What happened
On August 25, 2026, McKinsey's QuantumBlack division published its latest Global Survey on the state of AI, drawing on responses from 1,719 people across 97 countries, surveyed between May 4 and June 8, 2026. Among the findings: 32% of respondents at organizations that regularly use AI said their company had decided against buying at least one software product or feature because it could be built internally using agentic coding tools, the AI systems that can plan, write, test, and revise code with limited human intervention. The share was highest in technology (41%), followed by healthcare (39%), professional services and energy and materials (38% each), and financial services (36%). The same survey put this next to a less flattering number. Only 37% of respondents said AI had made any positive contribution to their organization's earnings before interest and taxes, a figure multiple outlets described as essentially unchanged from the prior year despite a year of heavier AI investment. Eighty percent said AI had improved their own personal productivity, so the gap between what individuals feel AI is doing for them and what companies can show on a balance sheet remains wide. Cost pressure is part of the picture too: one in five respondents said AI-related operating costs, including the cost of processing tokens, are already limiting how much their organization uses AI, even as 28% report spending more than 10% of their total technology budget on AI and expect that share to grow. McKinsey senior partner Lieven Van der Veken, quoted in coverage of the report, said the shift does not mean companies should build everything themselves or stop working with outside vendors. He described the more common pattern as leaders becoming more deliberate about what to build, what to buy, and what capability to develop in house. QuantumBlack senior fellow Michael Chui added that organizations are treating operating costs as a design constraint rather than an afterthought as they decide where AI coding agents fit into that mix.
Why it matters for business owners
Most small and medium businesses have never had the option to build their own internal software at a reasonable cost. That used to require a development team most companies could not justify hiring, so buying an off-the-shelf tool or paying an agency to build something custom were the only realistic paths. AI coding agents genuinely change that math for a specific slice of work: narrow, well-defined internal tools where the business already understands exactly what it needs. That is a real shift, and it is worth taking seriously rather than dismissing as hype. But the McKinsey data does not say building has become free, easy, or safe by default. It says a third of AI-using companies chose to build instead of buy for at least one product or feature, out of many software decisions they made during the year. It is a new option on the table, not a new default answer.
What owners should not misunderstand
This is not evidence that most companies are abandoning purchased software. The 32% figure describes organizations that skipped buying at least one product or feature, not organizations that have stopped buying software altogether or that build the majority of what they use. McKinsey's own commentary on the survey is explicit that a mixed model, buying core platforms while building specific, differentiated pieces around them, is the more likely outcome than a wholesale shift to in-house development. It is also not evidence that building is cheaper once you account for the full cost. The survey measured a purchasing decision, not a total cost comparison between the software an organization chose not to buy and the cost of building, securing, and maintaining what replaced it over its full life. A tool built quickly by an AI coding agent still needs someone accountable for its security, its behavior when something breaks, and its upkeep as the business changes, the same way any other piece of internal software does. None of that goes away because AI wrote the first version faster. And this is not proof that AI spending is paying for itself. The same survey that found rising build-instead-of-buy activity also found that only 37% of organizations can attribute any earnings impact to their AI investment, a number that has not moved much even as spending keeps climbing. Treat the build-versus-buy shift and the ROI picture as two separate findings from the same report, not one story where more building automatically means a better return.
The operational lesson
The useful question is not whether to build or to buy in general. It is which specific pieces of software are actually worth building in house, and McKinsey's own guidance points to a reasonable test: build what is genuinely specific to how your business runs and gives you an edge competitors do not have, and buy or use existing tools for anything that is not a meaningful differentiator, no matter how easy AI coding agents make it to build a version yourself. A narrow internal tool, an approval workflow, a simple dashboard that pulls from systems you already use, is a reasonable candidate for an AI-assisted internal build, because the requirements are well understood and the cost of getting it slightly wrong is low. A system that touches customer data, financial records, or a core part of how the business operates is a different calculation entirely, because the cost of a security gap, a bug, or an unmaintained dependency is much higher than the subscription fee it might have replaced. The other lesson sits in the ROI gap itself. High performers in the survey, the organizations most likely to report a real financial return from AI, were also the ones most likely to have decided against a software purchase in favor of building, at roughly one and a half times the rate of other respondents. That correlation does not prove building caused the better return. It is more consistent with a simpler explanation: businesses that are already disciplined about where AI creates real value make more disciplined build-versus-buy decisions too, rather than building because they can.
What a serious business should do next
Before treating an AI coding agent as a reason to skip a software purchase, ask who will own the result after it ships. A vendor's product comes with support, security patching, and someone whose job depends on it working. An internally built tool needs the same things assigned to a specific person or team inside your business, not left to whoever happened to prompt the agent that built it. Separate the software decisions in front of you into two piles: things that are genuinely specific to how your business operates, where a custom internal tool could be a real advantage, and things that are common problems every business in your position already has a solved, supported answer for. AI coding agents make the first pile more achievable. They do not make the second pile worth reinventing. If your team has already built something internally with an AI coding agent, check who is responsible for it today, not just who built it. A tool without a named owner for security, maintenance, and what happens when the person who built it leaves is a liability whether AI wrote it in an afternoon or a contractor wrote it over a month. Do not use this survey as a reason to freeze software purchases you already know you need while you wait to see if AI can build a version instead. The 32% figure describes a shift in what companies consider, not a signal that buying software is now the wrong move by default.
The Atlacis view
This survey captures something real: AI coding agents have expanded what a business can reasonably build for itself, and that is a genuine change in the build-versus-buy calculation many owners have never had to make before. The part worth sitting with is the other number in the same report. Most organizations still cannot show a financial return on their AI spending, and a new, easier way to build software does not fix that on its own if the decision to build is not held to the same discipline as any other technology investment. Atlacis helps business owners work through exactly this kind of decision before money and time are committed: which pieces of software are genuinely worth building around your own workflow, which are commodity problems better solved by buying, and who inside your business will actually own what gets built once an AI coding agent has produced the first version.
The short version
- McKinsey's 2026 State of AI survey (1,719 respondents, 97 countries, May 4 to June 8, 2026) found that 32% of AI-using organizations decided against buying at least one software product or feature because they could build it internally with AI coding agents, highest in technology (41%), healthcare (39%), and professional services and energy and materials (38% each).
- The same survey found only 37% of organizations attribute any positive earnings impact to their AI spending, a figure described as roughly flat versus the prior year despite rising AI investment, even as 80% report improved individual productivity.
- The 32% figure describes organizations that skipped one purchase, not organizations abandoning purchased software altogether. McKinsey's own guidance points to a mixed model: buy core platforms, build what is genuinely specific to your business.
- An AI-built internal tool still needs an owner for its security, maintenance, and upkeep after it ships. None of that cost disappears because an AI coding agent wrote the first version quickly.
- High performers were about 1.5 times more likely to report building instead of buying, which is more consistent with general AI discipline than with building itself causing better returns.
- The decision worth making is which specific pieces of software are worth an in-house AI-assisted build versus which are commodity problems already solved by existing tools, not a blanket choice to build or to buy.
Where ATLACIS can help
Sources
- McKinsey & Company: The state of AI in 2026: On the road to ROI (August 25, 2026)
- CIO Dive: Enterprises bet on agents to build in-house software, boost productivity (Paige Gross, August 27, 2026)
- The Times of India: One in three AI-using firms forgo some software buys: McKinsey (Supriya Roy, August 27-28, 2026)
- The Register: McKinsey says enterprise AI is finally 'on the road to ROI' (August 25, 2026)