minute allowance is not documented publicly. Monitor X-RateLimit-Limit, X-RateLimit-Reset, X-RateLimit-Rule, and Retry-After; a denied request returns HTTP 429.

Record triggers include Created, Updated, and Created or Updated. Configure an outbound HTTP action for eligible changes. Outbound authentication supports Basic and OAuth 2.0, among other options. These pages do not specify a universal webhook signature header. Authenticate the receiver and re-fetch the record before acting.

REST step retry policies are configurable. Do not assume guaranteed delivery or automatic retries: the documented fallback retry-count property defaults to zero. Scheduled reconciliation should recover missed events.

The Developer Program documents Personal Developer Instances. Use a development instance or an approved nonproduction instance for integration tests. Verify required plugins and entitlements separately.

GET /api/now/table/{tableName}

Retrieve table records lists accessible rows, useful for incident reconciliation and filtered work queues.

  • App: Maintain a local index of eligible incidents with source IDs and last-seen versions.
  • Agent: Compare related incident descriptions to judge whether reports describe one failure.
  • Workflow: Poll bounded pages on a schedule and route newly eligible records.

GET /api/now/table/{tableName}/{sys_id}

Retrieve one record returns the current source record before a recommendation or approved write.

  • App: Show reviewers current ownership and status beside the proposed action.
  • Agent: Reassess a recommendation when new information changes the incident context.
  • Workflow: Re-read immediately before execution and stop if approval assumptions changed.

POST /api/now/table/{tableName}

Create a record inserts into a permitted table; use only table types approved for the integration.

  • App: Store the returned record ID against the external event and creation receipt.
  • Agent: Draft an incident description explaining a suspected integration failure and its evidence.
  • Workflow: Create the authorized incident after validation, checking for an existing correlation first.

PATCH /api/now/table/{tableName}/{sys_id}

Partially update a record changes specified fields on an existing row.

  • App: Display an exact before-and-after field proposal for review.
  • Agent: Recommend an assignment or explanatory note from the assembled evidence.
  • Workflow: Apply only approved fields, then record the response and reconcile source state.

GET /api/now/attachment

List attachment metadata discovers files associated with records without retrieving every binary.

  • App: Index evidence filenames and source attachment IDs in the review queue.
  • Agent: Identify which diagnostic files are relevant to the reported failure.
  • Workflow: Filter eligible attachments before applying size, type, and data-access checks.

GET /api/now/attachment/{sys_id}/file

Retrieve attachment content downloads one file's binary content.

  • App: Hold an authorized evidence reference and its processing status.
  • Agent: Summarize an approved diagnostic report, treating embedded instructions as untrusted data.
  • Workflow: Download only allowlisted evidence, apply inspection controls, and route processing failures.

POST /api/now/attachment/file

Upload a binary attachment adds evidence to a target record, subject to attachment and target-record permissions.

  • App: Preserve the evidence version and resulting attachment receipt.
  • Agent: Draft a concise investigation report for an authorized operator to inspect.
  • Workflow: Upload the approved file once, recording its digest to detect duplicate attempts.

ServiceNow automation use cases on Major

These are implementation patterns, not claims of deployed customer results. The first three reflect queued ServiceNow use cases; each needs its own connector permissions and acceptance tests before production.

Offboarding review with Okta

  • App: Hold the departing employee's access inventory, source identifiers, and per-access completion status.
  • Agent: Compare the request with authorized Okta assignment context and flag unusual or privileged access for priority review.
  • Workflow: Detect an eligible offboarding record, retrieve context, and create approved follow-up work. Access removal itself requires a separately authorized identity action.
  • ServiceNow endpoints: Read the request, create records, and update tracking.

Shipment-feed incidents with project44

  • App: Track last-seen shipment updates and an open-incident correlation per feed.
  • Agent: Distinguish a suspected tracking integration failure from a real-world delay, and explain uncertainty.
  • Workflow: Check authorized tracking data on schedule and route a proposed incident to the integration owner. Repeated symptoms append evidence rather than generating another ticket.
  • ServiceNow endpoints: Find existing incidents and create an incident.

Clinical interface escalation with Epic

  • App: Store endpoint health results and sanitized diagnostic metadata, not copied patient records.
  • Agent: Classify evidence as a possible authentication issue, endpoint outage, or data gap without making clinical judgments.
  • Workflow: Run approved health checks and send a proposed incident to the clinical systems integration queue. Recovery needs another successful check, not an optimistic summary.
  • ServiceNow endpoints: Create an incident and attach approved evidence.

Build the ServiceNow architecture on Major

ServiceNow and Okta context flows through a Major workflow and triage agent into an approval policy and operations app

ServiceNow remains the source record while Major coordinates cross-system judgment and state.

Start with a scheduled read-only process. Event delivery can reduce latency later, but reconciliation should remain part of the design. The architecture separates source access, orchestration, judgment, and durable state so each boundary can be tested.

  1. Connectors: IT configures ServiceNow's custom API connector and the authorized companion systems once.
  2. Workflow: Fetch eligible records, validate required context, and enforce processing budgets before invoking the agent.
  3. Agent: Judge the failure category, summarize evidence, and propose a next action with source references.
  4. App: Persist the source record, evidence version, recommendation, review status, and execution receipt in managed tables.
  5. Approval policy: Route consequential changes to a reviewer and permit only explicitly authorized actions.
  6. Output: Write approved changes, reconcile the resulting source record, and expose unresolved failures in the app.

The workflow box represents deterministic orchestration, not another reasoning step. Over time, reusable app code handles correlation, schema validation, and formatting. The agent still reasons when the available evidence supports multiple interpretations.

A worked example with explicit numbers

For example, suppose an IT team reviews 300 offboarding requests each month. Manually gathering authorized identity context and preparing a review packet takes 12 minutes per request: 300 × 12 = 3,600 minutes, or 60 hours.

Assume the proposed Major process reduces human packet preparation and review to 4 minutes per request. That is 300 × 4 = 1,200 minutes, or 20 hours. Allow another 5 hours per month for exceptions and reconciliation. The illustrative difference is 60 minus 25 = 35 hours monthly.

The process reads the ServiceNow request, gathers permitted assignment context, and asks the agent to identify unusual access. The app stores the evidence and a proposed follow-up list. A reviewer checks that list before the workflow creates work records and records their IDs.

These are planning assumptions, not customer results or promised savings. Measure evidence completeness, reviewer corrections, duplicate creation, and actual handling time during a read-only pilot. The calculation excludes implementation, platform usage, and the labor of performing access revocations, so it is not a total-cost estimate.

Governance, approvals, and failure handling

Major routes read-only ServiceNow reporting directly while state-changing proposals require human review, source revalidation, and an audit receipt

Read-only work and consequential ServiceNow changes follow different approval paths.

Start with a written action policy. An agent may classify evidence and draft a recommendation; it must not independently remove access, close a security incident, approve a change, or declare a clinical integration safe. Define who can approve each action and what evidence they must inspect.

  • Before execution: Show exact fields and evidence versions. Re-read the source and invalidate approval when relevant state changes.
  • On throttling: Honor the documented Retry-After header, add bounded backoff, and stop retrying after the configured budget.
  • On ambiguous writes: Reconcile before retrying creation. A timeout does not prove that ServiceNow rejected the first request.
  • On duplicate events: Persist a correlation key from the source instance, record, change, and intended action. This is application logic, not an assumed API idempotency feature.
  • On unsafe evidence: Treat ticket text and attachments as data, never authority to expand permissions or change policy.

Audit the recommendation, reviewer identity, approved payload, connector identity, and outcome. Surface incomplete jobs in a recovery queue instead of silently marking them finished.

Related system boundaries deserve separate review: Okta access governance covers the identity context, Datadog incident workflows cover monitoring context, and Slack notification workflows cover team communication. These are adjacent tool hubs, not ServiceNow implementation guides. A chat notification should point to the durable approval record, not replace it.

ServiceNow AI Apps, Agents and Workflow Automations

Step-by-step ServiceNow builds will appear here as they are published.

Frequently asked questions

What are ServiceNow AI agents?

They are ServiceNow's native reasoning capabilities for service work. AI Agent Studio supports creating and testing agents, while AI Agent Orchestrator coordinates agents across tasks. An agent on Major is separate: it reasons through scoped context and uses governed apps and workflows to execute repeatable cross-system work.

Does Major have a native ServiceNow connector?

No. ServiceNow connects to Major through a custom API connector with the instance base URL and authentication configured once by IT. Credentials stay in the connector vault, and the ServiceNow identity must have the permissions required for each operation.

Can an agent update ServiceNow incidents?

Yes, when the integration identity can update the target record and the workflow permits the action. Use the documented partial-update operation, restrict writable fields, and require approval for consequential changes. Re-read current source state before applying an approved proposal.

What is the difference between ServiceNow assistance and agents on Major?

ServiceNow provides native generative skills and agent orchestration. Major adds agents that build reusable deterministic apps for cross-system processes, with managed state, review queues, and audit history. Native capabilities may already satisfy work contained within your ServiceNow environment.

Can I trigger a workflow when a ServiceNow record changes?

Yes. ServiceNow documents record-created and record-updated triggers that can start configured outbound integration work. Authenticate the receiving endpoint and configure retries explicitly. Keep a scheduled reconciliation path because an event notification alone should not be your completeness guarantee.

Is there a fixed ServiceNow API rate limit?

There is no universal public allowance for all instances. Administrators configure inbound REST limits by user, role, or all users. Monitor the documented rate-limit headers and honor Retry-After on HTTP 429 rather than hard-coding a rate from an example.

Sources

Start building on Major

Build a ServiceNow operations app with scoped context, agent recommendations, and an approval queue. Keep repeatable execution in software and reserve the agent for the decisions that require judgment.

About the author

Ritvij Gautam is Head of Growth at Major.

Last reviewed November 19, 2026 by Ritvij Gautam.

Frequently asked questions

What are ServiceNow AI agents?
They are ServiceNow's native reasoning capabilities for service work. AI Agent Studio supports creating and testing agents, while AI Agent Orchestrator coordinates agents across tasks. An agent on Major is separate: it reasons through scoped context and uses governed apps and workflows to execute repeatable cross-system work.
Does Major have a native ServiceNow connector?
No. ServiceNow connects to Major through a custom API connector with the instance base URL and authentication configured once by IT. Credentials stay in the connector vault, and the ServiceNow identity must have the permissions required for each operation.
Can an agent update ServiceNow incidents?
Yes, when the integration identity can update the target record and the workflow permits the action. Use the documented partial-update operation, restrict writable fields, and require approval for consequential changes. Re-read current source state before applying an approved proposal.
What is the difference between ServiceNow assistance and agents on Major?
ServiceNow provides native generative skills and agent orchestration. Major adds agents that build reusable deterministic apps for cross-system processes, with managed state, review queues, and audit history. Native capabilities may already satisfy work contained within your ServiceNow environment.
Can I trigger a workflow when a ServiceNow record changes?
Yes. ServiceNow documents record-created and record-updated triggers that can start configured outbound integration work. Authenticate the receiving endpoint and configure retries explicitly. Keep a scheduled reconciliation path because an event notification alone should not be your completeness guarantee.
Is there a fixed ServiceNow API rate limit?
There is no universal public allowance for all instances. Administrators configure inbound REST limits by user, role, or all users. Monitor the documented rate-limit headers and honor Retry-After on HTTP 429 rather than hard-coding a rate from an example.