Major

Greenhouse automation: AI, APIs and governed hiring workflows

Explore Greenhouse automation with native AI, eight verified Harvest v3 endpoints, and Major apps, agents, and workflows for controlled onboarding and recruiting operations.

Ritvij GautamUpdated
recruiter interviewing candidate office (Greenhouse AI agents and automations)

Greenhouse automation connects recruiting records to the work that follows: onboarding handoffs, hiring-stage reviews, and recruiter briefs. This guide covers Greenhouse's native AI, eight Harvest v3 endpoints, and how Major adds governed apps, agents, and workflows around your ATS.

Major's deterministic app layer keeps mappings, approvals, and execution history durable. Agents handle ambiguous interpretation, while workflows control when approved actions run.

What Greenhouse is and who relies on it

Greenhouse is a recruiting system organized around candidates, applications, jobs, interviews, and offers. Talent acquisition teams use these records to run structured hiring, while recruiting operations maintains the process connecting them.

The distinction between person and application matters. A candidate profile can have several applications, each representing a journey on a particular job. An onboarding handoff should follow the correct application, not simply the most recently edited person.

Recruiters need an accurate view of the current stage and submitted feedback. Hiring managers review evidence against the interview plan. HR operations needs the approved offer, department, and start date before preparing an employee record.

For these teams, Major is the enterprise platform where agents build the software they run on. A handoff app stores the relationship between hiring records and downstream worker records. An agent interprets a department mismatch, and the workflow waits for HR approval before executing the change. That separation makes recruiting operations inspectable without moving the hiring system of record out of Greenhouse.

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

Start with the Greenhouse AI feature inventory. Its native capabilities already cover several recruiting tasks:

  • Talent Matching compares resumes with defined, weighted job criteria.
  • Talent Matching assistant turns a role description into draft calibration and filters using the configured job post and scorecard attributes.
  • Greenhouse Notetaker records and transcribes interviews, then prepares notes mapped to scorecard questions.
  • Scorecard summaries collect agreements, disagreements, and key quotes for the hiring team.
  • Job post description suggestions and Email generator help draft recruiting content.

These features are useful inside the recruiting process. Check your subscription, enabled settings, and permissions rather than assuming every feature is available everywhere.

Major adds cross-system execution and durable operational state. For example, an agent can interpret why a Greenhouse department differs from the approved Workday organization, then propose a mapping with its evidence. The app preserves the proposal and reviewer response; the workflow controls the eventual handoff.

A native AI feature is not proof of a matching public API. The AI opt-out endpoint, for example, explicitly does not expose Talent Matching scores. Do not design an integration around unavailable scoring data.

How Greenhouse connects to Major

Greenhouse connects through Major's Pipedream connector, app p_greenhouse. There is no native Greenhouse connector. This is the connection method listed in the catalog, not a claim that every Harvest v3 operation is already packaged as a connector action.

IT sets up the connector once. Credentials stay centrally stored in the connector vault, scoped to the approved work, never copied into code or prompts. SSO, role-based access control, and audit logs cover the app, agent, and workflow using that access.

Before implementation, inspect the configured connector's authentication and operation coverage. A legacy action should not be assumed to use OAuth or support the v3 paths below merely because the vendor now documents them. Validate any coverage gap before committing to the design.

For verified v3 access, the API host is https://harvest.greenhouse.io and requests use Bearer tokens under the current authentication guide. The following reference describes vendor capabilities, not guaranteed connector action coverage.

Major orchestrates Greenhouse; it never replaces it. Keep hiring records authoritative in Greenhouse and store only the operational state needed to coordinate approved work elsewhere.

Greenhouse API guide for apps, agents, and workflows

Authentication and least privilege

Harvest v3 authentication uses OAuth 2.0. Custom integrations use Client Credentials; partner integrations use Authorization Code. The token service is https://auth.greenhouse.io/token. Older v1/v2 Basic authentication is transitional, not the model to assume for new v3 work.

Exact documented read scopes include harvest:candidates:list, harvest:offers:list, and harvest:aioptout_statuses:list, verified on their endpoint pages. Scope is not the whole permission model: the authentication guide says list operations require Site Admin authorization. Offer reads have additional private-data permission conditions when acting as a named user, and can return an empty list rather than a permission error, as the offer reference explains.

Pagination, limits, events, and testing

Pagination is cursor-based. Follow the next URL in the Link response header. The cursor parameter must not be combined with other query parameters. Initial list requests accept per_page, default 100 and maximum 500 on the endpoints below.

Rate limits use fixed 30-second windows, with token requests on a separate 60-second interval. A universal numeric allowance is not documented publicly on that guide: 75 appears as a header example, not a promised quota. Monitor X-RateLimit-Limit, X-RateLimit-Remaining, and X-RateLimit-Reset; honor Retry-After on HTTP 429.

Recruiting webhooks cover application, candidate, interview, and job events. Verify the Signature header with an HMAC-SHA256 digest over the unchanged request body and the configured secret. The docs specify up to seven delivery attempts over roughly 15 hours after failures, and identify deliveries through Greenhouse-Event-ID.

Greenhouse also documents a sandbox for API and integration testing. Use separate sandbox credentials and job-board identifiers, not production keys. Confirm availability with your account team.

Candidate identity: GET /v3/candidates

Returns candidate profiles, with related applications and attachments on separate resources in v3.

  • App: Store the candidate ID alongside the approved handoff record.
  • Agent: Flag inconsistent contact or identity context for a reviewer, without guessing legal identity.
  • Workflow: Retrieve only the profiles needed for the current application batch.

Hiring status: GET /v3/applications

Returns candidate journeys by job, including lifecycle status and current-stage references.

  • App: Track each application separately when a person applies to multiple roles.
  • Agent: Explain an exception where the onboarding request references a different job.
  • Workflow: Recheck hired status immediately before an approved downstream handoff.

Requisition context: GET /v3/jobs

Returns hiring plans with department, office, and custom-field context.

  • App: Maintain a versioned mapping from job IDs to target organizational codes.
  • Agent: Propose how an unfamiliar job label relates to an approved department taxonomy.
  • Workflow: Refresh job context before assembling a reviewer packet.

Offer details: GET /v3/offers

Returns versioned offers with status, proposed start date, and configured custom fields.

  • App: Record the exact offer version a reviewer approved.
  • Agent: Identify ambiguous compensation-field labels without inventing missing values.
  • Workflow: Request current_only=true when preparing the latest offer and invalidate approval if its version changes.

Stage history: GET /v3/application_stages

Returns per-application stage visits, with entry and exit timestamps.

  • App: Calculate stage duration and preserve the reporting cohort in code.
  • Agent: Summarize unusual aging patterns using the calculated evidence.
  • Workflow: Refresh stage histories on the reporting schedule and route incomplete histories for review.

Interview plan: GET /v3/jobinterviewstages

Returns the job-specific stage definitions, including ordering and active status.

  • App: Resolve historical stage IDs without discarding retired definitions.
  • Agent: Suggest comparable stage groupings across differently named hiring plans.
  • Workflow: Require analyst approval before applying a proposed grouping to compliance reporting.

Feedback evidence: GET /v3/scorecards

Returns interview evaluations, distinguishing complete feedback from draft working copies.

  • App: Display submitted feedback with source IDs and access restrictions.
  • Agent: Draft a briefing that distinguishes interviewer agreement from unresolved disagreement.
  • Workflow: Include only complete scorecards in a scheduled briefing. A draft does not establish that feedback is overdue.

AI review preferences: GET /v3/aioptout_statuses

Returns confirmed Talent Matching opt-out decisions per candidate and job, subject to product access and permissions.

  • App: Preserve the job-specific decision and its confirmation time.
  • Agent: Explain why a record requires manual review under the approved policy.
  • Workflow: Route opted-out records to people. Absence of a row is not affirmative opt-in, and this endpoint is not general consent for other AI processing.

Greenhouse automation use cases on Major

Greenhouse and Workday: a verified new-hire handoff

  • App: Hold source IDs, the approved field mapping, review status, and downstream receipt.
  • Agent: Interpret role or department mismatches against the approved Workday schema.
  • Workflow: Read applications, offers, candidates, and jobs, then require HR approval.

Keep offer acceptance separate from the marked-hired event. Workday worker retrieval is not worker creation; use the tenant-approved hire process or existing integration, then capture its receipt.

Greenhouse and BambooHR: onboarding exceptions

  • App: Track hire-to-employee mappings and outstanding required fields.
  • Agent: Propose a department match when source and target labels differ.
  • Workflow: Combine current offers, candidate profiles, and hired applications, then release only the approved payload.

Evaluate the existing handoff first. Add controls around missing fields and corrections rather than creating a second competing export.

Greenhouse and Google Sheets: controlled stage-review packets

  • App: Calculate selection rates in code using approved cohorts and separately authorized demographic inputs.
  • Agent: Draft caveats and questions from the calculated results.
  • Workflow: Join stage histories, stage definitions, and applications, then share only an approved aggregate packet.

Those three resources do not themselves provide demographic survey responses. Restrict sensitive data separately, suppress small groups, and leave legal interpretation to qualified reviewers.

Greenhouse and Slack: hiring-manager briefs

  • App: Keep a restricted briefing queue with source references.
  • Agent: Summarize complete scorecards against job context, preserving disagreement.
  • Workflow: Send a minimal notification linking to the authorized app view, not private interview notes in a broad channel.

The hiring manager remains responsible for the hiring decision. A convenient summary must not become an unreviewed recommendation to reject someone.

Build the Greenhouse operating architecture on Major

Greenhouse and Workday hiring records flow through a Major workflow, mapping agent, HR approval, and stateful handoff app

Greenhouse remains authoritative while Major coordinates the governed handoff.

Build one control surface for the process owner, with tables for source snapshots, mapping versions, approvals, execution attempts, and target receipts. The diagram's workflow box represents orchestration, not another judgment step.

  1. Receive a verified event or run a scheduled reconciliation.
  2. Fetch the current records through scoped connectors.
  3. Run required-field, version, and duplicate checks in deterministic code.
  4. Ask the agent to interpret exceptions needing judgment, with evidence attached.
  5. Route the proposal to an authorized HR reviewer.
  6. Execute the approved handoff and record its downstream result.

The agent can build reusable validation and mapping software, then run that software for routine cases. It should not reason through a known department-code lookup on every hire. Store that approved lookup in the app and reserve reasoning for genuinely unfamiliar cases.

The app never decides whether a candidate deserves a job. It holds state, enforces validations, and exposes the review queue. This is how Major makes the process stateful without treating a conversation as an operational database.

A worked example with numbers

For example, suppose an HR operations team processes 100 onboarding handoffs each month. Manually assembling and checking each packet takes 15 minutes, so preparation consumes 100 x 15 = 1,500 minutes, or 25 hours.

In an illustrative Major design, 80 packets have exact approved mappings and need three minutes of human review each. The remaining 20 require eight minutes each because an agent's proposed mapping needs inspection. Total review effort becomes 80 x 3 + 20 x 8 = 400 minutes, or six hours and 40 minutes.

The arithmetic difference is 1,100 minutes, or 18 hours and 20 minutes. These are planning assumptions, not a customer result, and exclude setup, maintenance, and service latency.

Follow one exception end to end: the application is hired, the current offer has a start date, but the target department code is unclear. The agent proposes a mapping, HR approves that offer version, and the workflow checks the source again before submitting. If the offer changes, the app returns the packet to review instead of reusing stale approval.

Governance, approval, and failure handling

Greenhouse to BambooHR handoffs split between exact mapping checks and agent-reviewed exceptions before HR approval

Exact mappings skip reasoning, while every employee handoff still requires HR approval.

Define the prohibited actions first. The agent must not reject candidates, change compensation, invent missing identity fields, or certify compliance on its own. Demographic data should stay out of onboarding mapping and recruiter briefings.

Use a delivery ledger plus a business-level duplicate key. The webhook guide identifies deliveries with Greenhouse-Event-ID; that alone is not a guarantee that separate deliveries cannot describe the same business change. Persist the application, source version, action, and target receipt before allowing another write.

For transient failures, use bounded retries and jitter. Honor the vendor's Retry-After guidance, and pause authorization failures for an owner rather than retrying indefinitely. After an ambiguous downstream timeout, reconcile before creating another employee.

Approval must bind to a particular proposal and source version. Record who approved, what changed, what the agent suggested, and what actually executed. Keep sensitive payloads out of general logs, set retention periods, and give reviewers enough evidence to correct a mapping without exposing unrelated candidates.

Greenhouse AI Apps, Agents and Workflow Automations

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

Frequently asked questions

What can you automate in Greenhouse?

Start with hiring-data reconciliation, approved onboarding handoffs, stage-aging reviews, and recruiter briefing preparation. Major holds workflow state in an app, uses an agent for ambiguous interpretation, and routes consequential actions through human approval.

Does Greenhouse have built-in AI?

Yes. Greenhouse documents Talent Matching, Greenhouse Notetaker, scorecard summaries, job post description suggestions, and an email generator. Check your subscription and configuration for availability; native features do not imply that equivalent public API operations exist.

How does Greenhouse connect to Major?

Greenhouse connects through Major's Pipedream connector, app p_greenhouse. There is no native Greenhouse connector. IT configures scoped access once, with credentials stored centrally rather than in code or prompts.

Which authentication does the Greenhouse Harvest API use?

Harvest v3 uses OAuth 2.0 Bearer tokens. Customer-built custom integrations use Client Credentials, while partner integrations use Authorization Code. Verify the actual connector's authentication and endpoint coverage before planning a v3 migration.

Is an accepted offer the same as a hired application?

No. An accepted offer is an offer state, while hired is an application lifecycle state. Use the business event your HR team has approved, re-read the current records, and do not create an employee merely because an older offer version was accepted.

Can AI automatically reject candidates or certify hiring fairness?

The workflows described here do neither. People approve hiring outcomes, and deterministic calculations support compliance review. Selection-rate ratios and AI summaries are review inputs, not legal conclusions or proof that a hiring process is fair.

Sources

These official references support the resource distinctions, authentication model, and operational controls above. Endpoint documentation describes vendor capability; implementation still requires checking the configured connector, account permissions, and current product access.

Start building on Major

Build a Greenhouse handoff app with versioned mappings, a review queue, and an agent for the exceptions that need judgment. Keep HR in control of consequential writes.

About the author

Ritvij Gautam is Head of Growth at Major.

Last reviewed November 19, 2026 by Ritvij Gautam.

Frequently asked questions

What can you automate in Greenhouse?
Start with hiring-data reconciliation, approved onboarding handoffs, stage-aging reviews, and recruiter briefing preparation. Major holds workflow state in an app, uses an agent for ambiguous interpretation, and routes consequential actions through human approval.
Does Greenhouse have built-in AI?
Yes. Greenhouse documents Talent Matching, Greenhouse Notetaker, scorecard summaries, job post description suggestions, and an email generator. Check your subscription and configuration for availability; native features do not imply that equivalent public API operations exist.
How does Greenhouse connect to Major?
Greenhouse connects through Major's Pipedream connector, app p_greenhouse. There is no native Greenhouse connector. IT configures scoped access once, with credentials stored centrally rather than in code or prompts.
Which authentication does the Greenhouse Harvest API use?
Harvest v3 uses OAuth 2.0 Bearer tokens. Customer-built custom integrations use Client Credentials, while partner integrations use Authorization Code. Verify the actual connector's authentication and endpoint coverage before planning a v3 migration.
Is an accepted offer the same as a hired application?
No. An accepted offer is an offer state, while hired is an application lifecycle state. Use the business event your HR team has approved, re-read the current records, and do not create an employee merely because an older offer version was accepted.
Can AI automatically reject candidates or certify hiring fairness?
The workflows described here do neither. People approve hiring outcomes, and deterministic calculations support compliance review. Selection-rate ratios and AI summaries are review inputs, not legal conclusions or proof that a hiring process is fair.