McKinsey: 32% of companies have skipped buying software because AI coding agents let them build it instead
McKinsey's State of AI 2026 survey of 1,719 organisations found that 32% have forgone buying a software product or feature because agentic coding tools let them build it internally instead, with tech (41%) and healthcare (39%) leading the shift — but the survey never asked whether those builds shipped, passed audit, or are still running a year later.
9 September 2026
McKinsey’s State of AI 2026 report, drawn from 1,719 responses across 97 countries, found that 32% of organisations have decided against buying an off-the-shelf software product or feature because agentic coding tools made it feasible to build the equivalent internally. The shift isn’t evenly spread: technology firms lead at 41%, followed by healthcare payers and providers at 39%, professional services and energy and materials at 38%, financial institutions at 36%, media and telecom at 34%, and pharma at 33%. This is a live re-run of the classic build-vs-buy decision, except the “build” side of the ledger just got dramatically cheaper to start.
The number everyone’s quoting, and the number nobody’s asking
The 32% figure is being repeated everywhere without its most important caveat: McKinsey asked whether organisations had decided to build instead of buy — not whether the replacement actually shipped, whether it passed a security or compliance audit, or whether it’s still running and maintained a year later. The same report notes that the share of companies seeing measurable profit impact from AI hasn’t moved in a year, sitting at 37%. Put those two data points together and a pattern emerges: a lot of “we built it ourselves” decisions are happening at the enthusiasm stage, before anyone has priced in the ongoing cost of running, patching, and securing software that an agentic tool assembled quickly but nobody architected for the long term.
Why this is a genuine shift, not just hype
It would be wrong to dismiss this as noise. Agentic coding tools have made the first 80% of a software build — scaffolding, CRUD, integrations, a working prototype — fast enough that “just build it” is now a real option for problems that used to default to a SaaS subscription or an off-the-shelf tool. For well-scoped, low-complexity internal tools, that’s often the right call. The risk sits in the remaining 20%: production-grade auth, data governance, the integration points that fail under real load, and the maintenance burden six months after the person who “vibe coded” it has moved to another project.
So what
The organisations getting genuine value from this shift are the ones treating agentic tools as an accelerant for a properly engineered build, not a replacement for engineering judgment — knowing which internal tools are safe to spin up quickly and which need the same rigour as anything shipped to customers. If you’re weighing whether to build a piece of software in-house or bring in a team that can take it from working prototype to something you can actually run and support, that’s exactly the conversation to have early. See our custom software development work, or start a project to scope it properly before you commit engineering time either way.