AI agents for Adobe Experience Platform: schemas, datasets and audiences with a person in control
An AI agent for Adobe Experience Platform reads sandboxes, XDM schemas, datasets and audiences through the Experience Platform APIs, answers questions about your data model and audiences, and proposes audiences and schema changes for approval while moving data always waits for a person.
What is an AI agent for Experience Platform?
It is software connected to one Experience Platform sandbox with your own credential. It reads the data model (schemas, field groups, classes), the datasets and their batches, and the audiences with their PQL — and explains them in plain language.
It can draft changes — a new audience, a schema change — but moving data in or out of Experience Platform is a person's decision every time.
What it is good at
- Data-model audits — which schemas use the Profile class, which field groups are unused, what changed.
- Dataset health — which datasets stopped receiving batches.
- Audience reviews — streaming vs batch audiences, their expressions and merge policies, and a proposed new audience with its PQL shown first.
Where it is easy to get wrong
- Treating 401 or 403 as empty — a product profile without Data Management or Profile Management answers that way; the agent names the permission.
- Activating an audience casually — activation is a dataflow to a destination; it waits for a person.
- Working in the wrong sandbox — every call carries the connection's sandbox unless a call names another one.
Guardrails
- Agent Planners' defaults: by default each change waits for your approval (in the app or by email), auto-approve is opt-in per task, schedule or project, and the person class below waits for a person even with auto-approve on.
- Person class for Experience Platform: ingestion, dataflows and destinations, merge policies, Profile system jobs, Query Service statements and any delete. Sandboxes, access control and privacy jobs are refused.
Frequently asked questions
- Can the agent create audiences in Experience Platform?
- Yes, with your approval — and it shows the PQL expression first. Activating it to a destination waits for a person.
- Does it read profile data?
- It reads the data model, datasets and audience definitions; Profile system jobs and privacy jobs are not something it runs.
- Which permissions does it need?
- Sandboxes, Data Modeling, and — for datasets and audiences — Data Management and Profile Management in the credential's product profile.