If your team has ever waited on a vendor’s roadmap for a feature you needed six months ago, or watched a subscription renewal jump 20% with no new functionality to show for it, you’ve felt the frustration that’s now showing up in hard numbers. According to a global survey published in late August, nearly one in three companies has walked away from a software purchase this year because they could build the same functionality themselves with AI coding tools. That’s not a fringe behavior anymore — it’s mainstream enough that the question every business now has to ask isn’t “build or buy?” in the abstract, but “which one, for this specific need?”

The headline number, and the one that matters more

McKinsey’s 2026 State of AI survey, fielded between May and June and published August 25, polled 1,719 business leaders across 97 countries. The finding getting the most attention: 32% of organizations report deciding against buying at least one software product or feature because agentic coding tools let them build it in-house instead. Among the companies McKinsey classifies as AI “high performers,” that figure climbs to nearly half.

But the survey also contains a quieter number that deserves equal weight. The share of organizations attributing any earnings impact to AI held flat at 37% — essentially unchanged from a year earlier — even as enterprise-wide AI scaling jumped from 38% to 44%. In plain terms: companies are building more, but the group actually making money from it hasn’t grown. Doing more with AI and profiting from AI turned out to be two different projects.

Why building got easy, and why that’s not the whole story

The shift is real and easy to understand. Tools that write, test, and ship working code have matured fast enough that a dashboard or workflow automation that once took a development quarter can now come together in days. For a narrow but growing set of use cases — internal tools, reporting layers, glue code between systems — that changes the math on “just buy the SaaS seat.”

What separates the companies actually benefiting from the ones just generating more code is not the tooling — it’s what they did before they opened an AI coding assistant. McKinsey’s own analysis of its highest-performing respondents found they were far more likely to redesign a workflow first and build the software to fit the new process, rather than bolt AI onto how things already worked. Nearly three-quarters of high performers reported rebuilding a workflow from scratch because of AI, compared with about a quarter of everyone else. Declining a vendor’s quote was the last step in that sequence, not the first.

The bill that shows up later

There’s a cost side to this that’s easy to miss in the excitement of shipping something fast. A software subscription has a known, mostly fixed price. A system you build in-house trades that for a different set of costs: security patching, uptime monitoring, compliance reviews, and — quietly — the token and compute costs of running AI tools at scale, which the survey found already constrains roughly one in five organizations’ AI use. The companies building the most also reported hitting that cost ceiling most often, which is less a warning against building than a reminder to price the ongoing cost of ownership before celebrating the sticker price of “free.”

There’s also a durability question. A feature you build in six weeks with a coding agent still needs someone who understands it eighteen months from now, when the engineer who wrote it has moved on or the underlying AI model has changed. A license fee doesn’t disappear when you stop buying it — it becomes a maintenance line item with your name on it.

What this means if you’re weighing a build vs. buy decision

If the software touches something generic to your business — payroll, accounting, identity, payments: ask a prospective partner why buying still wins here. The right answer involves compliance coverage, uptime guarantees, and the thousand edge cases a mature vendor has already handled, not just feature lists.

If the software would shape a process that’s actually unique to how you operate: ask whether the workflow gets redesigned before anything gets built, or whether the plan is to automate the current process as-is. The research above suggests the second approach rarely produces a financial return worth the effort.

If you’re being pitched a custom build: ask who owns it in year two, and what the realistic ongoing cost looks like once support, security review, and model or infrastructure changes are priced in — not just the cost of the initial sprint.

The takeaway

The capability to build instead of buy is now widely available, and for the right use cases, that’s genuinely good news for businesses tired of paying for software that doesn’t quite fit. But the data is equally clear that capability alone doesn’t produce a return. The companies pulling ahead are the ones treating the build-or-buy question as a strategic decision made after rethinking how the work should actually happen — not a shortcut taken because the tooling made it possible.

Kode Vox helps businesses sort exactly this kind of decision: which systems are worth building around your own workflow, which are better left to a mature vendor, and how to price the real cost of ownership either way. If you’re weighing a build vs. buy decision and want a second opinion before you commit, email us at info@kodevox.com or reach out through our contact page.

Sources and further reading:

• McKinsey & Company — The state of AI in 2026: On the road to ROI (25 August 2026)
• Digital Applied — A Third of Companies Skipped Buying Software and Built It (1 September 2026)
• Under30CEO — Build vs Buy Flips as AI Lets Companies Ship In House (4 September 2026)

— The Kode Vox Team