McKinsey Survey: 32% of Companies Abandon Software Purchases Due to AI Coding Agents; High-Performer Rate Approaches 50%

On August 25, 2026, McKinsey released "The State of AI: 2026 Global Survey," confronting the software industry with a number it cannot ignore: among 1,719 survey responses completed across 97 countries over five weeks (May 4 to June 8), 32% of respondent companies said they had proactively abandoned the purchase of at least one software product or feature because AI coding agents can replicate the same functionality in-house. This is the first time an authoritative institution has quantified the actual incidence of "AI coding agents replacing SaaS purchases" through a large-scale survey.

The number itself is hardly surprising; what is surprising is its distribution. The report defines "high-performing enterprises" as those that can attribute at least 5% of earnings before interest and taxes (EBIT) to AI and find that AI delivers substantial value—a threshold met by only 6% of respondents. Among this 6%, the share that abandoned at least one software purchase approaches 50%, compared with just 31% among other respondents. Companies that have genuinely embedded AI into production processes and realized financial returns are more willing to replace purchases with in-house builds than to simply layer AI on top of existing SaaS subscriptions.

How Coding Agents Rewrote the Economics of Procurement

In the traditional "build vs. buy" decision framework, the hidden costs of building in-house are often underestimated: recruiting developers, maintaining and iterating, and handling compliance and security. Stacked together, these costs usually make in-house development far more expensive than purchasing. But AI coding agents are rewriting this algorithm. When an AI system capable of code generation, testing, and debugging can deliver a functional module in hours, rather than requiring engineer schedules spanning weeks, the default logic of "buy first" begins to loosen.

For enterprises, what truly changes is the decision threshold: as long as AI can roughly satisfy a need, the short-term cost of building in-house becomes competitive with annual SaaS subscription fees. As a result, more and more teams are asking "can we do this ourselves?" before signing new contracts. The depth of this shift varies across industries—according to the McKinsey report, the share choosing in-house builds over purchased alternatives is highest in technology at 41%, healthcare payers and providers at 39%, professional services and energy and materials at 38% each, and financial institutions at 36%. No major industry falls below one-third.

Large enterprises with annual revenue exceeding $1 billion are leading the way: 40% have deployed AI agents at scale in at least one business function, up 13 percentage points from the prior year; for small and mid-sized enterprises, that figure remains at 22%, barely changed. This means the wave of purchase substitution is not unfolding across the board but is being driven by well-capitalized, engineering-strong front-runners willing to bear the risks of building in-house. The most direct pressure falls on vertical tools with standardized functionality covering a single function—report management, data export, simple workflow automation. Such products are the easiest to reproduce with a precise sequence of prompts and are often the first test cases enterprises tackle.

Procurement Decisions Are Shifting, but the Overall Market Has Not Shrunk

A key benchmark comes from market research firm Gartner: global software spending in 2026 is projected to reach $1.468 trillion, an increase of 15.5% from 2025. This shows that while the build/buy balance tilts toward the "build" side, the overall software market is still expanding—new AI infrastructure, collaboration platforms, and cloud spending have partly filled the gap left by vertical SaaS products displaced by in-house alternatives.

Enterprise research platform Retool's February 2026 survey (817 users, skewed toward engineers) found that 35% of customers had replaced at least one SaaS product with in-house tools, and 78% expected to build more internal tools in 2026. Retool's sample structure means its figures run higher and cannot be compared directly with McKinsey's 32%, but the two datasets point in the same direction: substitution has moved from fringe experiments into mainstream procurement decisions.

Capital markets have priced in this structural shift even before actual revenue losses materialized. After Anthropic released its Claude coding agent with legal, sales, and data-analysis automation capabilities this February, RELX fell 14% in a single day, Thomson Reuters dropped 16%, Wolters Kluwer declined 12%, and LSEG fell 13%. The share-price repricing occurred before any meaningful revenue losses appeared, indicating that the market is pricing the structural risk to business models separately rather than merely reflecting current-period performance.

Two Quantifiable Constraints on In-House Builds

This wave of substitution does not come without costs. The McKinsey survey shows that roughly one-fifth of respondent companies said AI operating costs (including token fees) had constrained the scale of their AI usage; among high-performing enterprises, the share constrained by token costs is three times that of other respondents—precisely because they use AI the most, their bills are the clearest. McKinsey's recommendation in the report to treat operating costs as a design constraint rather than an afterthought is itself a signal: building in-house does not mean low cost, and runtime token consumption may prove harder to predict and control than an annual SaaS subscription fee.

The second constraint is security. According to a report released by security firm Veracode on July 28, 2026, the overall security pass rate of AI-generated code averages just 56%—Python performs relatively well at 85%, while Java is the lowest at 30%. Retool's survey found that 93% of CIOs/CTOs are concerned about unreviewed AI-generated code entering production environments, and 60% of in-house developers have effectively circumvented IT departments' oversight processes. In other words, while replacing a SaaS subscription, enterprises take on new shadow IT risks.

One figure deserves a closer look: the McKinsey survey shows that 37% of respondent companies believe AI has a quantifiable impact on their EBIT, essentially flat from the prior year; high-performing enterprises that truly derive more than 5% of EBIT from AI still account for only 6% of all respondents. This indicates that "32% of companies abandoning software purchases" and "most companies gaining financial returns from AI" are two separate things. Companies are accelerating in-house building, but the rate at which in-house builds are converted into profit has not risen in step.

What Signals to Watch Next

The SaaS industry's bifurcation will become measurable over the next 12 to 18 months. Tools with standardized functionality and low coupling to core business processes will remain under pressure; platforms that deeply integrate business processes and benefit from network effects or compliance moats will experience less real impact than current market pricing implies. Investors have preemptively sold off, but differentiated SaaS products and generic tools will trace entirely different curves.

The other signal is the trajectory of token costs. The 6% of high-performing enterprises identified by McKinsey are a leading indicator for the entire market—they are already feeling the pressure of token bills, and they are also the most aggressive cohort in substituting in-house builds for purchased software. If inference costs continue to decline, the economics of in-house development will tilt further in a favorable direction, and mid-sized enterprises will follow suit at a significantly faster pace; if costs stay high, this wave of substitution will remain concentrated among front-runners with the capacity to absorb operating expenses, and 32% could remain a ceiling rather than a starting point for a considerable time.