A third of companies have already skipped buying a piece of software because they decided to build the same solution in-house using agentic coding tools. That's what McKinsey's global "State of AI 2026" survey found this week — and the number comes with a caveat that changes how it should be read.
What Changed
According to McKinsey, 32% of organizations have already skipped purchasing at least one software product or feature because they could build the alternative in-house with AI coding agents. The share varies widely by sector: 41% in technology, down to 19% in insurance and 17% in the public sector. The pattern is strongest among the best-performing companies on AI — the 6% of respondents who attribute at least 5% of EBIT to AI use — nearly half of them skip software purchases in favor of building, versus 31% of everyone else. Overall, about 20% of organizations are already scaling coding agents past the pilot stage, rising to 31% among large enterprises (revenue above $1 billion), which also raised the share scaling AI agents in at least one function from 27% to 40% in a year.
Why It Matters
Here's the caveat that changes everything: despite the acceleration in adoption and in "doing it yourself," the share of organizations attributing any measurable EBIT impact to AI held flat at 37% — and only 6% clear the high-performer bar (5%+ of EBIT). In other words: building with a coding agent got easier and faster, but for most companies that isn't translating into proven financial return. The decision to "build instead of buy" is being made, in a good number of cases, based on technical feasibility — because it's now possible — rather than on proof that it delivers a better return than an off-the-shelf solution.
The Impact for Brazil
For Brazilian companies weighing building in-house instead of buying ready-made software — an increasingly viable choice with coding agents — McKinsey's data works as a practical warning: the technical ease of "doing it yourself" isn't, on its own, a business justification. It's worth applying the same ROI rigor to a build decision that you'd apply to any software purchase: total cost of maintaining what gets built, internal team time consumed, and an honest comparison against what a mature market solution already delivers. Going by the numbers, that rigor is mostly present at companies that are already "high performers" on AI — it isn't automatic for those just starting to scale.
Entercast's Take
This data speaks directly to the theme that has anchored this blog from the start: the gap between a pilot and scale with real results. McKinsey is showing, at global scale, exactly the pattern we've seen in specific cases here — from AT&T cutting cost through model routing to NVIDIA's low-cost agent framework: a better tool isn't automatically the same thing as a better result. Before deciding to build in-house just because it's now technically possible, it's worth measuring — with the same rigor as any investment decision — whether it actually moves the company's bottom line, or just looks cheaper at first glance.