AEM Sites Playbook, Chapter 2: Platform, hosting and experience architecture
Which platform, authoring and delivery choices fit the requirements?
November 10, 2026
Separate the platform from the delivery choice
Make two decisions explicitly. First, identify the AEM deployment and operating arrangement. Second, decide how authors create content and how customers receive it. Conflating them hides responsibilities: moving to Cloud Service does not by itself select a frontend, content model or measurement design.
Scroll horizontally if needed.
| Context | Establish before choosing a design |
|---|---|
| AEM 6.5 | Exact service pack, runtime, extensions, hosting/support responsibilities and upgrade path |
| AEM 6.5 LTS | Supported source/target path, runtime and dependency compatibility, removed legacy dependencies and authoring behavior |
| AEM as a Cloud Service | Cloud-compatible application behavior, deployment/configuration boundaries, API access and responsibility for custom code |
| AMS | Contracted hosting and operational services, the underlying AEM release and the team's remaining responsibilities |
Adobe's LTS FAQ and Cloud Service development guidelines describe different requirements. Use the appropriate branch when assessing a design. Record the review date because release support and compatible components change.
Choose an authoring and delivery contract
AEM-rendered pages give the team one set of composition and rendering responsibilities. Headless delivery separates structured content from the consuming application's presentation. A hybrid approach combines these where useful. Edge Delivery offers another delivery approach, with document-based authoring and AEM/Universal Editor choices that need their own assessment. Use the headful/headless explanation and Edge Delivery overview to establish the boundaries.
Ask authors to demonstrate the expected editing experience. Can they understand references, preview a change, identify where it will appear and request the right approval? Ask frontend teams to demonstrate delivery, authentication, caching, analytics and failure handling. A content API returning JSON is not a complete editorial preview or customer experience.
The Experience Fragment and document-based Edge Delivery discussion illustrates a reuse assumption that needs testing. Suggestions about fragment blocks and separately managed content do not prove that traditional Experience Fragments migrate unchanged. Keep document-based examples distinct from Universal Editor examples when comparing implementations.
Run a bounded architecture trial
Choose a representative slice: a localized page with a shared image, or a product fact used by both a website and an app. Include the difficult handoffs, such as a protected API, translated reference or third-party dependency. Measure the work required to create, preview, approve, deliver, correct and remove the content.
Document who owns each boundary: content schema, frontend, public domain, access control, cache policy, instrumentation, alerting and support. If two teams believe the other handles freshness, the trial has exposed an operating gap. Resolve it before estimating a wider rollout.
For an existing application, inventory dependencies on instance identity, local state and editor-specific behavior. In the AMS-to-Cloud Publish-job discussion, the design depended on a fixed Publisher sending an external update once. That assumption belongs in the migration assessment, even when the application's pages render successfully.
Turn architectural preferences into testable choices
Community discussions illustrate two questions to test before choosing an architecture: whether headless delivery improves the actual operating model, and whether existing shared content can be reused with the selected Edge Delivery authoring approach. The examples leave those benefits or reuse outcomes unconfirmed. Include editorial work and representative existing content in the architecture trial.
First classify each requirement as mandatory, desirable or an assumption. A regional author's ability to preview an approved composition may be mandatory. A preference for a particular framework is an implementation choice. Keep them in separate columns so a technically attractive option cannot compensate for a failed editorial requirement.
Scroll horizontally if needed.
| Requirement or constraint | Compare these options | Evidence needed |
|---|---|---|
| Authors control composed pages | Retain AEM rendering; assess another editor/delivery model | A normal author builds and previews the required composition |
| Multiple applications consume the same facts | Structured content with bounded consumers; existing duplication | Model ownership, consumer compatibility and update/withdrawal tests |
| Independent frontend releases are required | Separate frontend delivery; coordinated application releases | Frontend operation, preview, identity and observability ownership |
| Existing shared fragments must survive a move | Adapt the reuse model; retain selected experiences temporarily | Representative reference migration and authoring, not just old HTML rendering |
| Infrastructure/release responsibilities must change | Applicable hosting/platform options | Actual contract, support model, source/target compatibility and operating capability |
Write a responsibility map for every candidate, even the preferred one. A diagram showing an API between two boxes should identify who owns its availability, schema changes, credentials, freshness and incident response. Unassigned responsibility is an unresolved requirement, not an implementation detail to settle after procurement.
Worked scenario of a mixed delivery decision
This scenario and its results are hypothetical. A company has an established AEM website in three languages and a separately operated mobile application. It wants product facts shared with the app, while website authors must retain visual page composition. Its sponsor proposes replacing the entire website frontend to achieve consistency.
The architect compares three candidates on one product change, one local edit and one withdrawal:
Scroll horizontally if needed.
| Candidate | Intended advantage | New obligations | Trial question |
|---|---|---|---|
| Keep the current website and copying process | Small immediate change | Continued reconciliation | Can ownership/process changes alone meet the requirement? |
| Retain website rendering and introduce a structured-content consumer for the app | Bound the shared-data problem | Model/API contract, app preview and support | Can both channels use the approved fact without losing website composition? |
| Replace website and app delivery together | A common frontend approach | Wider migration, new author preview, routing and release operations | Is that wider change required by a mandatory need? |
The trial uses existing representative content rather than a new empty model. An editor changes a fact, a regional author retains approved local copy, and the mobile team demonstrates its own consumption and fallback. The team records the responsible person at each transition. Current structured-content API guidance supplies the applicable management/delivery boundary; it does not choose the organization's operating model.
Suppose the trial finds:
Scroll horizontally if needed.
| Check | Illustrative observation | Consequence |
|---|---|---|
| Existing website composition | Works unchanged in the bounded candidate | Keep this capability in scope |
| App receives the approved fact | Works for the agreed payload | Consumer contract provisionally supported |
| App preview of an unpublished proposal | No owned implementation | Mandatory review requirement remains open |
| Withdrawal | App retains content beyond the agreed requirement | Freshness/removal contract needs repair |
| New website frontend | No demonstrated need in this trial | Do not include it solely to solve data reuse |
The decision is to continue the bounded candidate, with rollout held until preview ownership and withdrawal behavior meet requirements. The trial supports a direction, not an unqualified launch. The sponsor can distinguish necessary work from optional redesign, and the mobile team knows which capabilities it must operate.
Define the point at which the architecture decision can change
Record assumptions that would invalidate the decision: another consumer needs a different schema, an editorial requirement cannot be supported, a legacy extension is incompatible, or the contract changes operating responsibility. Give each assumption a review owner and a way to observe it.
For Cloud Service, review interruption/cluster assumptions. For 6.5 or LTS, use the supported runtime and extension matrix for that branch. For Edge Delivery, verify the selected authoring model. Do not treat a successful trial on one branch as proof for the others.
The business owner accepts the editing/customer consequences. The platform owner accepts the responsibility map and compatibility. Each consuming team accepts its contract. A decision remains conditional until the named mandatory gaps are resolved or the approved scope is narrowed.
Make the decision reviewable
Complete W02 with the current state, requirements, alternatives, chosen design, responsibility map and trial results. Record unresolved constraints and the condition that would change the decision. Business owners approve the experience and authoring trade-offs; architects approve compatibility and operating responsibilities; practitioners retain the repeatable trial. Do not describe one successful demo as evidence that every brand, language, integration or volume requirement has been met.
W02 Architecture decision
Use with chapter 2. Record deployment and delivery/authoring choices separately.
Scroll horizontally if needed.
| Record | Entry |
|---|---|
| Current/target deployment, hosting and support arrangement | [enter] |
| Authoring, preview, frontend and delivery requirements | [enter] |
| Alternatives and reasons for selection/rejection | [enter] |
| Responsibilities for content, frontend, domains, security and operation | [enter] |
| Entitlements and contract assumptions needing confirmation | [enter] |
| Trial fixture, observed result and unresolved constraint | [enter] |
| Condition that would change the decision | [enter] |
