A Third of Companies Have Skipped Software Purchases Because AI Can Build It

Nearly a third of organizations have decided against buying at least one software product or feature in the past year because they could build it internally using AI coding agents, according to McKinsey’s State of AI 2026 survey. The survey drew 1,719 responses across 97 countries and was published August 25.

32% is the headline number. But the survey also found that internally built systems succeed roughly half as often as vendor tools — 33% success rate for internal builds versus 67% for purchased software.

That gap is the part worth paying attention to.

Who’s doing this most

The build-versus-buy shift isn’t uniform across industries. Technology firms lead at 41%, meaning two in five tech companies have skipped a software purchase because AI coding could cover it. Healthcare payers and providers come next at 39%, followed by professional services and energy at 38%, financial institutions at 36%, and pharma at 33%.

Among the top 6% of AI performers — companies generating at least 5% of EBIT from AI — the share is much higher: nearly half have skipped software purchases, compared to 31% of their peers. The companies that are extracting the most value from AI are also the ones most aggressively using it to avoid software spending.

Large enterprise adoption is also moving fast. Organizations with more than $1 billion in revenue report 40% now scaling agents across one or more functions, up from 27% in the prior year’s survey.

The success rate problem

McKinsey’s data includes a cautionary counterweight to the build trend. Internal systems built with AI coding agents succeed about 33% of the time. Vendor software succeeds at about twice that rate.

This doesn’t necessarily mean the 32% skipping purchases are wrong to do it — software budgets are large, and even a partially successful internal build can save significant spend. But the number suggests that a meaningful portion of these projects are generating costs, delays, or failures that don’t show up in the “we skipped a purchase” statistic.

Gartner’s CIO survey data from 2026 adds more context: only 17% of organizations have deployed agents, and only 11% have systems that are actually production-ready. That’s a wide gap between deciding to build and shipping something that runs.

Gartner’s forecast is that 40% of agentic AI projects will be canceled before the end of 2027.

The operating cost side

McKinsey partner Lieven Van der Veken noted that organizations need to treat operating costs as a design constraint when deciding where to build. That’s the part the build-versus-buy analysis often misses: software licenses have predictable costs, while AI agent operating costs can compound unexpectedly as usage scales.

20% of organizations surveyed are already experiencing strain from AI operating costs — a number that will likely grow as more projects move from pilot to production.

The decision to skip a purchase isn’t just about whether the feature can be built. It’s also about whether the team can maintain it, whether the operating cost stays reasonable at scale, and whether internal capacity would be better spent elsewhere.

What this means for software vendors

The survey is a real signal for enterprise software companies. A third of their customers are actively evaluating whether to build rather than buy, and AI coding agents have made that calculation plausible for a much broader range of features than it was two years ago.

The features most at risk are narrow, workflow-specific tools that don’t require deep domain expertise to replicate — internal dashboards, reporting pipelines, basic automation, integrations that are expensive to license but straightforward to write. Those are exactly the kinds of tasks AI coding agents handle well.

The features least at risk are ones that require compliance certifications, audit trails, legal accountability, or integrations that took years to build. Software vendors that can point to those properties as core to the product have more durable positioning than vendors competing primarily on features that an AI agent can write in an afternoon.

Sources: McKinsey State of AI 2026, Yahoo Finance, Startup Fortune