The short answer
Nearly a third of organizations — 32% — have skipped buying at least one software product or feature because they could build it internally with AI coding agents, according to McKinsey's The State of AI in 2026 report published August 25, 2026. The shift is sharpest among high performers, where nearly half declined a purchase. But the same coverage flags a hard truth: internally built systems have historically succeeded only about a third of the time versus roughly two thirds for vendor software. Agentic coding tools cut the cost of the first version — not the cost of owning it. The teams that win this shift are the ones that pair AI speed with real custom software engineering discipline.
What the survey actually says
McKinsey's annual State of AI survey, subtitled On the road to ROI this year, was fielded from May 4 to June 8, 2026 and drew 1,719 responses across 97 countries. The headline finding that caught the industry's attention: 32% of respondents said their organization decided against buying one or more software products or features specifically because they could build the capability internally with agentic coding tools. That is a structural change in the classic build-versus-buy calculus, and it is being driven by tooling that did not exist at scale a year ago.
The appetite is concentrated among the winners. The roughly 6% of respondents that McKinsey classifies as high performers — organizations attributing at least 5% of their EBIT to AI — are the most aggressive, with nearly half skipping a purchase compared with 31% of everyone else. By industry, technology leads at 41%, followed by healthcare payers and providers at 39%, and professional services and energy tied at 38%. Large enterprises are leaning in too: 40% of companies with over $1 billion in revenue are now scaling AI agents in one or more functions, up sharply from 27% a year earlier.
One number that did not move is worth holding onto. The share of organizations reporting any bottom-line profit impact from AI stayed flat at 37% year over year. Adoption is accelerating; measured returns are not yet keeping pace. That gap is the whole story behind the build-versus-buy question, and it is why a decision to build in-house needs the same rigor you would apply to hiring a dedicated engineering team — because, functionally, that is what owning software is.
Why the sticker price is the small number
The purchase price of a SaaS product is visible, negotiated, and easy to point at in a budget review. That is exactly why it is a misleading anchor. When you decline to buy and build instead, you do not eliminate cost — you convert a predictable subscription line into an open-ended run cost your own team now carries: maintenance, security patching, dependency upgrades, uptime and on-call, infrastructure, and the continuous engineering time to keep the thing alive as requirements and platforms shift underneath it.
The historical base rate is sobering. Coverage of the McKinsey findings noted that internally built systems have succeeded only about a third of the time, against roughly two thirds for vendor solutions — and Gartner has projected that more than 40% of agentic AI projects will be cancelled by the end of 2027, largely over escalating costs and unclear business value. Roughly 20% of organizations already report feeling the pinch of AI operating costs. As McKinsey senior partner Lieven Van der Veken framed the underlying shift, leaders are increasingly asking what their organizations need in order to build AI tools themselves — but wanting to build and being able to run are two different capabilities.
The practical lesson is that operating cost has to be a design constraint from day one, not a surprise in year two. An agent can generate a working prototype in an afternoon; it cannot, on its own, make that prototype observable, secure, compliant, and cheap to keep running. Those are engineering decisions, and they are where the real total cost of ownership is decided.
When building beats buying
None of this argues against building. It argues for building deliberately. The cases where an in-house build with AI coding agents genuinely wins share a few traits: the capability is a real competitive differentiator rather than a commodity; the requirements are specific enough to your business that no off-the-shelf product fits without heavy customization; the data or workflow is sensitive enough that keeping it inside your own systems has strategic value; and you have — or can assemble — the engineering discipline to own the result for years, not just ship a demo.
Buying still wins for the large surface area of commodity capability: authentication, billing, analytics, ticketing, generic CRM. A mature vendor has already absorbed the edge cases, the compliance certifications, and the maintenance burden, and no amount of agent velocity makes rebuilding that a good use of senior engineering time. The most durable strategy in 2026 is not "build everything now that we can" but a portfolio: buy the commodity, build the differentiator, and be honest about which is which.
What it means for US & EU dev teams
For engineering leaders in the US and EU, the McKinsey data reframes a familiar decision rather than settling it. The first implication is that build-versus-buy is now a live question for far more of your stack than it was a year ago, because agentic tools have collapsed the cost of the first working version. That is a genuine opportunity — and a genuine way to accumulate unmaintained internal software if the decisions are made on enthusiasm rather than economics.
The second implication is that the constraint has moved. When an AI agent can produce a plausible implementation quickly, the scarce resource is no longer people who can write code — it is people who can review it, secure it, and take responsibility for it in production. Teams that scaled agent usage without scaling senior review are the ones most likely to show up in Gartner's cancelled-project statistic. For regulated sectors — FinTech, HealthTech, anything under SOC 2, ISO 27001, or the EU's NIS2 and GDPR regimes — a fast internal build that skips security and compliance review is not a saving; it is a deferred liability.
The third implication is about staffing. The organizations getting real EBIT impact from AI are not the ones that replaced engineers with agents; they are the ones that put experienced engineers in charge of agents and treated operating cost as a first-class design goal. That is a product-engineering competence, and it is exactly the gap most in-house teams underestimate when they decide to build instead of buy.
A build-vs-buy checklist for 2026
- Name the differentiator. If the capability is not something customers would notice or pay for, default to buying. Reserve building for what actually sets you apart.
- Price the run, not the build. Estimate three years of maintenance, security, infrastructure, and on-call before you compare against a subscription. The build cost is the small number.
- Check the base rate. Assume an in-house build is more likely to stall than a vendor rollout, and ask what specifically makes yours the exception.
- Staff for review, not just generation. Every AI-generated component needs a senior engineer who can read, test, and stand behind it. Budget that review time explicitly.
- Make compliance a gate, not a phase. For regulated data, security and privacy review belong in the design, not after launch. A fast build that fails an audit is slower than buying.
- Set a kill criterion. Decide up front what "not working" looks like and when you would switch to a vendor. Sunk-cost momentum is what turns a build into a cancelled project.
- Revisit yearly. A capability worth building today may become a commodity a vendor solves better next year. Treat the portfolio as something you rebalance, not set once.
Frequently asked questions
What did McKinsey's State of AI 2026 survey find about build vs buy?
In McKinsey's The State of AI in 2026 report, published August 25, 2026, 32% of respondents said their organization decided against buying at least one software product or feature because they could build it internally with agentic coding tools. The survey drew 1,719 responses across 97 countries and was fielded from May 4 to June 8, 2026. The trend was strongest among high performers — the roughly 6% of respondents that attribute at least 5% of EBIT to AI — where nearly half skipped a software purchase versus 31% of their peers.
Does building software in-house with AI agents actually save money?
Not automatically. The purchase price you avoid is only the sticker cost. Internally built software carries the full run cost: maintenance, security patching, on-call, infrastructure, and the engineering time to keep it current. Coverage of the McKinsey data noted that internally built systems succeed roughly a third of the time compared with about two thirds for vendor solutions, and Gartner has projected that more than 40% of agentic AI projects will be cancelled by the end of 2027 over cost and unclear value. AI agents lower the cost of the first version, not the cost of owning it.
Which industries are building instead of buying the most?
According to the McKinsey survey breakdown, the technology sector led at 41% of respondents deciding against a purchase, followed by healthcare payers and providers at 39%, and professional services and energy at 38%. Large enterprises with more than $1 billion in revenue were also scaling agents, with 40% now running them in one or more functions, up from 27% the previous year.
When should a team build with AI coding agents versus buy?
Build when the capability is a genuine differentiator, the requirements are specific to your business, and you have the engineering discipline to own the result long term. Buy when the capability is a commodity, a mature vendor already solves it, and your team's time is better spent elsewhere. Treat operating cost as a design constraint from day one rather than an afterthought, and staff the build with senior engineers who can review agent output — because the bottleneck has moved from writing code to reviewing and maintaining it safely.
Sources
McKinsey & Company — The State of AI in 2026: On the road to ROI (published August 25, 2026)
Forkast / Yahoo Finance — The Build-vs-Buy Shift: 32% of Enterprises Bet on Agentic Coding Tools (September 1, 2026)