Adobe AI agents: catalogue, shopping and marketing 2026
Adobe’s AI agents serve different catalogue, customer and marketing tasks, which a retailer must distinguish before connecting store records.
- Published
- Last updated
Adobe AI agents handle different jobs across customer conversations, catalogue management and internal marketing. A retailer needs to choose the job first: helping a customer compare products requires different records and permissions from changing a campaign audience.
The practical distinction is between an agent that answers from your store records and an agent authorised to change them. Adobe’s product family contains both, with access controlled by the application and connection involved.
Identify the Adobe agent and its job
Adobe Brand Concierge is the customer-facing conversational product, while Adobe Experience Platform agents help employees work inside business applications. Adobe’s business AI overview spans customer engagement and content work, alongside other enterprise tasks.
The names below identify different parts of that family. A connection or orchestration layer supplies capabilities to an agent, so buying access to one application does not establish access to every row.
| Product or component | Retail job | Records involved |
|---|---|---|
| Adobe Brand Concierge | Help customers ask questions and explore products on a brand’s digital properties | Approved website content and connected product information |
| Adobe Commerce Catalog Discovery and Product Comparison skills | Search a live catalogue and compare named products within Brand Concierge | Commerce product details and searchable attributes |
| Adobe Commerce Catalog Agent | Propose and apply product-name and description enrichment through Adobe Brand Visibility | Existing catalogue names, descriptions and product attributes |
| Audience Agent | Help employees manage audiences in Real-Time CDP or Adobe Journey Optimizer | Customer profiles and audience definitions |
| Content Advisor Agent | Help employees find content and create visual variants in Adobe Experience Manager Assets | Approved assets and their metadata |
| Journey Agent | Help employees analyse and improve journeys in Adobe Journey Optimizer | Journey configuration and customer engagement data |
| Adobe Experience Platform Agent Orchestrator | Coordinate specialised agents through the AI Assistant interface | The data and actions available to the employee’s authorised applications |
Adobe’s agent application map identifies each employee agent’s supported applications. These agents respect the underlying application’s access controls. Their presence in the same family does not give an audience-management agent authority over a customer’s basket.
Follow catalogue facts into an agent answer
A catalogue answer depends on the selected skill calling a connection that returns the correct product record. Keep the product identifier attached to the result throughout that path, particularly when several variants share a name.
For Brand Concierge, Adobe documents Commerce MCP as an integration backed by Adobe Live Search. Model Context Protocol (MCP) lets an agent call defined tools. The available Brand Concierge tools cover product search and details, with variant and facet discovery. Adobe calls its two Commerce skills catalogue-only capabilities in the skills and integrations framework.
A retailer can trace the answer through four responsibilities:
- The Commerce catalogue supplies the product identity and attributes. Use the variant’s identifier to distinguish, for example, a navy medium coat from its black equivalent.
- The connected Commerce integration supplies the product information available to the skill. Price and availability must belong to that same product and store context.
- The selected skill uses the returned information to answer the customer’s question or build a comparison.
- The retailer checks that the answer still matches the source record after a catalogue change.
Adobe’s Brand Concierge setup explanation describes live product, price, stock and variant information through a Commerce connection. A concierge created from a website URL begins with crawled knowledge instead. Connecting a website and connecting a live catalogue are separate setup decisions.
Catalogue enrichment has a different direction of travel. The Commerce Catalog Agent reads existing attributes and proposes names or descriptions that a practitioner can edit before publishing. Adobe explicitly excludes price and inventory from that enrichment. Its catalogue enrichment documentation describes applying changes directly to Commerce and rolling them back.
Keep these permissions separate. Retrieving a product for a comparison does not establish permission to rewrite it. A published description change also reaches other channels that use the Commerce catalogue, so merchandising must own its approval.
Separate a shopping conversation from marketing work
A customer choosing a product needs current sale information, while an employee planning a campaign needs approved content and customer-engagement records. The same product name can appear in both tasks without making their answers interchangeable.
Consider a hypothetical outdoor retailer. A customer asks for a waterproof jacket in their size. Catalogue discovery needs the actual variants and their attributes, plus current price and stock. A question about returning that jacket needs the retailer’s approved policy content as well.
An employee preparing a wet-weather campaign has another task. Content Advisor Agent can help find relevant assets in Experience Manager Assets. Audience Agent works with audience records, and Journey Agent works with journeys in its supported application. A campaign image of the jacket is creative material, so staff must separately establish whether its advertised variant remains available.
Two common interpretations produce different outcomes:
- Incorrect: a successful product comparison proves that the agent can take payment. A comparison establishes that product information was retrieved and presented.
- Correct: checkout needs its own authorised transaction capability and an order result. The documented catalogue-only Brand Concierge skills establish discovery and comparison scope.
Apply the same distinction to employee work. An agent finding an approved image does not prove that a campaign has been published. The team needs the relevant application’s saved result or publication record.
Check prerequisites before connecting the store
Match the product licence and account access to the selected job before connecting customer or catalogue records. Adobe’s application map distinguishes agents enabled in existing applications from AI-first applications with their own licences.
For employee agents, the documented routes include an Adobe Experience Platform Agents AI Credits licence or an eligible trial arrangement. Jobs consume AI credits, and users retain their underlying application permissions. For Brand Concierge, the AI-first application licence includes its agentic features without a separate Agent Orchestrator licence.
An administrator must add Brand Concierge users to the Experience Platform organisation and assign a role with Manage Brand Concierge. That role also selects accessible sandboxes. Adobe’s permission setup guide identifies the required administrator access.
For the live catalogue route, attach the Commerce MCP integration and supply the correct Commerce environment configuration and credentials. The baseline website concierge does not include that integration. Building a connector absent from the available integrations requires engineering work.
Catalogue enrichment also needs its own prerequisites. Adobe documents it as a beta for Adobe Commerce customers, with access through the account manager. Adobe Brand Visibility must be connected to Commerce before changes can be applied.
Assign a merchandising owner to catalogue facts, a content owner to policies and an administrator to connection health. Give the customer-experience team responsibility for the resulting answers. Before exposing the selected workflow to customers, use these acceptance questions:
| Change or failure | Acceptance question | Evidence to retain |
|---|---|---|
| Updated product | Does the answer use the changed attribute, price or availability for the same variant? | Source product identifier and the resulting answer |
| Withdrawn policy | After removing the old policy from its connected source, does the answer stop presenting it as current? | Retired source reference and a replacement answer |
| Failed connection | Does a catalogue request avoid presenting an old price or stock claim as live? | Connection error and the customer-visible response |
| Applied enrichment | Does Commerce contain the approved name or description, with unrelated fields unchanged? | Saved catalogue record and rollback reference |
Use Adobe’s previews and evaluations to assess those cases before deployment. Keep discovery acceptance separate from transaction acceptance: a product answer passes only when it matches its source, and an order passes only when its authorised transaction produces the required record.
Give your store its own shopping agent
Customers ask in their own words. Your agent finds, compares and checks out from your catalogue, right on your site.