SitecoreAI now has a capability at most stages of the journey from a marketing intention to a live, measured experience. Brief types capture intent in a consistent structure. Brand context grounds agents in approved brand information. Agentic Studio runs and sequences agent work, including inside Page builder. Audience builder, component personalisation and A/B/n tests decide who sees what. Marketer MCP and the Agent API let the same actions happen from AI clients outside the SitecoreAI interface.
Each connection between those capabilities creates an organisational question: who owns intent, how context is governed, where approval lives, what an agent may change, and how learning returns to planning. The platform gives each question a place to be answered.
Start from the lifecycle and place the features on it
Sitecore publishes its progress as a changelog, one feature at a time, and most commentary follows the same shape. Placed against the work a marketing team does, the features form a chain from intent to measurement, and the weak points are the hand-offs between them.
The lifecycle we use runs from intent to a structured brief, then to context and research, content and action, authoring, audience, personalisation or experiment, publication and measurement. It maps onto the model we use for every Digital Signal: context, signal, interpretation, decision, orchestration, execution and measurement, with measurement producing the next signal. When Sitecore announced SitecoreAI in November 2025, agentic orchestration was mostly a direction of travel. Ten months of releases later, there is enough shipped capability to place a named feature under each stage.
Context
What the organisation knows about its brand, audiences and constraints.
Brand context files and brand kits, applied in Agentic Studio and exposed through the Agent API and Marketer MCP.
Signal
Evidence that something has changed or deserves attention.
Agentic Studio skills such as competitive analysis; Signals, powered by Scrunch, for AI visibility (generally available, Scrunch subscription).
Interpretation
Judging what the evidence means for this organisation.
People. Agents can draft research in Agentic Studio, and the judgement about what it means stays with the team.
Decision
Committing to an objective, an audience and a timeline.
Brief types in the Strategy app, with briefs linked to campaigns and scheduled in the campaign calendar.
Orchestration
Sequencing the work between people, agents and systems.
Agent sequencing and workflow agents in Agentic Studio; Marketer MCP and the Agent API for AI clients outside SitecoreAI.
Execution
Changing the live experience.
Page builder with Page builder compatible agents and a review card before saving; Audience builder, component personalisation and A/B/n tests.
Measurement
Finding out whether the change did what the brief intended.
Personalisation and A/B/n test analytics in the authoring flow, and the Performance area.
Measurement produces the next signal. In SitecoreAI, carrying that result back into the next brief is the connection a team has to design most deliberately.
What each connection does
Intent becomes a brief
Since 18 June 2026 the Strategy app has supported brief types: reusable templates that set the fields a marketer completes, so that objectives, audiences and timelines are captured the same way every time. Admins create and manage brief types, and marketers choose one when they create a brief and complete it manually or with AI. Each field can carry help text and an “intent” that guides AI-generated content.
Briefs sit alongside campaigns. A Briefs tab arrived on the campaign overview page in March 2026, and since 12 August 2026 briefs can be linked, created and unlinked from that tab. The field-level intent setting is the most consequential detail. It is where a person writes down, once, what an agent should aim for when it fills that field in for every future brief of that type.
Context grounds the agents
Brand context, released on 31 August 2026, generates editable Markdown files from a brand website or an existing SitecoreAI site. An organisation can keep several contexts for different brands, regions or business units. All organisation users can view the files, and admins and owners can edit them. Brand context is applied in Agentic Studio and is also available through the Agent API and Marketer MCP.
Treat brand context as infrastructure. The same files can ground an agent in Agentic Studio and an external AI client connected through MCP, so an error or an outdated claim in one file reaches every route that reads it.
Agents do bounded work in Agentic Studio
Since the 30 March 2026 release, agents in Agentic Studio can run on SitecoreAI pages, briefs, artefacts and CSV files, and each run can be given its own context, skills, MCP servers and tools. Agents can run in sequence within a space, for example research, then content generation, then translation, reusing earlier artefacts as shared context. Workflow agents can call standard agents, external APIs and Agent API tools.
In May 2026 Sitecore added custom MCP connectors with OAuth, where each user authenticates separately, and parallel per-item execution for workflows that fan out across many pages. Several items in that release are marked preview or early access.
Agentic Studio has been available inside Page builder since 20 May 2026. Only agents that Sitecore describes as “Page builder compatible” run there, in the context of the site collection, site or page open in the canvas. Since 3 July 2026, the chat shows which tools the AI used and presents proposed changes on a review card, where a person can compare values, edit, skip or decline before anything is saved.
Audience, personalisation and experiment decide who sees what
Audience builder creates reusable, rule-based audiences from profile details, session information, events and traits, and estimates audience size before an audience goes live. Scheduled refreshes followed in July 2026. Component personalisation, which is rolling out in phases, uses a decision table in which rules are evaluated top to bottom and the first match chooses the component variant. A/B/n tests can be targeted at an audience, so only matching visitors take part.
An audience defined once can now shape a test, a personalised component and, through the tools below, an agent’s work. That makes the audience definition one of the most reused decisions in the whole chain.
MCP and the Agent API extend execution beyond the interface
Sitecore describes Marketer MCP as a server that translates natural-language prompts into actions in SitecoreAI through the Agent API. Its tools cover sites, pages, content, components, assets, personalisation, brand kits, brand context, briefs, experiments and flow definitions. The May 2026 experiment tools and personalisation and brief tools mean an AI client can create a component A/B/n test, update a personalisation variant or generate a revised brief draft. Since 6 July 2026, page operations can target a specific personalisation or test variant.
Sitecore’s own explanation calls the Agent API “the engine underneath” MCP and states that its actions follow Sitecore security, permissions and approval workflows. Setting up Marketer MCP requires an admin role. When Sitecore talks about connecting to other MCP-enabled tools, the coordination happens in the AI client. Marketer MCP works alongside other MCP servers in the same assistant, such as a project management tool, and the assistant decides how to combine them.
Five organisational questions the connections create
Each connection hands something from one party to another: from a marketer to a template, from an admin to every agent, from an agent to a page, from a live experience back to a plan. Each hand-off leaves a question.
| Connection | Organisational question | What the platform leaves open | A sensible first decision |
|---|---|---|---|
| Intent to brief | Who owns intent? | Admins define brief types, marketers complete briefs, and AI can fill fields. Nothing names the person accountable for what a brief is trying to achieve. | Name an owner for each brief type and agree which fields a person must complete before AI assistance is used. |
| Brief to context | How is context governed? | Brand context is generated from a website and editable by admins and owners. How often it is reviewed, and who approves a change, is a team process. | Treat brand context files as controlled documents with an owner, a review date and a record of changes. |
| Context to agent work | What may an agent change? | Agents can run on pages, briefs and files, use per-user MCP connectors, and through Marketer MCP create tests and personalisation variants. | Write down which agents, connectors and MCP tools are allowed for which content, and start with the narrowest set that is useful. |
| Agent work to live experience | Where does approval live? | The review card approves a change before it is saved; publishing workflow approves what goes live. Changes made from an external AI client reach the platform by a different route. | Decide which approval is the one of record, and make sure changes from every route pass through it. |
| Live experience to next brief | How does learning return to planning? | Test and personalisation results are visible in the authoring flow. Turning them into a revised brief or a changed brief type is left to the team. | Agree who reviews results on a fixed cadence and which brief types they are allowed to change. |
Every capability in the chain above can be switched on while all five questions remain unanswered. Agents then make changes faster than anyone can say whose intent they serve. Write the answers down before you connect the features: that written set is your operating model.
Trade-offs worth naming
- Reach against visibility. MCP lets people act on SitecoreAI from the tools they already use. Each additional client is another route into the platform that governance has to cover, and per-user connectors are private by design.
- Consistency against fit. Few brief types keep intent comparable across campaigns. Many brief types fit each campaign better and make results harder to compare when learning returns to planning.
- Review depth against throughput. A review card per change is proportionate for one page. Parallel per-item workflows across hundreds of pages produce more proposals than a person can review with care, so bulk work needs sampling rules or a narrower remit.
- Breadth of context against precision. One broad brand context is easy to maintain. Separate contexts for regions or business units ground agents more precisely and multiply the documents someone has to keep current.
Where to start
Our recommendations follow the order in which a failure would travel down the chain.
- Answer the five questions before connecting the next feature. A short written answer per connection is enough to begin.
- Start with brief types and brand context. Everything downstream reads them, so a weakness there is copied into every agent run, test and personalised variant.
- Open one route for agent change first. Agentic Studio inside Page builder, with the review card, keeps the change next to the page and a person in the loop. Add MCP clients once approval of record is settled.
- Design the return path explicitly. Decide which results can change which brief type, and who makes that call. This is the connection the platform leaves most open.
- Treat preview and early-access capabilities as experiments. Keep them out of processes that other teams depend on until they are generally available.
Most expensive mistakes in a connected platform begin before execution, when intent is vague or context is wrong. That is why our signal-to-execution model puts evidence and interpretation ahead of orchestration. It shows how a Digital Signal moves from outside-in evidence to a proportionate decision, and then to delivery that produces the next signal.
If AI visibility is the signal you most want to act on, the same five questions apply to the loop we describe in how Signals, powered by Scrunch, turns AI visibility changes into content opportunities.
