We tested Adobe's AI agents in a live AEM sandbox. Here's the unvarnished truth - plus the architecture, licensing, and enterprise adoption roadmap your team needs to act on today.
The paradigm shift you can't afford to ignore
Adobe isn't just adding AI features to AEM. It's fundamentally reimagining what a content management system is for.
The concept Adobe calls the "Agentic Web" signals a structural change in how digital experiences get built, delivered, and consumed. Content no longer exists solely for human visitors navigating pages in a browser. AI agents - from search engines like Google's AI Overviews to enterprise copilots and brand concierges - are now reading, interpreting, and redistributing your content at scale. Your CMS has to serve both audiences simultaneously.
For digital leaders, this reframes the AEM investment entirely. AEM is evolving from a page-rendering engine into what Adobe describes as a "brand experience management system" - a platform where content is structured for machine consumption, governed for brand integrity, and optimized by intelligent agents working alongside your team.
What this means for your team:
The question is no longer "should we use AI in AEM?" It's "how fast can we restructure our content operations, governance, and delivery architecture to work in an agent-driven world?" Organizations that treat this as a feature evaluation will fall behind those who treat it as an operating model transformation.
This article goes beyond the marketing narrative. We tested Adobe's AI agents in a live AEM as a Cloud Service sandbox, mapped the full agent ecosystem across Experience Cloud, identified what's real versus what's roadmap, and built an enterprise adoption playbook your team can start executing this quarter.
The full agent landscape: it's bigger than AEM
Most coverage focuses narrowly on AEM agents. That misses the picture. Adobe has built an orchestration layer - the Adobe Experience Platform (AEP) Agent Orchestrator - that coordinates specialized agents across the entire Experience Cloud. Think of it as the reasoning engine that decides which agents to activate based on your prompt, your data, and your permissions.
The key insight: your team interacts with a single conversational AI Assistant. Behind the scenes, the Agent Orchestrator determines which specialized agents to activate, coordinates multi-step workflows, and maintains conversation context. This is fundamentally different from isolated AI features - it's a coordinated intelligence layer across your entire experience stack.
The agents that matter most right now
- Experience Production Agent: Automates content updates, page creation, form building, and site migrations to Edge Delivery Services. The workhorse for day-to-day authoring teams.
- Content Advisor Agent: Discovers assets across the DAM, creates channel-ready variations using natural language, and streamlines asset reuse across campaigns.
- Development Agent: Analyzes pipeline failures in Cloud Manager, identifies root causes, and suggests fixes. Designed to reduce developer troubleshooting overhead.
- Governance Agent: Monitors rights expirations, enforces brand policies, manages permissions - all through natural language prompts instead of manual audits.
- Journey Agent (AJO): Create, analyze, and optimize customer journeys using natural language. Detect conflicts, analyze drop-off points, and replicate top performers.
- Data Insights Agent (CJA): Answers data questions in plain language, builds relevant visualizations in Analysis Workspace, and forecasts trends from your actual data.
What we actually tested - and what happened
Marketing narratives are one thing. We wanted to know what happens when you sit in front of AEM as a Cloud Service, open the AI Assistant, and start asking it to do real work. We ran structured tests across page authoring, asset operations, development workflows, and governance boundaries in an AEM Sites Trial sandbox.
Here's what we found.
Page content updates: the clear win
The Experience Production Agent performed well when editing existing pages. When prompted, it accessed the target page directly, identified the editable components on that page, and applied content updates without breaking the existing layout or component structure. For teams that spend hours coordinating simple content changes across components, templates, and approval chains, this is a meaningful productivity gain.
What worked:
The agent updated text, modified component content, and preserved page structure reliably. It understood the relationship between components and their editable regions - it wasn't blindly injecting text.
However, we hit a hard wall: the agent could not create new pages in the author environment during testing. It could modify existing pages but not generate new ones. This suggests the current implementation is optimized for content velocity on established page structures, not for site architecture creation.
Asset discovery and operations: mixed results
The Content Advisor Agent successfully located assets used on a given page and identified them within the DAM repository. Basic discovery worked as expected.
But deeper asset operations were restricted. The agent could not download assets locally, update asset metadata, or perform most asset modifications. One operation that did work - creating asset variations - required providing the delivery domain. Without the correct domain configuration, the agent couldn't generate variations at all.
Development and pipeline access: permission-gated
The Development Agent correctly recognized the sandbox environment and the user context but could not retrieve pipeline information from Cloud Manager. This appeared to be a permissions limitation within the trial environment, not a capability gap - but it highlights how tightly agent functionality is coupled to IAM configuration.
The big finding: permission-aware by design
Across every test, agents consistently recognized the environment they were operating in, the permissions available to the current user, and which actions were allowed or restricted. When an action required permissions the current user didn't have, the agent declined to attempt it. It didn't fail silently or produce errors - it acknowledged the boundary.
Why this matters for enterprise adoption:
This is the single most important finding for security and compliance teams. AI agents in AEM respect your existing RBAC model. They don't bypass permission boundaries, they don't escalate privileges, and they don't attempt unauthorized actions. For regulated industries, this is a prerequisite for adoption - and it's already built in.
Full test results at a glance
| Capability tested | Status | Notes |
|---|---|---|
| Update existing page content | Works | Reliable across text components; preserves layout |
| Create new pages | Blocked | Not available in current implementation |
| Locate assets on a page | Works | Correctly identifies assets and DAM paths |
| Create asset variations | Limited | Requires delivery domain configuration |
| Download / modify assets | Blocked | Not permitted in sandbox; likely permission-gated |
| Update asset metadata | Blocked | Not available during testing |
| Pipeline status retrieval | Blocked | Permission restrictions in sandbox |
| Environment awareness | Works | Consistently recognized context and permissions |
| Permission boundary enforcement | Works | Never attempted unauthorized actions |
The content strategy shift nobody is talking about
Here's an insight that has massive implications for your content team, and almost no agency is writing about it yet.
AI agents don't browse your website the way humans do. They don't click through navigation menus, admire hero images, or scroll through marketing copy. They consume raw HTML. They parse structured data. They read every word of your 2,000-line FAQ page - content that most human visitors ignore entirely - and redistribute that information to their users.
This fundamentally changes what content matters.
In AEM terms, this means content fragments become reusable knowledge units - not just building blocks for page composition, but durable records that agents can query, cite, and redistribute. Your metadata strategy, your content modeling, and your information architecture become the primary levers for how your brand appears in AI-driven discovery.
Edge Delivery Services reinforce this shift. By delivering content as clean, fast, JavaScript-light HTML directly from the edge, EDS ensures that AI systems can read and trust your content without needing to execute scripts or navigate complex rendering pipelines.
Action item:
Audit your AEM content model. How much of your "long-tail" content - FAQs, specs, policies, documentation - is structured in content fragments with proper metadata? If it's sitting in unstructured rich text components or buried in PDFs, AI agents can't effectively consume it. This is the single highest-ROI content architecture change you can make this year.
The elephant in the room: licensing, access, and the AEM 6.5 gap
Let's address what most vendor content carefully avoids: who can actually use this, and what does it cost?
Cloud-only. No exceptions.
AI agents in AEM are available exclusively for AEM as a Cloud Service and Edge Delivery Services. They are not available for AEM 6.5, AEM 6.5 LTS, on-premise deployments, or Managed Services environments. This is not a temporary limitation - it reflects the architectural requirements of the agent orchestration layer.
For the significant portion of the AEM install base still running on-premise or Managed Services, this creates an urgent migration question. Every quarter you delay cloud migration is a quarter your competitors are building agent-assisted content operations while you're still operating manually.
Three paths to access
Adobe currently offers three ways to access AEM agents: the "Try Before You Buy" program (rolling out in phases - contact your CSM/TAM for availability), a trial through your account team, and a full Agentic SKU license. For the broader AEP Agent Orchestrator, organizations need either an AI Credits license or the Agent Orchestrator Promo SKU (a time-bound trial). Agent usage consumes AI credits, which introduces a variable cost component on top of your existing AEM licensing.
AI-first apps are licensed separately
Sites Optimizer, GenStudio for Performance Marketing, and Brand Concierge are classified as "AI-first applications" with their own licenses - they don't require the Agent Orchestrator SKU but they also aren't included in your base AEM license. Budget accordingly.
Financial planning note:
Request a detailed licensing breakdown from your Adobe account team that maps each agent capability to its SKU and credit consumption model. Don't plan your adoption roadmap without understanding the cost structure - some agents are included in existing entitlements while others require new investment.
Don't wait for Adobe: the hybrid integration play
Here's the uncomfortable truth that Adobe's own content won't tell you: native Adobe AI capabilities are not the whole story. Enterprise-scale content operations often demand more flexibility, deeper integrations with non-Adobe systems, and orchestration patterns that go beyond what's available natively today.
The most successful organizations are adopting a hybrid approach - using Adobe's purpose-built agents where they excel while layering in complementary AI capabilities where gaps exist.
Where to layer in external AI now
Semantic search and discovery. RAG (Retrieval-Augmented Generation) pipelines using tools like Azure AI Search or Coveo AI can bring natural language understanding to content stored across AEM, SharePoint, and other systems - delivering more relevant search results than keyword-based approaches.
Developer productivity. Tools like GitHub Copilot, Cursor, and Windsurf are already accelerating AEM development workflows - generating Sling Models, component dialog stubs, and test scaffolding. These complement the Development Agent rather than compete with it.
Content intelligence at scale. Custom agents that identify stale or underperforming AEM content, auto-generate metadata for DAM assets using AI vision models, or enforce content quality standards can fill gaps in native capabilities today.
Cross-platform orchestration. If your marketing stack extends beyond Adobe (and whose doesn't?), custom agentic workflows that coordinate between AEM, Workfront, Microsoft 365, and other systems can deliver value that Adobe's ecosystem alone cannot.
Strategic guidance:
Don't position this as "Adobe vs. alternatives." The Agent Orchestrator is explicitly designed for multi-agent collaboration, including third-party agents through MCP (Model Context Protocol) and A2A protocols. Build with the grain of Adobe's architecture, not against it - but don't constrain yourself to only what Adobe ships natively.
The enterprise adoption roadmap
AI agents don't deliver value through a single proof-of-concept. They require a phased approach that builds foundations before scaling automation. Here's the roadmap we recommend.
Phase 1: Foundation (Weeks 1-4)
Confirm your AEM as a Cloud Service readiness
If you're on AEM 6.5 or Managed Services, agents are not available. Begin cloud migration planning now - every quarter delayed is a competitive gap widening. If you're already on AEMaaCS, confirm your entitlements and accepted GenAI terms.
Audit your content model for agent readiness
Map what percentage of your content is in content fragments vs. unstructured rich text. Identify long-tail content (FAQs, specs, docs) that lacks proper metadata. Prioritize converting high-value content into structured, fragment-based models.
Map your IAM model to agent capabilities
Agents respect your RBAC boundaries. Audit which roles need access to which agent capabilities. Ensure your permission model reflects what you want agents to be able to do - not just what humans can do today.
Phase 2: Pilot (Weeks 5-10)
Start with content updates and asset discovery
These are the most reliable agent capabilities today. Have your authoring team use the Experience Production Agent for routine content updates. Measure time saved versus manual workflows. Use the Content Advisor for asset discovery across your DAM.
Deploy the Development Agent for pipeline troubleshooting
Give your AEM developers access to the Development Agent in Cloud Manager. Track resolution times for build failures. This is a low-risk, high-visibility proof point that builds organizational confidence.
Phase 3: Scale (Weeks 11-20)
Extend beyond AEM to journey and analytics agents
If you're running AJO and CJA, activate the Journey Agent and Data Insights Agent. These extend agent value from content operations to customer experience orchestration - the bigger strategic play.
Layer in hybrid AI capabilities
Implement semantic search via RAG pipelines. Deploy developer productivity tools alongside the native Development Agent. Build custom metadata enrichment workflows for your DAM. This is where you differentiate from competitors using only native capabilities.
Phase 4: Optimize (Ongoing)
Evaluate AI-first applications and custom agents
Assess Sites Optimizer for automated site improvement. Evaluate Brand Concierge for customer-facing AI experiences. Build custom agents using Agent Composer and MCP protocols. Monitor AI credit consumption and optimize for cost efficiency.
Restructure content for the Agentic Web
Redesign your content strategy to serve both human visitors and AI agents. Invest in content fragments as durable knowledge units. Implement content authenticity signals. Position your brand to be accurately represented in AI-driven discovery.
The bottom line
AI agents in Adobe Experience Cloud are real, they're production-ready for specific use cases, and they're constrained by sensible enterprise guardrails. But they're not magic - they require cloud infrastructure, clean content architecture, proper IAM configuration, and a phased adoption approach to deliver value.
The organizations that will win aren't the ones waiting for agents to mature. They're the ones building foundations now: migrating to AEMaaCS, restructuring content models for machine consumption, and piloting agent-assisted workflows while their competitors are still reading announcement blog posts.
The Agentic Web isn't a future state. It's the present. The only question is whether your content operations are ready for it.
Ready to assess your agentic readiness?
Our team combines hands-on AEM engineering depth with enterprise AI strategy. We've tested these capabilities in live environments - not just read the documentation. Whether you need a cloud migration assessment, content model audit, or full agent adoption roadmap, we can help you move from exploration to execution.