32% of Companies Just Skipped a Software Purchase to Build It With AI. What It Means If You Sell SaaS.

This week the SaaS industry got the number it had been quietly dreading. In McKinsey's State of AI 2026 report, released August 25, 32% of organizations said they decided against buying at least one software product or feature — because they could just build it in-house with agentic coding tools. If you sell software, that sentence is either an obituary or an opportunity, depending on what you sell. Let's figure out which.

McKinsey surveyed 1,719 people across 97 countries between May and June 2026. The headline finding: nearly a third of organizations have already passed on a software purchase they would have made a year ago, because a coding agent can now stand up a workable version internally. Break it down by industry and it gets sharper — in technology, the figure is 41%.

And this isn't one survey with a spicy stat. Retool's 2026 Build vs. Buy report found that 35% of enterprises have already replaced a SaaS product with custom software. Two different methodologies, same trajectory. The build-vs-buy line that has sat in roughly the same place for two decades is moving, and it's moving toward build.

The most uncomfortable detail is who is doing the building. McKinsey's “high performers” — the ~6% of companies that attribute real EBIT to AI — skip software purchases at nearly half, versus 31% of everyone else. The customers most fluent in AI, the ones with budget and the ones you most want on your logo wall, are the fastest to build instead of buy.

The reflex read is “SaaS is dead,” and half the internet has already written that post. It's wrong — but it's wrong in an instructive way, so it's worth taking seriously.

Here's what the doomers get right: if your product is a thin layer over one API, a basic CRUD app, a form builder, or a dashboard over data the customer already owns, an agent can now clone it in an afternoon. The floor just fell out of that market. When building costs almost nothing, “we built the same thing” stops being a moat and becomes table stakes.

Here's what they get wrong: building the thing was never the expensive part. Keeping it alive is. A team that builds a feature in-house with a coding agent inherits the maintenance, the security review, the migrations, and the on-call — every line the vendor was quietly pricing into the licence.

So the market isn't collapsing. It's splitting. The question a buyer now asks isn't “does a tool exist for this?” — it's “is this cheaper to own than to license?” And the answer flips on one axis: not how hard it is to build, but how expensive it is to keep alive.

Everything in the “survives” column shares a trait: the cost of ownership stays high forever. Regulated workflows carry compliance and liability no internal team wants to sign up for. Marketplaces need a second side you can't conjure in-house. Systems of record are valuable precisely because of years of accumulated edge cases — the mess is the moat. And any product sitting on a web of 40 integrations is really selling “we maintain 40 integrations so you don't,” which no coding agent erases.

This is the same lesson we drew from Norm AI hitting a $1.2B valuation and from the broader collapse in the value of pure generation. When the act of making software approaches free, value migrates to everything around the code: distribution, trust, accumulated data, and the boring operational burden customers will happily pay to never touch.

Expect the build wave to crest and then partially reverse. Right now the industry is in the honeymoon — the agent ships v1, the demo dazzles, the purchase gets cancelled. The bill for maintenance, security and migration arrives on a lag. In 18 months, watch for the “build-buy-back” stories: companies quietly re-licensing the SaaS they replaced once they price the true cost of owning it.

Second, the middle hollows out. Generic horizontal tools with shallow moats get squeezed from both sides — buyers build the simple stuff, and the deep vertical players take the rest. If your product lives in that middle, the next two years are the window to pick a side and go deep.

Third, “buildable” becomes a due-diligence question. Investors and acquirers will start asking every SaaS founder the weekend-clone question directly. The startups that come out of this era strong aren't the ones that shipped fastest. They're the ones building something that's genuinely expensive to own any other way.