32% of Companies Now Build Instead of Buy: The AI Agent Shift
McKinsey's State of AI 2026 finds nearly a third of firms skipped buying software because agentic coding let them build it. What that means for you.
Nearly a third of organizations — 32% — say they've already skipped buying at least one software product or feature because agentic coding tools let them build it internally instead. The build-versus-buy decision, stable for decades, just flipped, and it's the clearest sign yet that AI agents are changing how companies think about software.
The number behind the headline
The figure comes from McKinsey's State of AI global survey for 2026. When asked directly, 32% of respondents reported that their organization decided against purchasing one or more software products or features because they could be built internally with agentic coding tools. It's not a niche experiment — it's a third of the surveyed companies changing a procurement habit that has defined enterprise IT for decades.
The adoption trend behind it is just as striking. According to the same survey, large enterprises scaling agents in one or more functions rose from 27% to 40% in a single year, while smaller firms stayed flat at 22%. The build-versus-buy shift is concentrated where agentic coding is actually being used.
Why build-versus-buy flipped
For decades, the math was simple: building software meant hiring engineers, and buying meant a predictable subscription. Agentic coding tools change the cost side of building. When a small team can have an agent scaffold, test, and iterate on an internal tool in days, the custom option no longer carries the premium it used to.
That doesn't mean companies are building everything. It means the threshold for "just buy it" moved. A feature that's core to how a business differentiates itself — something off-the-shelf software only half-serves — is now worth building internally. A commodity tool still isn't.
| What changed | Why it matters |
|---|---|
| Build cost fell | Agents cut the time and headcount to ship internal tools |
| Buy stays predictable | Commodity software is still cheaper than custom |
| The middle moved | Features that "almost fit" are now build candidates |
| Adoption is uneven | Large firms scale agents; small firms haven't caught up |
Build vs buy: the numbers behind the shift — hover for context
What this means for you
If you work in software or buy it for a team, this is a signal to re-run your own build-versus-buy math rather than inheriting last year's decision. The question is no longer "can we build it" but "is this core enough to justify building, now that building is cheaper?"
If you're a buyer of AI tools yourself, the same logic applies to the personal agents you use. The platforms that win your business aren't necessarily the biggest — they're the ones that let you do the most with the least custom code. A local-first agent you can extend yourself, for example, follows the exact build-versus-buy logic McKinsey is describing: you own the tool instead of renting it.
The caveat worth knowing
The build-versus-buy shift is real but not a guarantee of success. Industry forecasters also project that a meaningful share of agentic projects — 40% or more, by some Gartner estimates — will be cancelled before they mature, often for unclear ROI or escalating costs. The takeaway isn't "build everything." It's that the option to build is now cheap enough to be the default question, even when the answer is still "buy."
Takeaway
A third of companies have already changed how they decide between building and buying, and the agent that tipped the scale is the same technology reshaping your own tooling. The winners won't be the ones who build everything or buy everything — they'll be the ones who re-ask the question with accurate costs.
For more on where agent adoption is heading, see our look at AI agents becoming a daily habit at work and the economics of running agents.
