TL;DR: Agentic coding is changing the build-vs-buy decision. McKinsey’s 2026 survey found that 32% of respondents’ organizations skipped at least one software purchase because they could build the functionality in-house. That measures whether a purchase was skipped, not how much SaaS spending disappeared. This guide examines the implications for vendors and CTOs, including maintenance, governance and total ownership costs. Enterprise returns remain uneven: 37% report positive EBIT impact from AI and about 6% qualify as high performers.
In this article
- What McKinsey’s State of AI 2026 Actually Found
- McKinsey and Retool: What the Surveys Measure
- What This Means for SaaS Vendors
- The New Build-vs-Buy Decision for CTOs
- Frequently Asked Questions
What McKinsey’s State of AI 2026 Actually Found
McKinsey’s The State of AI in 2026, released August 25, 2026, is the consulting firm’s flagship annual read on enterprise AI adoption — and this year’s edition surfaced a finding that should worry every SaaS vendor. The survey gathered responses from 1,719 professionals and business leaders across 97 countries and multiple industries.
The finding that generated headlines: nearly one-third of respondents (32%) said their organization has decided against purchasing at least one software product or feature because it could build the functionality internally using agentic coding tools.
That’s a new kind of threat to the software industry. For two decades, the SaaS model rested on a simple premise — building software is hard, slow, and expensive, so renting it makes economic sense. Agentic coding tools attack that premise directly. When an AI agent can scaffold, write, test, and iterate on a working internal tool in hours, the calculation that made “buy” the obvious default starts to wobble.
McKinsey framed the broader report around cautious optimism, declaring enterprises finally “on the road to ROI” four years into the generative AI era. But the build-vs-buy data point is the one with the sharpest strategic edge, because it doesn’t just measure AI adoption — it measures AI displacement of existing software spending. That’s a different and more consequential signal.
McKinsey and Retool: What the Surveys Measure
The reason this finding is hard to dismiss is that it doesn’t stand alone. A separate survey from Retool, using a different sample and methodology, landed on nearly the same number.
Retool’s 2026 Build vs. Buy Report — subtitled “How Vibe Coding and Shadow IT Have Reshaped Enterprise Software” — surveyed 817 Retool customers and builders and found that 35% of teams have already replaced at least one SaaS tool with a custom build, and 78% expect to build more custom internal tools in 2026.
The two surveys point toward related behavior, but they are not directly comparable. McKinsey asks whether organizations skipped a purchase; Retool asks its customer and builder sample about replacing existing SaaS functionality. Similar percentages do not establish the same market-wide rate.
But read the number precisely — this is where careful analysis separates from the hype. 32% does not mean a third of all enterprise software has been replaced by AI. It means a third of surveyed organizations skipped at least one purchase. A company that built a single internal dashboard instead of buying one counts the same as a company that gutted its entire SaaS stack. The figure measures the presence of the behavior, not its magnitude.
That distinction matters because the breathless version of this story — “AI is killing SaaS” — overstates what the data supports today. What the data actually shows is an early, broad-based behavioral shift: across most industries, a meaningful minority of organizations have started choosing “build” in situations where they’d previously have chosen “buy.” Whether that minority becomes a majority is the open question. But as a leading indicator, two independent surveys agreeing is about as strong as early signals get.
The Shift Is Broad: Every Sector Is Affected
One of the most striking things in the McKinsey data is how little the behavior varies by industry. This isn’t a tech-sector quirk — it’s showing up nearly everywhere.
The sector breakdown, per McKinsey:
- Technology: 41% — unsurprisingly leading, since tech firms have the in-house engineering talent to exploit agentic tools fastest
- Healthcare (payers and providers): 39% — a notable second, given healthcare’s traditionally cautious procurement
- Professional services: 38%
- Energy and materials: 38%
- Financial institutions: 36%
- Media and telecom: 34%
- Pharmaceuticals and medical products: 33%
The seven sectors listed above span eight percentage points; this selected group is not the full industry distribution. When a behavior shows up this consistently across industries as different as software and pharmaceuticals, it suggests the underlying driver — the falling cost of building software with AI agents — is a general-purpose force, not a sector-specific fad.
For SaaS vendors, that breadth is the uncomfortable part. A trend concentrated in tech could be dismissed as early-adopter noise. A trend that’s nearly as strong in healthcare, energy, and financial services is a structural shift in how enterprises think about acquiring software capability. The vendors most exposed are those selling into every one of these verticals with products that an internal team could plausibly rebuild.
The Two-Speed Race: Why Big Enterprises Are Pulling Away
The single most important nuance in the entire report is that this is not happening evenly. AI agent adoption is splitting into a two-speed race — and the gap is widening.
Among organizations generating more than $1 billion in annual revenue, 40% now say they are scaling AI agents in at least one function — up sharply from 27% a year earlier. Among smaller organizations, the figure is just 22%, and it did not increase at all year over year.
The same divide shows up specifically for coding agents: around 20% of organizations overall report scaling the technology, but that rises to 31% among large enterprises.
This two-speed pattern has real strategic consequences, and they cut against the intuitive story. You might assume small, nimble startups would be the ones exploiting agentic coding to avoid SaaS bills. The data says the opposite: it’s the big, resource-rich enterprises pulling ahead, because scaling AI agents (and building software with them reliably) requires exactly what large enterprises have — engineering depth, platform teams, governance structures, and the budget to absorb the learning curve. Smaller organizations, paradoxically, may remain more dependent on SaaS precisely because they lack the internal capacity to build and maintain alternatives.
For the build-vs-buy question, this means the “build” option is realistically available to a narrower set of organizations than the 32% headline implies. Building internal software with AI agents is cheap to start and expensive to sustain — and sustaining it favors the players who were already big.
What This Means for SaaS Vendors
For the software industry, McKinsey’s finding is an early warning that the ground is shifting under a decades-old business model. The threat isn’t that AI writes better code than SaaS vendors — it’s that AI lowers the barrier to “good enough” internal alternatives for a meaningful slice of the market.
The vendors most exposed share a common profile:
Thin-wrapper products are the most vulnerable. If your SaaS is essentially a polished UI over an API and some CRUD operations, an internal team with agentic coding tools can now replicate the core of it quickly. The value you captured came from saving customers the effort of building — and that effort just collapsed.
High per-seat pricing on narrow feature sets invites replacement. When a tool costs a fortune at scale but only does a handful of things, the math for building an internal version gets attractive fast. Retool’s data pinpointed the categories under most pressure: workflow automations and internal admin tools first, followed by CRMs, BI tools, project management, and customer support.
The market has already reacted. Retool’s report noted that a recent Anthropic Claude update — improving agentic coding capability — triggered a market selloff among software companies, a sign that investors are pricing in exactly this displacement risk.
But the picture isn’t uniformly grim for vendors, and the honest analysis has to say so. The same McKinsey data shows the “build” behavior is concentrated among large enterprises with the capacity to sustain it. And the categories least vulnerable are those where the software encodes genuine, hard-won complexity — payments, tax, compliance, security — that no amount of agentic coding can shortcut, because the difficulty was never writing the code. It was accumulating the domain knowledge, certifications, and edge-case handling that the code represents. Vendors selling deep complexity are far safer than vendors selling convenience.
The New Build-vs-Buy Decision for CTOs
For engineering leaders, agentic coding has genuinely changed the build-vs-buy calculus — but not by making “build” the automatic answer. It changed where the line sits, not which side always wins.
When building now makes sense:
- Thin-wrapper SaaS you were paying mostly for a UI over an API you could call directly
- Internal admin tools — dashboards, workflow automations, internal CRUD apps (Retool’s top replacement category)
- High per-seat pricing where licence cost scales with headcount but the feature set is narrow
- Custom-fit needs where your process is unusual and the off-the-shelf tool forced awkward workarounds
When buying still wins:
- Deep domain complexity — payments, tax, compliance engines with years of edge cases you can’t cheaply replicate
- The maintenance tail — building is day one; security patches, uptime, and upgrades are forever, and an AI agent doesn’t own your on-call rotation
- Assurance and compliance — SOC 2 reports, HIPAA requirements, and audited controls a vendor already carries so you don’t have to
- Anywhere undocumented AI-generated code becomes a liability — a generated app nobody on the team fully understands is tomorrow’s legacy system
The hidden cost nobody budgets for. Retool’s most important finding wasn’t the 35% replacement figure — it was that 60% of teams reported building software outside IT oversight in the past year, with 25% doing so frequently. This is “shadow IT,” and agentic coding has supercharged it. When anyone can spin up a working internal tool in an afternoon, software gets built faster than governance, security review, and documentation can keep up.
That reframes the entire build-vs-buy question for a CTO. It’s no longer just “is it cheaper to build than buy?” It’s “can we build it and own it responsibly?” A tool built outside IT oversight skips the security review, the access controls, the audit trail, and the documentation that make software maintainable and safe. The cost of building collapsed. The cost of governing what you built did not — and it connects directly to the agentic-governance gap that McKinsey’s own AI trust research and Deloitte’s tech leadership study both flagged as the number-one barrier to scaling enterprise AI. The organizations that win the build-vs-buy shift will be the ones that build deliberately, inside their governance perimeter, not the ones that let shadow IT proliferate because the tools finally made it easy.
The Reality Check: AI Spending Still Isn’t Moving Earnings
Before anyone concludes that agentic coding has already transformed enterprise economics, McKinsey’s broader report delivers a sobering counterweight that belongs in any honest analysis.
Despite record AI spending, 37% of respondents report positive enterprise-level EBIT impact from AI, while about 6% qualify as AI high performers. McKinsey’s framing — enterprises are “on the road to ROI” — is deliberately cautious. Four years into the generative AI era, most organizations still can’t point to AI moving their actual earnings.
This matters for the build-vs-buy story in two ways. First, it’s a reminder that adoption and impact are different things. A company skipping a SaaS purchase to build internally is a real behavioral change, but it doesn’t automatically translate into measurable financial benefit — the internal build has to actually work, get maintained, and deliver value the purchased tool wouldn’t have. Second, it tempers the “AI is killing SaaS” narrative with the reality that most enterprises are still in the experimentation phase, where the economics remain unproven in either direction.
The most credible read of the full report is neither “SaaS is dead” nor “nothing has changed.” It’s that the option to build instead of buy has become real for a meaningful and growing share of organizations, that large enterprises are exploiting it fastest, and that whether it translates into durable value depends entirely on execution — including the unglamorous governance work most organizations haven’t done yet. That’s a genuine shift worth acting on, minus the hyperbole.
Frequently Asked Questions
What did McKinsey’s State of AI 2026 report find about build vs buy?
McKinsey’s The State of AI in 2026, published August 25, 2026, found that 32% of surveyed organizations decided against buying at least one software product or feature because they could build the functionality internally using agentic coding tools. The survey covered 1,719 respondents across 97 countries. It signals an early but broad-based shift in enterprise software procurement.
What are agentic coding tools?
Agentic coding tools are AI systems that can autonomously write, test, debug, and iterate on software with limited human direction — going beyond autocomplete to complete multi-step engineering tasks. Examples include Claude Code, Cursor Composer, and Codex CLI. They’re the technology behind the build-vs-buy shift, because they dramatically lower the time and cost of building internal software.
Does 32% mean a third of software has been replaced by AI?
No. The figure means 32% of surveyed organizations skipped at least one software purchase in favor of building internally — not that a third of all enterprise software has been replaced. A company that built a single internal tool counts the same as one that replaced its entire stack. It measures the presence of the behavior, not its scale. As an early signal it’s significant, but the “AI is killing SaaS” framing overstates what the data currently supports.
Which industries are most affected by the build-vs-buy shift?
Per McKinsey, technology leads at 41%, followed by healthcare payers and providers (39%), professional services (38%), energy and materials (38%), financial institutions (36%), media and telecom (34%), and pharmaceuticals and medical products (33%). These seven sectors span eight percentage points; they are a selected group, not the complete industry breakdown.
What is the two-speed AI race?
It’s the widening gap between large and small organizations in AI agent adoption. McKinsey found that enterprises with more than $1 billion in revenue scaling AI agents rose from 27% to 40% year over year, while smaller organizations stayed flat at 22%. Large enterprises are pulling ahead because scaling agents requires engineering depth, governance, and budget that smaller firms often lack — meaning the “build” option is realistically available to a narrower set of organizations than the headline suggests.
Does this mean SaaS is dying?
Not exactly. The most exposed SaaS products are “thin wrappers” (a UI over an API), internal admin tools, and high-per-seat products with narrow feature sets. Software encoding deep domain complexity — payments, tax, compliance, security — remains well protected, because the difficulty there was never the code but the accumulated domain knowledge, certifications, and edge-case handling. SaaS vendors selling genuine complexity are far safer than those selling convenience.
When should a company build instead of buy in 2026?
Building now makes sense for thin-wrapper SaaS, internal admin and workflow tools, high-per-seat products with narrow features, and cases needing unusual custom fit. Buying still wins for deep domain complexity, products where the maintenance tail (security patches, uptime, upgrades) is heavy, anything requiring independent assurance or regulatory compliance, such as SOC 2 reports and HIPAA requirements, and situations where undocumented AI-generated code would become an unmaintainable liability.
What is the hidden risk of building software with AI agents?
Governance. Retool found 60% of teams built software outside IT oversight in the past year — “shadow IT” that agentic coding has supercharged. Software now gets built faster than security review, access controls, documentation, and governance can keep up. The cost of building collapsed, but the cost of owning and governing what you built did not. The real build-vs-buy question is no longer just cost, but whether you can build it and own it responsibly.
Is AI actually delivering financial returns for enterprises yet?
Mostly not yet. McKinsey found that despite record AI spending, 37% of respondents report positive enterprise-level EBIT impact, while about 6% qualify as AI high performers. The firm describes enterprises as “on the road to ROI” rather than having arrived. This tempers the build-vs-buy story: skipping a SaaS purchase to build internally is a real behavioral change, but it doesn’t automatically produce measurable financial benefit without solid execution and maintenance.
How does this connect to broader enterprise AI trends?
It’s the procurement-side consequence of the agentic AI shift documented across 2026 research. McKinsey’s AI trust study and Deloitte’s tech leadership study both found that governance — not model capability — is the primary barrier to scaling enterprise AI. The build-vs-buy shift makes that concrete: agentic coding lets organizations build software they used to buy, but doing so responsibly requires exactly the governance maturity most organizations are still building.
Final Take
McKinsey’s State of AI 2026 finding — 32% of organizations skipping a software purchase to build internally with agentic coding tools — is one of the clearest signals yet that AI is reshaping not just how software gets built, but the fundamental economics of whether to build it at all. Backed by Retool’s near-identical 35% figure, the trend is real and broad, appearing across several industries.
But the honest reading requires holding three things at once. The shift is genuine: the option to build instead of buy has become viable for a meaningful, growing share of organizations, and SaaS vendors selling replaceable convenience should be worried. The shift is uneven: it’s large enterprises with engineering depth pulling away, not scrappy startups, and the “build” option is narrower in practice than the headline implies. And the shift is unproven at the bottom line: 37% report positive EBIT impact and only about 6% qualify as high performers, so adoption alone does not establish broad financial transformation.
For CTOs, the practical takeaway is that agentic coding moved the build-vs-buy line — it didn’t erase it. Thin wrappers, internal tools, and overpriced narrow products are now often cheaper to build. Deep complexity, heavy maintenance, and regulatory certification still favor buying. And the decisive new variable isn’t cost at all — it’s governance. The 60% of teams already building outside IT oversight are a warning: the organizations that win this shift will be the ones that build deliberately and own what they build, not the ones that let the falling cost of code outrun their ability to govern it. Building got cheap. Owning it responsibly is where the real work — and the real advantage — now lives.
Published August 2026 · The AI & Tech Society · digitalstrategy-ai.com
Sources: McKinsey & Company, “The State of AI in 2026” (published August 25, 2026, based on 1,719 respondents across 97 countries), via McKinsey QuantumBlack and reporting from BigDATAwire, The Register, TechTimes, and ANI/Asianet; Retool “2026 Build vs. Buy Report: How Vibe Coding and Shadow IT Have Reshaped Enterprise Software” (800+ respondents), via Business Wire; Gartner agentic AI cost analysis (March 2026) referenced for inference-cost context. The 32% and sector figures are McKinsey’s; the 35%, 78%, and 60% figures are Retool’s. All survey figures are self-reported by respondents about their own organizations and measure the presence of a behavior, not its magnitude. This article is analysis, not procurement or investment advice. Verified September 2026.
Primary sources: McKinsey’s State of AI 2026 survey and Retool’s 2026 Build vs. Buy report. The 37% EBIT-impact figure and 6% high-performer figure measure different things and should not be treated as complements.
Discover more from The Tech Society
Subscribe to get the latest posts sent to your email.