McKinsey surveyed 1,719 participants across 97 countries between May 4 and June 8, then published the results on August 25. The headline number that spread immediately: 32% of organizations have decided against buying off-the-shelf software, opting instead to build their own solutions using agentic coding tools. In tech specifically, the share is 41%. Read that alongside a second number and it gets interesting: 80% of respondents report individual productivity gains, but only 37% report any enterprise EBIT impact - flat versus last year's survey. That gap - widespread speed at the keyboard, unmoved bottom line - is the real story inside the data.
What the agentic coding build-vs-buy shift actually measures
The 32% figure counts organisations that declined at least one purchase - not ones that have replaced their entire software stack. The industry spread tells you something: insurance and the public sector sit at 19% and 17%, because those sectors buy audit evidence and liability, not code.
Healthcare payers and providers sit at 39%, professional services and energy at 38%, and financial institutions at 36%.
The effect is also concentrated in the companies already winning. McKinsey identifies "high performers" as the 6% of respondents who attribute at least 5% of their EBIT to AI. Nearly half of those high performers are skipping software purchases, compared to 31% of their peers. In other words, the build-vs-buy shift is driven by companies that already know how to extract profit from AI - not the broad middle.
A separate data point reinforces the direction. Retool's 2026 Build vs. Buy Report, based on a survey of over 800 professionals, found that 35% of enterprises have already replaced a SaaS tool with something they built themselves.
Teams are replacing SaaS with custom software, and many are doing it outside traditional IT oversight, as development outpaces procurement and governance processes.
Why the productivity-to-profit gap is the more important number
As organisations expand deployment, AI operating costs - including tokens - are becoming a meaningful consideration. One in five respondents says their organisation is limiting AI use because of operating costs.
That constraint sits right in the middle of the productivity-to-profit gap. A team that builds an internal tool with a coding agent runs that tool in production. It now pays inference costs every time the tool runs. It also owns the security surface, the upgrade cycle, and the incident response when something breaks. None of that shows up in the "we didn't renew our SaaS subscription" line.
The pattern has a name in software economics: the make-vs-buy decision has always understated total cost of ownership for internal builds. Agentic coding tools lower the cost to build from days to hours. They do not lower the cost to run, maintain, or secure what you built.
Large enterprises are scaling agents; smaller ones are not
Agent adoption is splitting along size lines in a way the headline numbers obscure. Use of agentic AI is increasing, mostly accounted for by large enterprises. Forty percent of respondents from large organisations (those with annual revenues of more than $1 billion) report scaling AI agents, up from 27% last year. The share of respondents from smaller organisations reporting scaling remained flat at 22%.
That flat 22% matters. Small and mid-sized teams have less tolerance for the maintenance tail that follows an internal build. A 12-person startup that uses a coding agent to build its own analytics dashboard instead of paying for a SaaS tool has committed its engineers to keeping that dashboard alive, debugging it when the underlying model API changes, and explaining to new hires why the system works the way it does.
| Large enterprise (>$1B revenue) | Smaller organisation | |
|---|---|---|
| Scaling AI agents | 40% (up from 27%) | 22% (flat) |
| Skipping a software purchase | Concentrated in high performers | Less common |
| Operating cost pressure | 1 in 5 limiting AI use | Similar |
| Build maintenance risk | Absorbed by larger eng teams | Falls on a small core |
The risk is asymmetric. Large teams can absorb a failed internal build. A small team that rebuilds its support ticketing system in-house and then loses its one expert on that codebase has a different problem.
Beagle in action#eng-ops, Thursday morningWhat this means for teams deciding right now
The agentic coding build-vs-buy calculation is real and the shift is durable. The McKinsey data is not hype. But the right frame is not "build is cheaper" - it is "build is faster to start and slower to own."
A few questions worth answering before skipping the next renewal:
- Who owns the production system? An internal build needs an owner who is on call for it. Name that person before you cancel the SaaS contract.
- What does the inference bill look like at full usage? Run the token math on a realistic monthly workload before the build, not after it's live.
- Is the workflow stable? Agentic coding tools accelerate initial builds dramatically. They do not help if the underlying process changes every quarter and someone has to rebuild the tool each time.
- What's the security surface? SaaS vendors absorb CVEs, supply chain patches, and SOC 2 audits. Internal builds hand those back to your team.
A teammate like Beagle can draft the structured comparison - pulling in the actual eng hours from your project tracker and the SaaS invoice from your finance channel - so the build-vs-buy conversation happens on real numbers rather than intuition.
Agentic coding build vs buy: common questions
What did McKinsey's 2026 State of AI survey actually find on build vs buy?
McKinsey published the 2026 edition of its State of AI survey on August 25, 2026. It found that nearly a third of respondents - 32% - report their organisations 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.
Is the build-vs-buy shift creating real business impact?
Not yet, broadly. The share of respondents reporting individual productivity gains is 80%, but only 37% report any enterprise EBIT impact - flat versus last year's survey. The productivity benefit is real and personal; the profit impact is lagging. The gap likely reflects attribution delays, operating cost offsets, and maintenance load that eats back the savings.
Which industries are most likely to build instead of buy?
McKinsey's State of AI 2026 survey found that 32% of organisations, and 41% in tech, have stopped buying software and are building it in-house with AI coding agents.
Healthcare payers and providers sit at 39%.
Insurance and the public sector sit at 19% and 17% - sectors that buy audit evidence and liability, not code.
Does building with a coding agent save money over SaaS?
It saves on the license. It does not save on inference costs, maintenance, security patching, or the engineering time needed to own the system in production. One in five organisations is already limiting AI use because of operating costs , which suggests the token and compute bill is not negligible. Run the full TCO before cancelling a renewal.
Are smaller teams experiencing the same build-vs-buy shift?
No. Forty percent of large organisations (over $1 billion in revenue) report scaling AI agents, up from 27% last year, while the share from smaller organisations remained flat at 22%. The build shift is driven by large enterprises with engineering capacity to absorb the maintenance tail that internal builds create.