The build-vs-buy line just moved | Thursday, September 3, 2026

Among the small group McKinsey calls high performers, the roughly 6 percent who attribute at least 5 percent of EBIT to AI, nearly half skipped a purchase, against 31 percent of everyone else. Large enterprises are leaning in hardest: 40 percent of companies above a billion dollars in revenue now scale agents in one or more functions, up from 27 percent a year ago. That is a real shift in procurement behavior, and it is easy to read as vindication for the build side of the oldest debate in enterprise IT.

It is also, on the current evidence, a setup for a wave of expensive disappointment. The same research that shows the enthusiasm shows the execution gap underneath it. MIT's NANDA initiative found that internally built AI systems succeed roughly a third of the time, while vendor-purchased tools land closer to two-thirds. Gartner projects that more than 40 percent of agentic AI projects will be cancelled by the end of 2027, citing escalating costs, unclear business value, and inadequate risk controls. And the production numbers remain stubborn: Forrester put agentic adoption near 75 percent this summer, but Gartner's 2026 CIO survey finds only 17 percent of organizations have actually deployed agents, and Deloitte counts just 11 percent with production-ready systems.

The trap is not the decision to build. It is the accounting. A purchased tool bundles a cost that a self-built one hides: the vendor's roadmap, security patching, uptime, compliance, and the salary of the person who fixes it at 2 a.m. When an agent writes the first draft in an afternoon, that hidden cost does not disappear. It moves onto your balance sheet, usually unpriced. McKinsey's own data shows the pressure already arriving. Twenty percent of organizations say AI operating costs, token spend included, are constraining how much AI they can use, and while 80 percent report individual productivity gains, only 37 percent see the EBIT impact that would justify a build program at scale. Lieven Van der Veken, the McKinsey senior partner behind the report, names the discipline that separates the two groups: the leaders treat operating cost as a design constraint rather than an afterthought.

Moving Pieces

On the Radar

The Number

Counter-Signal

It is worth steelmanning the side this edition is skeptical of, because the build enthusiasm is not irrational. For a high performer already extracting real EBIT from AI, agentic coding tools genuinely change the arithmetic: the internal team is fluent, the operating model exists, and a custom system built in the flow of their own work will beat a generic tool that assumes someone else's process. For those organizations, the 33 percent success rate is a population average that does not describe them, and Van der Veken's framing holds, they build selectively, price the run cost up front, and treat buying as the default only for the undifferentiated.

The failure statistics are real, but they are dominated by organizations that mistook a cheap prototype for a cheap product. The lesson is not build or buy. It is that neither is a strategy without the discipline to run the whole number, and the firms that have that discipline can build more this year than they safely could last year. The tools earned them optionality. What they do with it is still a management decision, not a technology one.

From the Field

I have watched more than one custom build get approved in a room where everyone was looking at the wrong number. A capable engineer stands up something that works in an afternoon, the room applauds, and the project is greenlit against a cost estimate that quietly assumed the afternoon was the hard part.

The hard part is the eighteen months after, when the model underneath the system gets deprecated, the one person who understood the prompts takes another job, a regulator asks how a decision was made, and the token bill for a workflow you now depend on turns out to scale with your success rather than your budget. None of that shows up in the demo, and all of it shows up on the balance sheet. So when someone on your team proposes building instead of buying this quarter, do not argue with the enthusiasm. It is often correct. Argue with the accounting. Ask for the five-year run cost, the named owner, the patching plan, and the exit if it fails. If they can answer, build it, because a system core to how you compete is worth owning. If they cannot, you have not found a project. You have found a liability that happens to work today. The tools got cheaper. The discipline did not.