A Third of Companies Now Build Software Rather Than Buy It

For years the safe corporate default was to buy software rather than build it. A new survey from McKinsey suggests that habit is starting to crack. In the firm's State of AI 2026 report, 32 percent of organizations said they had decided against buying at least one off-the-shelf product or feature because they could build it themselves with agentic coding tools.

Among technology companies the figure rises to 41 percent, according to coverage of the report by AIwire. The shift is sharpest at the top. McKinsey labels the roughly 6 percent of respondents who credit at least 5 percent of their profit to AI as "high performers," and nearly half of that group are now skipping software purchases, against 31 percent of everyone else.

The gap between trying and doing

The headline number is striking, but the survey's more useful finding is how few companies have actually made this work at scale. Roughly three quarters of organizations say they are adopting agentic AI in some form, yet only a small minority have moved past pilots into real production. On agentic coding specifically, about two in ten organizations are scaling it, rising to around a third at larger enterprises.

That gap matters. Building your own tool is easy to start and hard to finish. Someone has to maintain it, secure it, and keep it running when the person who wrote it moves on. The old logic of buying software was never only about the code. It was about handing the upkeep to someone else. Agentic tools lower the cost of writing something; they do not remove the cost of owning it.

The bill arrives later

McKinsey also flags a quieter problem. About 20 percent of organizations say they are already feeling the pinch of AI operating costs. The price of building has fallen, but running these systems, with their tokens and their compute, is a recurring expense that shows up every month rather than once at purchase. A build that looked cheaper than a subscription can turn out to cost more over its life.

None of this means the buy-it-all era is coming back. The direction is clear enough, and it fits a pattern we have tracked elsewhere, from AI writing half of a company's engineering tickets to the mixed evidence on whether it truly makes developers faster. What the McKinsey numbers really describe is a widening split: a small group getting real returns and pulling ahead, and a much larger group still stuck at the pilot stage, paying for the experiment without yet seeing the payoff.