Don’t Buy Someone Else’s AI Reference Model (A EACOE™ Position Paper)
Where David Linthicum Is Right, Where the Terms Need Correcting, and Why the Problem Was Never the Framework
On October 6, 2026, David Linthicum published “Don’t buy a consultant’s AI framework” in InfoWorld. His argument is direct: prebuilt generative AI and agentic AI blueprints from professional services firms anchor both the consultant and the client to a static idea, the client’s problem gets forced into it, and the result is an architecture that, in his direct experience, costs 10 to 20 times more to operate than one derived from the client’s own requirements. He closes with a rallying line: “Requirements first. Architecture second. Frameworks last, if at all.”
The Enterprise Architecture Center Of Excellence (EACOE) agrees with the substance of that argument. In fact, it restates a position EACOE has published for more than a decade: there is no such thing as an Enterprise Architecture that will fit any organization, buying “reference” or vendor-provided models is a shortcut that does not work, and you have to do the work.
Where we differ is in the vocabulary, and the vocabulary matters. What Linthicum calls a “framework” is not a framework. It is a reference model, a reference architecture, or a packaged blueprint: someone else’s architecture output, pre-filled with someone else’s answers. A true framework, as EACOE defines it, is a thinking tool. It has no output, so it cannot contain a vector database, a multi-agent orchestration layer, or an “AI brain.” Telling executives to put frameworks last, if at all, accidentally tells them to discard the one tool that makes their requirements complete.
This paper is a companion to two earlier Architectures Center Of Excellence articles: Large Language Models (LLM): The Latest Fool’s Gold? and Reference Models and Reference Architectures: Caveat Emptor. Linthicum’s article joins them as a third case of the same mistake: buying someone else’s model instead of doing the architecture of your own enterprise.
This paper makes four points:
1. Linthicum is right about the outcome. Buying someone else’s AI blueprint and architecting against it produces over-built, over-priced, slower implementations.
2. The term confusion matters. A framework is a thinking tool. A reference model is someone else’s answer. Collapsing the two leads organizations to throw out the thinking and keep the answers.
3. The corrected sequence is Framework first, Requirements second, Architecture third, and someone else’s reference model last, if at all. The framework is what tells you whether your requirements are complete.
4. “AI reference model” is simply the newest member of an old family. EACOE has argued for years that buying someone else’s “reference anything” (industry models, enterprise blueprints, packaged best practices, off-the-shelf ontologies, and now AI reference architectures and the large language models inside them) will cost more and take longer than doing the architecture of your own enterprise.
