AI needs a visible place in enterprise architecture
My proposed Cognitive Enterprise Architecture Framework makes the relationships between AI, business decisions, ownership, and execution explicit.
AI deserves an explicit place in the way an enterprise describes itself. When models influence recommendations, priorities, and operational actions, their relationships with strategy and governance need to be visible. Describing an application and its database alone may leave important questions about decision ownership unanswered.
That is the problem I addressed in my graduate capstone through the Cognitive Enterprise Architecture Framework, or CEAF. I proposed an explicit AI layer within an enterprise view connecting strategy, business, data, applications, and technology. The aim is to make AI-enabled decisions easier to trace and govern across those relationships.
CEAF is a conceptual proposal. It has not been empirically validated through organisational implementation. I see it as a contribution to examine and refine, with practical value to be demonstrated through use. That boundary matters because a coherent architectural argument and evidence of improved enterprise performance are different achievements.
Make the decision traceable
The central idea is a trace from strategic objective to business capability, supporting data, model behaviour, application workflow, decision, and outcome. Each connection asks a useful question. Which objective justifies the capability? What information supports the model? Where does its output enter the work? Who remains accountable for the resulting decision?
Take an illustrative material-allocation recommendation. The strategic aim could be dependable delivery. The relevant capability is allocation of constrained supply. The model may draw on demand, available stock, and production priorities, while an application presents a recommendation to a planner. The architecture should show who maintains those inputs and who may authorise a change in allocation.
Without that trace, a model can appear technically successful while its operational role remains ambiguous. A recommendation may be accurate against a test set yet arrive too late, use a definition the receiving team rejects, or suggest an action outside the user’s authority. The enterprise view connects these conditions to the original business purpose.
A visible layer does not mean an isolated function
In CEAF, the AI layer is a modelling device for making responsibilities and dependencies explicit. It does not require a new department between data and applications. AI also raises questions across the enterprise: workforce skills, information quality, procurement, infrastructure, monitoring, and governance. Those relationships should remain connected to their existing owners.
Established enterprise architecture frameworks can already represent AI-related concerns. My proposal draws on familiar architectural ideas and adds emphasis to decision behaviour, ownership, and execution. It is an extension of my own architectural approach, rather than a claim that existing frameworks have been formally revised or cannot address AI.
The same distinction applies to model change. Some deployed models remain fixed until deliberately updated; others operate within systems that change components or use new information. Governance should describe the actual mechanism. It should identify which changes require evaluation and approval, instead of assuming that every AI service continuously learns in production.
Evaluate usefulness in a bounded setting
I would begin testing CEAF around one AI-supported business decision. The first task would be to document the current ownership, dependencies, and failure points. The second would be to use the proposed view to clarify gaps and agree changes. The evaluation would examine whether that work improves decisions and reduces unresolved responsibilities.
Evidence could include clearer exception ownership, more complete change-impact assessments, or less time spent reconstructing why a model is used. These would be evaluation candidates, not results I can claim today. The framework should earn its place through the quality of the decisions it helps people make.
My broader position is that enterprises need to understand how AI participates in their operating system. CEAF offers a structured way to investigate that question while keeping strategy, people, and accountability in view.
Developed from my EA 878 graduate capstone, Cognitive Enterprise Architecture Framework (CEAF), dated 1 May 2026. CEAF is my conceptual proposal and has not been empirically validated through organisational implementation. The example and proposed evaluation approach are illustrative.