There is a version of the same meeting I keep having on both sides of the Atlantic.

A CCO or CRO walks me through their AI plan. They are smart. Their team is credible. The slides are good.

They want to build their own GPT. Their own agents. Their own workflows. Their own application layer, with Darwin's intelligence flowing in through MCP.

And every time we get to the buy-versus-build question, they say the same thing. Our data is unique. Our stakeholders are unique. Our regulatory environment is unique. We can build it better ourselves.

I have started asking them a question in response. Have you ever led an enterprise software transition before?

Most of them have. And most of them, when they think about it, remember the same pattern.

In the late 1990s, every Fortune 500 built custom ERP, CRM, and HR systems for exactly the same reasons. Our business is unique. Off-the-shelf will not fit. By the mid-2000s, almost all of them had migrated to SAP, Oracle, and Salesforce, and quietly dismantled the internal teams that built the custom stacks.

Then the same story played out with intranets in the mid-2000s. And again with data platforms in the late 2010s.

AI is now in the early-optimism phase of the same cycle. And it is moving faster than any of the previous three.

Here is the pattern that held across those cycles. The parts of the stack that are truly proprietary get built internally. The parts that require scale, specialized data, or continuous investment get bought. Companies that got this ratio right compounded advantage. Companies that tried to build everything spent a decade catching up.

The CCOs and CROs I am meeting with are not wrong to want to build. When they say they want to build the orchestration and workflow themselves and connect to a specialized intelligence layer through MCP, that is the correct architecture.

The mistake is when the build side of the ratio expands to include things enterprise software has already proven cannot be built and maintained by a single company.

The question is not build versus buy. It is where the line sits, and whether your organization has honestly done the math on which side each component actually belongs.

CCOs, CROs, and the boards they report to: which components on your AI roadmap are you building because they are genuinely differentiating, and which are you building because it feels safer than depending on someone else? The honest answer to the second is where the maintenance debt of 2027 gets created.