Adobe Marketo Engage AI Agents: APIs and Automation
A researched guide to Marketo's native AI, eight verified API operations, custom connectivity to Major, and governed engagement workflows across marketing and sales.

Adobe Marketo Engage AI agents can evaluate engagement, explain lead priority, and coordinate approved actions across marketing and sales. This guide covers Marketo's native AI, the APIs behind useful integrations, and how to build governed apps, agents, and workflows on Major.
Major keeps repeatable work in deterministic apps with durable state. Agents handle judgment, while workflows control when recommendations become actions.
What Adobe Marketo Engage does for marketing operations
Marketo Engage manages the operational side of B2B marketing: people records, engagement history, nurture programs, campaign execution, and measurement. Adobe's product glossary describes the marketing concepts behind these jobs, including programs, Smart Campaigns, and predictive content.
Marketing operations teams maintain the database, define qualification logic, and keep campaign inputs consistent. Demand generation teams use engagement signals to choose audiences and evaluate whether a program produces useful sales conversations. Revenue operations teams connect those signals to the records sales teams work from.
The practical problem is often coordination rather than another score. A person can engage heavily in Marketo while their CRM deal is closed, their account owner has changed, or their advertising eligibility needs review. A useful recommendation needs that surrounding context.
Major is the enterprise platform where agents build the software they run on. For this problem, that software can be a deterministic engagement review app: identity mappings, evidence, approved changes, and processing history live in managed database tables rather than disappearing with a conversation.
Adobe Marketo Engage AI and its built-in capabilities
Adobe already provides AI inside Marketo. Its AI Assistant content-generation documentation covers generating subject lines and preheaders, revising text or images, and creating full emails from templates. Access requires the relevant terms and an administrator granting the Access AI Assistant permission.
Predictive Audiences estimates event registration and attendance likelihood, supports predictive filters, and provides model insights. Adobe documents subscription and enablement requirements. Predictive Content, described as an add-on, recommends relevant material to web visitors and email recipients.
These features are useful for content production and targeting inside the marketing environment. An agent on Major adds a different layer: interpreting Marketo evidence alongside CRM ownership, advertising constraints, or a team's review history.
For example, the agent can explain why recent engagement merits sales attention but not advertising inclusion. The app preserves the evidence and decision record; the workflow requests approval before a destination changes. Nothing here assumes Adobe's AI predictions are exposed through the endpoints below.
How to connect Marketo to Major and build the review app
Adobe Marketo Engage has no native or Pipedream connector in Major's catalog. It connects as a custom API connector, with the base URL and authentication configured once by IT. Major orchestrates Adobe Marketo Engage; it never replaces it.
Adobe documents the REST base URL as an instance-specific address such as https://<Munchkin-ID>.mktorest.com/rest. Copy the actual Endpoint value from Admin, Integration, Web Services rather than constructing a production address from an example. Endpoint paths below include /rest, so avoid duplicating that prefix when configuring requests.
Build the initial implementation around these pieces:
- Connector: IT configures the Marketo instance and OAuth client credentials, with secrets stored centrally in Major's connector vault, never in code or prompts.
- App: Managed tables store identity mappings, activity cursors, recommendations, approvals, and write results. The UI exposes the review queue.
- Agent: Evaluate ambiguous engagement and draft a recommendation with evidence and uncertainty.
- Workflow: Poll on a schedule, fetch context, invoke judgment, and route consequential proposals for human approval.
- Governance: Apply SSO, RBAC, scoped connector access, and audit logs across the app, agent, and workflow.
Start with reads. Add write capabilities only after operators can inspect the evidence and reconcile proposed changes.
Marketo API guide: authentication, limits, and useful endpoints
Authentication and permission boundaries
Marketo uses OAuth client credentials associated with a custom LaunchPoint service and an API-only user. Tokens last one hour; send them using the Authorization bearer header. These are service credentials, not a personal token pasted into an agent conversation.
The user's role permissions determine allowed operations. Relevant exact labels include Read-Only Person, Read-Only Activity, and Read-Write Person. These are Marketo role permissions, not invented OAuth scope strings. Separate observation from write access and test the actual role against required operations.
Pagination, rate limits, webhooks, and testing
Activity pagination starts with a timestamp and continues through a returned cursor. Preserve the nextPageToken value exactly and URL-encode it. For activities, continue while moreResult is true, even after a short page. List membership uses nextPageToken and batchSize, with a maximum of 300 records.
Adobe's REST overview documents 100 calls per 20 seconds and 10 concurrent calls per instance. It states a 50,000-call daily subscription allocation; verify your actual allocation in Admin. The error guide identifies 606 for rate limiting and 607 for exhausted daily quota. Dedicated rate-limit response headers are not documented publicly in these references. Inspect the JSON success flag, errors, and per-record results, not HTTP status alone.
Marketo webhooks send outbound GET or POST requests from trigger campaigns. Custom headers and response mapping are supported, and requests time out after 30 seconds. This reference does not document a universal signing scheme or automatic retry guarantee. Configure a receiver-validated secret header, acknowledge quickly, and reconcile missed events through polling.
Adobe offers an optional sandbox instance. It starts empty and limits emails to 20 per campaign run. Test with synthetic people and nonproduction destinations.
GET /rest/v1/leads/describe.json
Discover available person fields and their API names before mapping records.
- App: Store validated field mappings for the review interface.
- Agent: Interpret evidence using field descriptions instead of guessing names.
- Workflow: Validate required fields before a scheduled run and stop on incompatible schema changes.
GET /rest/v1/leads.json
Find people using filterType and filterValues rather than repeatedly exporting the database.
- App: Associate Marketo IDs with existing CRM contact mappings.
- Agent: Examine candidate matches when identity is ambiguous.
- Workflow: Fetch bounded batches and route unresolved matches to review instead of accepting the first result.
GET /rest/v1/lead/{id}.json
Retrieve a specific person's current record by Marketo ID.
- App: Display current context beside the proposed action.
- Agent: Reassess an engagement recommendation against the latest person data.
- Workflow: Refresh the person immediately before an approved write so stale context cannot silently drive execution.
POST /rest/v1/leads.json
Create or update people, with documented batches of up to 300 records.
- App: Store an explicit before-and-after field diff.
- Agent: Recommend a justified classification, without directly changing protected fields.
- Workflow: Execute only approved, allowlisted updates and record each result, including skipped records and reasons.
GET /rest/v1/activities/types.json
Retrieve the activity types available for interpreting engagement events.
- App: Maintain a readable activity dictionary for reviewers.
- Agent: Distinguish meaningful engagement from less informative event types.
- Workflow: Refresh definitions deliberately and validate configured activity filters rather than hard-coding unexplained numeric IDs.
GET /rest/v1/activities/pagingtoken.json
Initialize activity retrieval using an ISO 8601 sinceDatetime value.
- App: Preserve the initial lookback boundary and checkpoint history.
- Agent: See which observation period supports its recommendation.
- Workflow: Bootstrap a new stream or an intentional replay without silently substituting the current time and losing earlier events.
GET /rest/v1/activities.json
Read activities using activityTypeIds and nextPageToken, with up to 10 activity types per request.
- App: Store evidence IDs and observed timestamps.
- Agent: Evaluate recency, intensity, and contradictory engagement patterns.
- Workflow: Page until moreResult is false, deduplicate repeated evidence, and advance the durable checkpoint only after successful storage.
GET /rest/v1/lists/{listId}/leads.json
Read people belonging to a selected static list.
- App: Show membership alongside the campaign's intended audience.
- Agent: Explain why a person belongs in a proposed review cohort.
- Workflow: Fetch all pages of a controlled test list before comparing recommendations with an operator's expected results.
Marketo automation use cases on Major
Marketo and Google Ads: review engagement-based audiences
The use-case context includes connecting Marketo engagement to paid-media audiences. Treat this as an eligibility-controlled audience process, not a direct instruction to increase spend whenever a score rises.
- App: Store cohort versions, person mappings, exclusion reasons, and audience sync receipts.
- Agent: Judge which recent engagement patterns support a high-intent cohort and explain borderline cases.
- Workflow: Schedule activity reads, check approved identity and consent rules, request approval, and hand eligible records to Google Ads. Keep budget changes outside this workflow.
- Marketo endpoints: Activity reads, person lookup, and static-list reads provide the evidence.
See the Google Ads automation hub for the destination system's broader role.
Marketo and HubSpot: make engagement useful to deal owners
Another planned use case joins recent marketing activity to HubSpot deals. A concise explanation of why a person is newly engaged is more useful than copying an unexplained score into every deal.
- App: Track the Marketo person, linked HubSpot deal, owner, evidence window, and last delivered recommendation.
- Agent: Assess engagement recency and intensity, then draft a contextual note with uncertainty.
- Workflow: Resolve the deal mapping, route ambiguous identities to review, and update the confirmed deal only under approved policy.
- Marketo endpoints: Activity types and activities, plus current person retrieval, support that decision.
The HubSpot AI agents hub covers adjacent sales-record processes.
Marketo and Slack: resolve marketing-data exceptions
An additional build pattern is a queue for missing qualification fields or contradictory CRM mappings. The notification should point to the governed review record, not scatter personal data into chat.
- App: Hold exception records, proposed corrections, reviewer ownership, and the complete field diff.
- Agent: Classify the exception and propose a correction with supporting evidence.
- Workflow: Notify the owner in Slack, wait for approval in the review app, then apply an allowlisted change or close the exception without writing.
- Marketo endpoints: Field description, person retrieval, and approved person updates keep the mapping and execution explicit.
The Slack automation hub covers the notification side, not a Marketo-specific tutorial.
Architecture: durable evidence before agent judgment

Marketo engagement context becomes a governed cross-system recommendation.
The architecture separates data collection from interpretation. A scheduled workflow reads Marketo activities and linked HubSpot context through scoped connectors. Deterministic code normalizes identifiers, checks required fields, deduplicates events, and stores evidence before the agent receives the case.
The agent makes the judgment call: whether the pattern merits attention, what the owner should know, and how uncertain the recommendation is. The app holds that recommendation in managed tables alongside evidence IDs, timestamps, policy version, and review status.
A decision step routes consequential changes to a person. After approval, the workflow rechecks current context and executes only the authorized change. The output receipt returns to the app so the next run knows what already happened.
This is also where Major's software-building model matters. Once a repeatable validation or matching rule is understood, the agent can build it into the app instead of reasoning through it afresh on every record. Judgment remains with the agent; repeatable execution becomes software the team can inspect.
For example: calculate the value of a review queue
Consider an illustrative marketing operations team reviewing 200 engagement exceptions each week. Without a shared queue, suppose each takes six minutes: finding the person, checking recent activity, locating the CRM deal, and writing a note.
That is 200 × 6 = 1,200 minutes, or 20 hours per week. These are planning assumptions, not customer results or measured Major performance.
In the proposed design, the workflow collects evidence and the agent drafts a recommendation. Suppose 160 routine cases need a one-minute human check, while 40 ambiguous cases still take six minutes. Review time becomes 160 × 1 + 40 × 6 = 400 minutes. Add 80 minutes for monitoring and exception reconciliation: 480 minutes, or eight hours.
The illustrative difference is 12 hours weekly. Validate it with a read-only pilot, measuring collection time, reviewer corrections, false matches, and failed deliveries. If matching quality is poor, keep the review requirement and improve the deterministic identity mapping before expanding write permissions.
Governance, approvals, and failure handling

Separate read-only observations from approved Google Ads audience changes.
Define an action boundary before enabling writes. Internal observations can be saved automatically; changes to audience membership, protected person fields, campaign execution, or spending require explicit authorization. An agent must never infer consent, overwrite suppression rules, or merge ambiguous identities on its own.
Store approvals against the exact proposed change. If the record, consent evidence, or destination mapping changes while approval is pending, return the case to review rather than treating yesterday's approval as unlimited authority.
Handle failure as application state:
- Expired credentials: Refresh through the configured authentication path, then stop and notify IT if authorization remains invalid.
- Rate pressure: Throttle centrally, apply bounded backoff with jitter for transient failures, and pause quota-exhausted work until capacity is available.
- Partial writes: Parse each record result and retry only appropriate failures. Marketo errors can appear inside HTTP 200 responses.
- Duplicate events: Use stable evidence identifiers and destination action keys. Reconcile uncertain writes before resending.
- Audit: Retain the evidence, agent rationale, reviewer identity, approved diff, and execution receipt under the team's retention policy.
Adobe Marketo Engage AI Apps, Agents and Workflow Automations
Step-by-step Adobe Marketo Engage builds will appear here as they are published.
Frequently asked questions
What are Adobe Marketo Engage AI agents?
They are agents that reason over Marketo engagement and business context to recommend or carry out permitted work. On Major, the agent makes judgment calls, an app holds durable state and review queues, and a workflow handles triggers, retries, routing, and approvals.
Does Marketo Engage have built-in AI?
Yes. Adobe documents AI Assistant for email content generation and Predictive Audiences for event registration and attendance likelihood. Predictive Content recommends material to website visitors and email recipients. Access depends on the feature and subscription.
How does Adobe Marketo Engage connect to Major?
It connects through a custom API connector, not a catalog-native or Pipedream connector. IT configures the instance base URL and OAuth client credentials once, stores credentials centrally, and grants only the required API permissions.
Can a Marketo agent update people or launch campaigns automatically?
Updating people is possible with the appropriate API permission and endpoint, but permission is not approval. Begin with read-only recommendations, require approval for consequential changes, and keep campaign launches outside the agent's independent authority.
What are Marketo REST API limits?
Adobe documents 100 calls per 20 seconds and 10 concurrent calls per instance, plus a daily subscription allocation. Its overview states 50,000 daily calls; confirm your actual quota in Admin. Inspect response bodies because an HTTP 200 can contain API errors.
Is there a Marketo sandbox for testing AI workflows?
Adobe offers a separate sandbox instance as an optional purchase. It starts empty and limits emails to 20 per campaign run. Use synthetic records, separate credentials, and nonproduction destinations before allowing writes to production.
Sources
- Marketo Engage glossary: Product concepts, marketing terminology, and the definition of Predictive Content used in this guide.
- Generate email content: Native AI Assistant capabilities and the administrator-controlled access requirements.
- Predictive Audiences: Event predictions, targeting features, and enablement conditions.
- REST API overview: API limits, authentication guidance, and response conventions.
- Authentication: LaunchPoint services, API-only users, and OAuth token lifetime.
- Base URL: Finding the correct instance endpoint in Marketo Admin.
- Role permission descriptions: Exact permission labels for scoped API access.
- Leads: Field discovery, filtered retrieval, individual records, and person updates.
- Activities: Activity types, retrieval filters, and cursor handling.
- Paging tokens: Initial timestamp and continuation-token behavior.
- List membership: Static-list reads and membership pagination.
- Webhooks: Outbound trigger-campaign calls, custom headers, response mapping, and timeout.
- Error codes: Response-level failures, record-level failures, and retry considerations.
- Marketo sandbox: Separate test instances and email restrictions.
Start building on Major
Build a Marketo engagement review app with durable evidence, scoped connectors, and human approval before consequential changes. Start with observation, then expand only the actions your team can inspect and govern.
About the author
Ritvij Gautam is Head of Growth at Major.
Last reviewed November 20, 2026 by Ritvij Gautam.
Frequently asked questions
- What are Adobe Marketo Engage AI agents?
- They are agents that reason over Marketo engagement and business context to recommend or carry out permitted work. On Major, the agent makes judgment calls, an app holds durable state and review queues, and a workflow handles triggers, retries, routing, and approvals.
- Does Marketo Engage have built-in AI?
- Yes. Adobe documents AI Assistant for email content generation and Predictive Audiences for event registration and attendance likelihood. Predictive Content recommends material to website visitors and email recipients. Access depends on the feature and subscription.
- How does Adobe Marketo Engage connect to Major?
- It connects through a custom API connector, not a catalog-native or Pipedream connector. IT configures the instance base URL and OAuth client credentials once, stores credentials centrally, and grants only the required API permissions.
- Can a Marketo agent update people or launch campaigns automatically?
- Updating people is possible with the appropriate API permission and endpoint, but permission is not approval. Begin with read-only recommendations, require approval for consequential changes, and keep campaign launches outside the agent's independent authority.
- What are Marketo REST API limits?
- Adobe documents 100 calls per 20 seconds and 10 concurrent calls per instance, plus a daily subscription allocation. Its overview states 50,000 daily calls; confirm your actual quota in Admin. Inspect response bodies because an HTTP 200 can contain API errors.
- Is there a Marketo sandbox for testing AI workflows?
- Adobe offers a separate sandbox instance as an optional purchase. It starts empty and limits emails to 20 per campaign run. Use synthetic records, separate credentials, and nonproduction destinations before allowing writes to production.