McKinsey’s State of AI 2026 survey, fielded May–June across 1,719 organisations in 97 countries and published 25 August, put a number on a shift a lot of software buyers have been sensing anecdotally: 32% report they’ve decided against purchasing at least one software product or feature because an agentic coding tool could build the functionality in-house instead. In technology companies that rises to 41%, and among the “high performers” — the 6% of organisations attributing at least 5% of EBIT to AI — it’s nearly half, against 31% of everyone else.
The number underneath the number
Here’s the part worth sitting with before treating this as a straightforward “build, don’t buy” verdict: the share of organisations reporting that AI is actually moving their EBIT held flat at 37% year-on-year. Build volume is rising fast; proven financial return isn’t rising at the same rate. That gap matters for anyone making the call today — a growing number of teams are choosing to build rather than buy, but a growing appetite to build isn’t the same evidence as a growing track record of it paying off. The organisations skipping purchases most confidently are the ones already running disciplined AI programmes with measurement in place, not the ones building opportunistically because a coding agent made it look easy.
So what
If your organisation is weighing whether to commission a bespoke build instead of licensing an off-the-shelf tool, this data says you’re not early to that decision — a third of comparable organisations have already made it. But it also says the decision is only as good as your ability to measure whether the build actually delivers, not just whether it shipped. That’s the gap between a prototype an AI coding agent can produce and a piece of software an organisation can run its operations on — which is where a properly scoped custom software development engagement earns its cost over a quick internal build. If you’re weighing build-versus-buy on a real project, get in touch and we’ll help you scope it honestly.