For most of my career, the enterprise architecture repository has had a readership problem. We put real effort into application inventories, capability maps, and technology standards, and then watched people make decisions without looking at any of it. The data was "there for anyone who needed it." Mostly nobody went looking.
A few things I read this week suggest that's changing, and not because people suddenly love fact sheets. The new readers of our repositories are AI agents. That raises a question we haven't had to answer so directly before: if a machine is going to reason over our architecture data and hand the answer to a finance lead or a project team, is that data good enough?
SAP LeanIX customers are wiring the repository straight into their assistants. In a November 3 post, SAP LeanIX described what it calls "architecture intelligence": exposing EA data to agents such as Claude, Joule, and Microsoft Copilot through its MCP server. The customer examples are the useful part. Kimberly-Clark's EA team built ARCHie, a Copilot Studio agent that combines their SharePoint standards site with the live LeanIX repository so employees can ask architecture questions in plain language. Norfolk Southern connected LeanIX to NSIQ, a company-wide assistant that also covers HR, operating rules, and the CMDB. According to the post, the team tracks 300 business applications against 422 business capabilities and has reached 99% coverage of business capabilities. Philip Morris International reportedly used clusters of agents in its architecture review process and cut document preparation from two days to two hours.
Reporting is becoming a conversation. On November 9, practitioner Rene de Daniel described a LeanIX assistant feature that turns a plain-language question ("show me all active applications in the Retire lifecycle phase tagged as Business Critical") into a ready-to-use report built from the workspace's real fact-sheet types, tags, and lifecycle states.
Bizzdesign is moving the same way. Bizzdesign's November 11 Unify release notes say its product documentation is now available to AI agents over MCP. The same release says multi-conversation management for its AI co-worker is ready for everyday use. In a LinkedIn post that week, the company said it wants more people to work with enterprise information "without needing specialist architecture skills."
And the agents themselves need to be inventoried. Gartner published a press release today based on a survey of 298 infrastructure and operations leaders in the U.S. and U.K. Ninety-five percent of their organizations had at least one agent-related incident in the past 12 months, and only 15% have automatic circuit breakers for runaway agents. Gartner's first recommendation is to treat every AI agent as a machine identity and track it in a centralized registry "so nothing runs in the shadows." A November 11 IASA Global article on the "Agent Control Plane" lists the fields that registry should hold: ownership, purpose, models, data access, tools, risk classification, and lifecycle status.
These stories point in two directions, and both lead back to the repository.
First, the repository is becoming context for AI. Once an assistant answers "what are we spending on sales and marketing tooling?" from your EA data, the repository is no longer just documentation. It's a production data source with real consumers, and its errors show up in other people's answers without anyone noticing. A stale lifecycle date or an orphaned application used to be a minor housekeeping problem. Now an agent can repeat it with confidence to someone who will never open the tool.
The Norfolk Southern detail I keep coming back to isn't the AI assistant. It's the 99% capability coverage. That's the boring groundwork that makes the assistant worth trusting. DataGalaxy made a similar point in a recent post: MCP carries context, but it doesn't make that context accurate, approved, or current. If you build AI connectivity before deciding what the AI should trust, you've done it in the wrong order.
Second, agents are becoming things the repository has to describe. Look at the questions LeanIX suggests people ask: "Does this type of AI agent already exist in the organization?" "Does it enhance or create AI tech debt?" Your repository can only answer those if agents are modeled as first-class objects, linked to the capabilities they support, the applications and data they touch, and the people who own them. Compare that list with the registry fields IASA describes and the inventory Gartner is telling I&O leaders to build. They overlap a lot. If EA doesn't step up, someone in security or operations will build a separate agent registry, and we'll have yet another inventory to reconcile.
There's also a quieter shift in who uses EA. Kimberly-Clark's team said EA had been seen as "the team that always tells us no." A plain-language front door, backed by the standards and the live repository, changes that. Routine questions get answered without a meeting, and architects can spend their time on the decisions that actually need judgment.
Treat repository quality as a service level. Choose the handful of attributes an agent is most likely to be asked about, like lifecycle, ownership, business criticality, cost, and capability mapping. Measure their completeness and freshness and publish the numbers. If you wouldn't stand behind an answer built on that data, don't expose it yet.
Expose a curated slice first. Start with a read-only connection over well-maintained data and published standards, the way Kimberly-Clark paired the repository with its standards site. Expand only as quality improves.
Add "AI agent" to the metamodel now. Model agents with owner, purpose, models used, tools and data accessed, risk tier, and lifecycle status. Link them to capabilities and applications. Then offer that view to security and I&O as the registry they're being told to build, so they don't have to start from scratch.
Make every answer traceable. Answers should link back to the fact sheet or standard they came from. That's how users learn to trust the assistant, and how you find the data problems it exposes.
Keep humans on the decisions that matter. Let agents handle intake, document preparation, and routine lookups. Keep architects accountable for exceptions, trade-offs, and anything with real business consequence.
For years we've talked about making architecture "actionable." It turns out the fastest way there may be making it machine-readable and trustworthy enough that an agent can act on it. That's less glamorous than a new diagramming feature, but it's the work that will decide whether AI makes our practice more relevant or just more visibly wrong.
Catalina Butnaru, "Architecture intelligence: EA's next defining moment," SAP LeanIX, November 3, 2026. https://www.leanix.net/en/blog/architecture-intelligence
Rene de Daniel, "Your question deserves a working report, not a configuration session," LinkedIn, November 9, 2026. https://www.linkedin.com/posts/rene-de-daniel-b108a413_sapleanix-sapeaf-enterprisearchitecture-activity-7525452564117700608-zSPI
Bizzdesign, "Bizzdesign Unify release notes 2026" (November 11, 2026 update). https://help.bizzdesign.com/articles/unify-help/bizzdesign-unify-release-notes-2026
Bizzdesign, "With Bizzdesign Unify, we wanted more people to be able to work with enterprise intelligenceā¦," LinkedIn, November 2026. https://www.linkedin.com/posts/bizzdesign_with-bizzdesign-unify-we-wanted-more-people-activity-7525520225992241153-BZmY
Gartner, "Gartner Says Weak AI Governance Will Lead to Business-Critical Service Disruptions for 40% of I&O Organizations by 2028," November 16, 2026. https://www.gartner.com/en/newsroom/press-releases/2026-11-16-gartner-weak-artificial-intelligence-governance-will-lead-to-business-critical-service-disruptions-for-40-percent-of-infrastructure-and-operations-organizations-by-2028
IASA Global, "Beyond the AI Pilot: The Agent Control Plane as An Enterprise Architecture Capability," published via ATD Solution, November 11, 2026. https://atdsolution.com/resource/the-agent-control-plane-as-an-enterprise-architecture-capability/
DataGalaxy, "MCP and AI Context: How MCP Connects Agents to Enterprise." https://www.datagalaxy.com/en/blog/mcp-and-ai-context/