A third of firms skipped a software purchase to build it

The line that should stop every SaaS founder cold is buried in a survey of 1,719 executives: nearly a third of organizations skipped buying at least one software product or feature because they could build it internally with agentic coding tools. That is McKinsey's State of AI 2026, and it is the first time the build-versus-buy question has moved on a number this large. Before you cancel a subscription, read the rest of the same survey — because it also says almost nobody is converting AI into earnings.

What actually happened

McKinsey surveyed 1,719 professionals and business leaders worldwide for The State of AI: Global Survey 2026. As The Register summarized it, nearly a third of respondents said their organizations chose not to purchase software products or features and built the functionality internally with agentic coding tools instead.

The adoption numbers underneath it are real. 40% of respondents at organizations above $1 billion in annual revenue say they are scaling AI agents, up from 27%. 80% of respondents who use AI in their own work say it improved their individual productivity.

Then the financial numbers land. 37% attribute at least some EBIT impact to AI — about the same share as the 2025 survey. Only 6% clear McKinsey's high-performer bar of more than 5% of EBIT attributed to AI plus self-reported significant impact. A year of aggressive deployment moved the earnings line for roughly one company in sixteen.

(Some outlets have published a precise 32% figure and a per-industry breakdown for the build-not-buy stat. We could not confirm those against McKinsey directly, so we are using "nearly a third.")

Why build-versus-buy matters for your business

Two facts sitting next to each other: the cost of building software fell hard, and the returns on AI stayed flat. That combination does not mean "build everything." It means the cost of building fell faster than most teams' judgment about what is worth building.

Here is the failure mode we already see in the wild. A team cancels a $400/month vendor and ships a replacement in a weekend with Claude Code. Real win. Six months later that replacement has no owner, no tests, an expired API token, a schema nobody documented, and a founder who cannot take a vacation. The $4,800 a year they saved is now a $30,000 rebuild. The vendor was never selling the feature — it was selling the fact that someone else gets paged at 2am.

The rule we use, and it has not changed because the tooling got better:

Build what is your differentiation. Your pricing logic, your fulfillment rules, your quoting math, the workflow that makes you faster than the shop down the street. Nobody sells that as a product because it is yours. Agentic coding tools make this cheap enough to finally be worth it.

Buy the commodity with a compliance surface. Payments, payroll, email deliverability, auth, tax. You are not buying code there, you are buying somebody else's liability and audit trail. An agent can write a Stripe integration; it cannot be PCI DSS compliant on your behalf.

Price the build honestly. The build cost is not the weekend. It is the weekend, plus hosting, plus the on-call, plus the upgrade you do in eighteen months when a dependency dies. If the honest number still beats the subscription, build it — and write down who owns it before you merge.

The 6% high-performer number is the real lesson in this survey. The winners are not the teams doing the most AI. They are the teams that picked a small number of things where automation actually touches revenue, and finished them.

Sources: McKinsey, The State of AI: Global Survey 2026, The Register.