McKinsey published the 2026 edition of its State of AI survey on August 25, 2026, and one finding travelled faster than the rest. “Nearly a third of respondents (32 percent) report that their organizations have decided against buying one or more software products or features because they could be built internally with agentic coding tools.” The survey drew 1,719 responses across 97 countries between May 4 and June 8, weighted by each country’s share of global GDP. The report gives no prior-year figure for it, so there is no trend line yet. There is, however, a companion finding that gives the 32% its meaning.
The share of organisations attributing any impact on earnings before interest and tax to AI is 37%, which McKinsey describes as “essentially unchanged from a year ago.” The share of high performers, organisations that credit at least 5% of EBIT to AI and call the impact significant, is flat at about 6%. So the survey describes a year in which AI became capable enough to replace purchased software, and in which the number of companies making money from it did not grow. Both things are true, and the useful question for a business is what separates the companies in the first group from the second.
- 01The build-instead-of-buy decision is now mainstream.32% of all respondents, and nearly half of AI high performers versus 31% of everyone else, declined at least one software purchase because coding agents could produce the functionality in-house. Most common in technology and healthcare, then professional services and energy.
- 02Profit impact did not move.37% attribute any EBIT impact to AI, the same as 2025, and high performers stay at 6%, even as 44% now report AI scaling across the enterprise, up from 38%. Individual productivity gains, reported by 80%, have not turned into enterprise returns for most.
- 03High performers redesign the workflow, not just the tooling.McKinsey’s account of the 6%: they pursue growth or innovation alongside efficiency, “fundamentally redesign workflows that are enabled by AI rather than insert AI into existing ones,” and are twice as likely to have leaders visibly committed.
- 04Cost is the constraint arriving next.One in five organisations say AI operating costs, including tokens, already limit their use. High performers report cost constraints on coding agents about three times as often as others, because they use them most.
01 — The findingOne question, and what the unit is.
The survey’s own framing is careful. The full sentence in the report reads: “Coding agents are emboldening companies to build their own software rather than purchase it. Nearly one-third of respondents (32 percent) report that their organizations have decided against purchasing at least one software product or feature because they were able to build the functionality in-house using agentic coding tools.” Note the unit. It is one product or feature, at least once. A company that built a single reporting dashboard instead of buying a seat licence counts the same as one that replaced its customer platform. McKinsey adds that this “could be a sign that AI is beginning to reshape how technology budgets are allocated,” which is about as far as the evidence goes. The table below puts the number in the context of the survey’s adoption figures, all from the published report.
| Finding | 2026 | 2025 | Note |
|---|---|---|---|
| Declined a software purchase because agents could build it | 32% | no prior figure given | Nearly half of high performers vs 31% of others; most common in technology and healthcare |
| Scaling software coding agents | about 2 in 10 | — | 31% at large enterprises; high performers twice as likely |
| Large organisations scaling AI agents | 40% | 27% | Smaller organisations flat at 22% |
| AI scaling across the enterprise | 44% | 38% | 54% of large organisations vs one-third of smaller ones |
| Attribute any EBIT impact to AI | 37% | about 37% | “Essentially unchanged” |
| AI high performers (≥5% of EBIT, “significant”) | about 6% | about 6% | Flat |
| AI operating costs constrained use | about 20% | — | Consistent across sizes and industries; 60% still plan to invest more |
| Expect AI-related workforce decline next year | 39% | 32% | Only 14% report an actual decline over the past year |
02 — The tensionBuilding rose. Benefiting did not.
Read the table top to bottom and a pattern appears that the headline hides. Every measure of doing more with AI went up: enterprise-wide scaling, agent deployment at large companies, functions covered, budget share. The one measure of getting paid for it stayed still. McKinsey’s own summary is direct about this: “Organizations’ conviction in AI is growing faster than the immediate financial returns they can attribute to it.” Eighty percent of respondents say AI improved their individual productivity and half say it helps them make better decisions, and those gains, in the report’s words, “have yet to translate into broad financial impact for organizations.”
The build decision sits inside that gap. Declining to buy a product because you can build it is an act of capability, not of return. Whether it pays depends on what happens next: whether the thing you built replaced a real cost, whether it changed how work is done, and whether anyone maintains it in year two. The survey cannot see any of that, and it does not claim to. What it can see is that the group most likely to build, the high performers, is also the group most likely to report the other behaviours that the report associates with returns. That correlation is the interesting part, and it runs through workflow design rather than through tooling.
McKinsey’s question counts a decision, not an outcome. A team that builds a feature in-house with a coding agent takes on the maintenance, the security review and the migration burden that the vendor was pricing into the licence. The survey offers one hint about where that lands: high performers report being constrained by costs in their use of coding agents about three times as often as others. The companies building the most are the ones already paying attention to what building costs.
03 — The 6%What the companies making money actually do.
McKinsey defines high performers as respondents who attribute at least 5% of EBIT to AI and describe its impact as significant. They are 6% of the sample, the same as last year, and the report devotes its longest section to how they differ. Four differences matter for the build-versus-buy question specifically.
They redesign the work
The report’s central claim: high performers “fundamentally redesign workflows that are enabled by AI rather than insert AI into existing ones.” Building a tool to fit an unchanged process is the pattern the 37% describes; changing the process and then building for it is the pattern the 6% describes.
They build more, and buy less
High performers are twice as likely as others to be scaling coding agents and 2.7 times more likely to be scaling other agentic AI. Nearly half report deciding against a software purchase because they could build it, against 31% of other respondents.
They aim past efficiency
Around 80% of both groups pursue efficiency. Most high performers also use AI to pursue growth or innovation, and they are 3.3 times more likely to intend to fundamentally transform their business within three years. A tool built to cut a cost has a ceiling; one built to sell something new does not.
They pay for it, and manage the risk
High performers are more than twice as likely to spend over 15% of their technology budget on AI, twice as likely to report visibly committed senior leaders and defined measurement processes, and much more likely to be working on AI-driven vulnerabilities and unintended actions.
None of those four is a technology choice. The survey is about a group of companies that changed how they operate and then found that coding agents let them build what the new operation needed. The reverse order, buying the agent first and looking for something to build, is what produces a feature that was cheaper than the vendor’s and changes nothing. Our earlier case for custom tools over branded software made the same argument from cost; McKinsey’s data makes it from outcomes.
04 — ConstraintThe cost that is starting to bite.
About one in five respondents say AI-related operating costs, which McKinsey specifies as including token costs, have constrained their organisation’s use of AI. The report calls this “a meaningful consideration, but not yet a widespread constraint,” and the share is broadly consistent across company sizes and industries. For each of the three tool types the survey asks about, chatbots, agents and coding agents, about one in ten say cost has limited use. At the same time, 28% of organisations spend more than a tenth of their technology budget on AI and 60% expect to increase AI investment next year.
The build-versus-buy reader should hold two of those facts together. A purchased product has a price that is known in advance and mostly fixed. A built one has a token bill that scales with use, and the survey shows the heaviest builders are the ones already feeling it. That does not argue against building. It argues for pricing the run cost of a built feature the way a vendor would have, before the decision rather than after. The arithmetic for doing that on the current model price sheets is in our maintained per-million-token price index.
05 — The decisionWhich purchases to decline, and which to keep paying for.
The survey does not say which products the 32% declined, only that technology and healthcare organisations declined most often, followed by professional services and energy. The high-performer findings do suggest a rule for deciding, and it is the one we apply in our own work: build where the workflow is yours and the software would have shaped it; buy where the workflow is generic and the vendor’s scale is the value.
The last row is the one the survey’s optimism can obscure. A coding agent makes the build fast; it does nothing for the adoption. Our guide to replacing an internal system safely covers the read-only rollout pattern that turns a built replacement into a used one. And for the marketing function in particular, where “features” are often reports and automations rather than systems, the marketing tasks worth handing to a coding agent lists the ones that pay back first.
06 — ConclusionThe capability is real. The return is a separate decision.
A third of companies can now build what they used to buy. The six percent making money from AI changed the work first, then built for it.
The 32% figure will be quoted for a year as evidence that software procurement is changing, and it is. It should be quoted next to the 37%, which says that for most organisations the change has not yet reached the income statement. McKinsey’s reading is that conviction is running ahead of returns. Ours is that building has become the easy half of the decision.
The companies the survey holds up as the model did not start by asking what they could build. They redesigned a workflow, set a growth goal rather than only a cost goal, committed budget and leadership to it, and then found that coding agents let them make exactly what that redesign required. Declining a vendor quote is the last step in that sequence. Taken first, it is a cheaper way to change nothing.