Build vs Buy Flips as AI Lets Companies Ship In House

Roughly a third of organizations have turned down at least one software purchase and written the thing themselves with agentic coding tools, according to McKinsey’s latest global survey of business leaders. The build vs buy question, settled for most of the past decade in favor of buying, is open again.

That matters to anyone selling software. Your buyer now has a credible third option beside your product and your competitor’s, and it costs a sprint rather than a subscription.

What the Survey Measured

McKinsey fielded the study between May 4 and June 8, 2026, and published it on August 25. The sample covered 1,719 business leaders across 97 nations, so this is not a niche read of one industry.

The headline figure is 32%. That share of organizations decided against buying an off-the-shelf product or feature because coding agents made an internal version viable.

Adoption of those tools is still uneven. About two in ten organizations are scaling agentic coding tools overall, while the figure reaches 31% among larger enterprises.

The Number Under the Number

Behind the build vs buy shift sits a less flattering statistic. The share of companies attributing any EBIT impact to AI held flat at 37%, and only 6% clear McKinsey’s high-performer bar of at least 5% of EBIT from AI.

So individual teams are shipping more, yet the profit line has not caught up. Read that as a warning about maintenance rather than proof the trend is hollow.

Who Is Actually Doing This

The behavior clusters at the top. Nearly half of the high performers skipped a software purchase, compared with 31% of everyone else, per the full McKinsey report.

That pattern is worth noting because high performers usually set the template. Their vendors feel the change first, and everyone else follows a year or two later.

Meanwhile the tooling keeps improving. Anyone tracking AI coding tools has watched the gap between prototype and production shrink month after month.

Smaller companies are further behind, though not by much. The tools are priced per seat rather than per enterprise, so the gap is one of confidence and engineering time rather than access.

The Risk Sitting Behind a Cheaper Stack

Building internally moves cost rather than deleting it. A licence fee becomes a run cost: on-call rotations, security patches, compliance reviews and the engineer who understands the code leaving in eighteen months.

There is a second trap. Teams that build on a single model provider have simply swapped one dependency for another, which is the same vendor lock in problem in different clothing.

Ask what happens when the model changes, the price moves or the API is withdrawn. If the answer is a rewrite, the internal build was never cheaper.

What Software Founders Should Change This Quarter

Start by naming the run cost out loud in your sales conversations. Buyers comparing a subscription against a weekend of agent output are comparing the wrong numbers, so put maintenance, uptime and audit trails on the same slide.

Next, examine which parts of your product are genuinely hard to replicate. Workflow wrappers are exposed. Proprietary data, regulated integrations, network effects and hard-won distribution are not.

Finally, look at where buyers still refuse to build. Niche operational software keeps winning, which is one reason a company like Rundoo can raise on small business technology that larger vendors ignore.

Then reprice around outcomes rather than seats. A per-seat licence invites a build comparison, whereas pricing tied to a result the buyer cannot easily reproduce does not.

Reading the Next Six Months

Watch three signals. Renewal rates on generic tooling will tell you whether the 32% figure keeps climbing, and procurement teams tend to move at renewal rather than mid-contract.

Watch the EBIT number too. If it stays near 37% into next year, expect a wave of companies quietly buying back the software they built and could not maintain.

Watch hiring, finally. Internal builds need owners, so a rise in platform engineering roles would confirm the shift is real rather than experimental.

Build vs Buy Questions Founders Ask

Does this apply to small companies? Partly. The behavior concentrates in larger enterprises, though the tooling is available to anyone with an engineer.

Which categories are most exposed? Thin workflow layers over a database face the most pressure, while regulated and data-heavy products hold up better.

Is buying software finished? No. The survey shows a majority still bought normally, so this is a change in negotiating power rather than the end of a market.

The short version for founders: your product now competes with your customer’s engineering team, so price and pitch accordingly.