McKinsey Survey Finds a Third of Companies Skipped Software Purchases Because AI Agents Could Build the Same Tool

The pressure on enterprise software vendors just got a number attached to it.

According to McKinsey's State of AI 2026 survey, published August 25, nearly one in three organizations has now passed on buying a software product or feature because an AI coding agent could build the equivalent in-house. The survey drew responses from 1,719 organizations across 97 countries, with fieldwork running from May through early June.

The headline figure is 32%. In the technology sector specifically, that rate climbs to 41%. Among what McKinsey classifies as "high performers", the 6% of respondents that attribute at least 5% of EBIT to AI, the share approaches 50%, according to analysis of the report by byteiota.

The mechanism here is straightforward: agentic coding tools have lowered the floor on internal development enough that procurement teams are now running a build estimate alongside every vendor evaluation. The question used to be "can we build this?", which implied months of engineering time. Now it's "how long does it take the agent?"

But the McKinsey data has a second number that software buyers should take seriously before they start cancelling contracts. The share of organizations reporting any EBIT impact from AI sat flat at 37% year over year, with only 6% clearing what the report calls the high-performer bar. Skipping a purchase isn't the same as realizing the savings. As noted in analysis published by The D*ai*ly Brief, the cancelled SaaS line is visible in a budget; what replaces it mostly isn't, until compute bills, token costs, and internal maintenance accumulate in year two.

Security is the other variable that doesn't show up in the build-vs-buy spreadsheet. Veracode's 2026 GenAI Code Security Report, which tested more than 100 AI models on standardized code-generation tasks, found an average security pass rate of 56%, meaning 44% of AI code-generation tasks introduced a security vulnerability in testing. That rate has barely moved from the previous year. Vendor SaaS absorbs established compliance certifications and liability. Internal builds carry the full burden.

The trend is sharpest in industries where speed and customization matter most. According to the McKinsey report as covered by ANI, the shift was particularly pronounced in technology and healthcare, followed by professional services and energy. Regulated sectors like insurance and the public sector are buying audit evidence and liability coverage, not just code, which is harder to replicate with an agent script.

Enterprise adoption of agents is accelerating regardless. The McKinsey survey found that 40% of respondents at organizations with more than $1 billion in annual revenue said they were scaling AI agents in at least one function, up from 27% a year earlier.

For software vendors, the near-term risk isn't existential, it's category-specific. Tools with commoditized feature sets and no proprietary data or compliance moat are the first candidates to get replaced. Products that offer guaranteed support, audit trails, or specialized workflows that are difficult to replicate as a prompted agent script have more durable ground.

The more honest read of the 32% figure is this: enterprise buyers now treat internal development as a live option at the start of a procurement cycle, not as a fallback. That's the structural shift. Whether the builds actually deliver is a question the next round of surveys will need to answer.

Sources cited:
- McKinsey State of AI 2026 (via ANI) (https://aninews.in/news/business/ai-coding-agents-threaten-to-reshape-software-spending-as-companies-choose-to-build-rather-than-buy-mckinsey20260906210256/)
- The D*ai*ly Brief (beri.net) (https://www.beri.net/article/mckinsey-state-of-ai-2026-agentic-coding-build-vs-buy-run-cost)
- byteiota (https://byteiota.com/agentic-coding-build-vs-buy/)
- AI Unfiltered (arturmarkus.com) (https://www.arturmarkus.com/mckinsey-32-of-companies-killed-a-software-purchase-because-coding-agents-could-build-it/)