An AI agent for Adobe Target and Launch: offers, activities, libraries and what waits for a person
Adobe Target decides what your visitors see; Adobe Launch decides which code runs on your site. Both are worth an assistant — reading activity results, preparing offers, checking which rules changed — and both are places where one wrong change reaches every visitor. This post covers what an AI agent reads and prepares in each, what always waits for a person, what we measured on a real Adobe organization, and goals to start with.
Two products, two kinds of risk
Each Adobe product is its own connection in Agent Planners: you connect Adobe Target and Adobe Launch on their own cards, each with only that product's permissions, using an OAuth Server-to-Server credential from your own Adobe organization. The product profiles the credential belongs to in Adobe Admin Console decide what it may read and change — a Target workspace and its properties, the Launch properties and their publishing rights.
The changes they allow are different in kind. A Target offer is content; an activity turned live changes what visitors see. A Launch rule is configuration; a built and published library is code running on your site. The agent's rules follow that line.
What the agent reads
| Product | Reads |
|---|---|
| Adobe Target | Activities and their state (live, paused, saved), one activity's experiences, offers, audiences, and each activity's results by experience — visitors, conversions and lift; locations (mboxes), environments and properties through target_get |
| Adobe Launch | Companies, properties, rules, data elements, extensions and libraries with their publishing state; anything else Adobe documents through launch_get |
What it prepares in Adobe Target
- JSON offers: created and edited, for example a personalization payload for an app or a single-page site.
- Audiences: created through the Target write gateway.
- Activities: created as saved — not live — so a person can look at them in Target first; a saved activity can be deactivated.
What it prepares in Adobe Launch
- Development libraries: created with the resources you name. When a rule or data element changed since its last revision, the agent revises it first and adds the new revision — and says so in the result.
- Rules: switched on or off in the draft.
- Data elements: created and renamed through the Launch write gateway.
How approvals work here
The ordinary changes above follow your approval setting: by default each waits for your approval card, and auto-approve is opt-in per task, schedule or project. The changes named as "always a person" wait under auto-approve too, on a schedule, and from an MCP client or the REST API. Sandboxes, access control and credentials are refused outright — those stay in Adobe Admin Console.
What we measured on a real Adobe organization
- Target: an HTML offer created, changed, read back and deleted; an audience created and deleted; an A/B activity created as saved, never turned live, deactivated and deleted.
- Launch: in a throwaway property created for the test (no site used it): a data element created, renamed and deleted, an Adobe-managed host, a development environment, a library with the Core extension, and a build that succeeded. The property was deleted afterwards.
- Not run live: turning a Target activity live and publishing a library to a production site.
Goals to start with
| Goal | What the agent does |
|---|---|
| "Which Target activities are live, and which experience is winning in each?" | Lists live activities and reads each one's results by experience: visitors, conversions and lift. |
| "Prepare a JSON offer for the spring banner with this text and image URL." | Creates the offer for your approval; it does nothing until an activity uses it. |
| "List the Launch rules on our website property that changed since the last published library." | Compares the rules' revisions with the published library and lists them. |
| "Put the new checkout rule in a development library." | Creates the library, revising the rule first if needed; the build waits for a person. |
| "Which data elements does the purchase rule use?" | Reads the rule and the data elements it references. |
Setting it up
- 1In Adobe Developer Console add Adobe Target and the Experience Platform Launch API (Tags) to a project with an OAuth Server-to-Server credential, and put it on the product profiles it should use.
- 2On Integrations press Connect on the Adobe Target card and on the Adobe Launch card; the same project JSON works on both.
- 3Press Test connection. Target asks for its tenant (the name after @ in your Target address — we suggest and confirm it); Launch asks for the company.
Frequently asked questions
- Can an AI agent make an Adobe Target activity live?
- Only with a person's approval, every time — auto-approve never applies it. The agent creates activities as saved and prepares offers and audiences.
- Can an AI agent publish Adobe Launch?
- It can prepare a development library and switch rules in the draft. Building or publishing a library, custom code and extensions always wait for a person.
- Can the agent create HTML offers in Target?
- It can prepare one, but an HTML offer is code on your site, so it always waits for a person.