McKinsey's 2026 State of AI survey, published in August by QuantumBlack, AI by McKinsey, draws on 1,719 respondents across 97 countries and nearly a decade of prior fieldwork.1 Three of its findings frame everything else in the report. Eighty percent of respondents say AI has improved their individual productivity. Thirty-seven percent say AI has contributed to their organization's EBIT, essentially unchanged from a year ago. Six percent qualify as what McKinsey calls AI high performers, a share that is also flat.2
The tension between the first two figures is real, but the interpretation matters. The gap doesn't prove that AI is overstated. It shows that individual capability is advancing faster than organizations are redesigning to capture it. That reading aligns with McKinsey's own conclusion and the frameworks BIP Capital Managing Partner Mark Buffington has developed in this series since early 2026.3
This piece takes the argument one step further. McKinsey measures the gap as an adoption problem. From BIP Capital's vantage point, as an investor in private companies and a partner to advisory firms, it is also a valuation problem in progress. The sorting that public markets began in 2025, between businesses made more durable by AI and businesses made less durable, is now taking place inside organizations, one operating model at a time. The six percent are the first cohort to have visibly sorted themselves.
What the survey found
Adoption has moved past experimentation. Nearly nine in ten respondents report regular AI use in at least one business function, and 44 percent say AI is scaling across their enterprise, up from 38 percent a year earlier.4 The share using AI in three or more functions rose from 51 to 56 percent.4
The scale is concentrating. Among organizations with more than $1 billion in revenue, 54 percent report on enterprise-wide scaling, compared with roughly one-third of smaller organizations. The divergence is sharper for agentic AI: large enterprises scaling agents rose from 27 to 40 percent in a year, while the smaller cohort held flat at 22 percent.5
Build is displacing buy. Thirty-two percent of respondents say their organization has declined to purchase at least one software product or feature because it could be built in-house with agentic coding tools. The figure is 38 percent in professional services and 36 percent in financial institutions.6
Cost is becoming a design variable. One in five respondents say AI operating costs, including tokens, have constrained their organization's AI use. Even so, 60 percent expect AI investment to rise next year, and 28 percent already spend more than a tenth of their technology budget on it.7
The strain is unevenly distributed. Individual benefits are consistent across levels, but 47 percent of midlevel managers and individual contributors report at least one negative effect from AI at work, compared with 31 percent of executives and senior managers. Twenty percent of midlevel managers say AI hurts their ability to think critically; nine percent of C-level respondents say the same.8
Workforce reductions arrived more slowly than expected. Last year, 32 percent of respondents expected AI-driven headcount declines; this year, 14 percent reported one. Two-thirds report little or no AI-related change in total employment. Expectations for the coming year nonetheless rose to 39 percent.9
The repricing has moved inside the organization
In April, in "The Great Repricing," Mark argued that the market was asking every business one question: Is it more durable because of AI, or less? Multiples compressed across software and expanded in sectors supplying AI's physical inputs, but the process was surgical, not uniform. The market was distinguishing adapters from the disrupted, but imperfectly, pricing many businesses as though the answer were already known.3
McKinsey's data show what that sorting looks like from the inside. High performers do spend more on AI, but McKinsey's commentary makes it clear their advantage isn't just budget.10 What distinguishes them is coherence. They set growth and innovation objectives alongside efficiency (74 and 65 percent, versus 47 and 49 percent for all other respondents). Nearly three-quarters have fundamentally redesigned workflows because of AI, up from 55 percent last year; one-quarter of other respondents have done the same. They are twice as likely to report senior-leadership commitment and defined processes for measuring impact, and more likely to manage AI-related risk in areas such as unauthorized actions and technical vulnerabilities.11 In the language of "The Great Repricing," these are the Scenario B businesses: the ones using AI to become more efficient, more scalable, and more durable, whether or not the market has noticed yet.
The build-versus-buy finding deserves particular attention. In April, the thesis that AI would erode per-seat software economics rested on early and highly visible cases. The survey now suggests the substitution is broad-based and reaching well beyond technology companies into professional services, healthcare, energy, and financial institutions. Among high performers, nearly half report forgoing a software purchase in favor of building.12 The decision to build rather than buy is being made in ordinary procurement meetings across industries, not only in a handful of publicized announcements.
Where the survey and our frameworks converge
Four points of agreement stand out.
- Workflow redesign is the dividing line. The "Five Framework Models" piece argued that AI-forward adoption is a condition for remaining competitive, not an advantage in itself, and that the advantage accrues to organizations that redesign how work gets done.13 McKinsey's high-performer data make the same distinction empirically: inserting AI into existing workflows and redesigning workflows around AI produce different financial outcomes.
- Cost reduction is the wrong primary objective. The "Five Framework Models" article cautioned against treating cost reduction as the main benefit of AI adoption and pointed to labor redeployment as the larger source of value.13 McKinsey's high performers pursue efficiency at roughly the same rate as everyone else. The difference is that they also pursue growth and innovation.11
- Leadership owns the redesign. The C-suite recommendations in "The Impact of AI on HR" included building personal working knowledge of the tools and appointing an AI integration lead with direct access to the CEO.14 McKinsey finds high performers twice as likely to report senior leaders who demonstrate visible ownership of AI initiatives.11
- The middle of the organization carries the strain. "The Impact of AI on HR" argued that the most exposed position in an organization is the information-relay layer between leadership and execution.14 McKinsey's role-level data show mid-level managers reporting the highest rates of career anxiety, mental fatigue, and diminished critical thinking of any group surveyed.8
Where the data push back
Agreements are less useful than disagreements. Three findings in the report sit uneasily with our arguments, and each deserves evaluation.
- The pace of structural change is slower than the diamond model implies. "Five Framework Models" describes organizations narrowing at the base as automation absorbs repetitive individual-contributor work.13 McKinsey's data show that the head-count reductions respondents expected a year ago largely did not materialize, and that a plurality still expect no change in most functions.9 The defensible reading is that the direction is supported and the timing is not. The diamond model explicitly assumes that displaced contributors are not redeployed. McKinsey's high performers are doing the opposite, pursuing growth and innovation that require redeployed capacity. The organization's shape is changing; total headcount may not, at least not on the schedule many leadership teams assumed in 2025.
- AI is infrastructure that compounds, and it is also an operating cost that scales with use. We recommend that executives treat AI investment as infrastructure rather than software.14 McKinsey's data introduce a complication: high performers are three times as likely as others to report that costs have constrained their use of software coding agents, and the report's commentary notes that per-token cost declines have been outpaced by growth in token consumption.15 Both framings are correct, and the tension between them is the discipline. Compounding infrastructure still requires a budget. Organizations that treat operating costs as a design constraint from the outset, rather than an afterthought, are the ones McKinsey describes as moving fastest.10
- Individual transformation has not been translated into enterprise transformation. The HR article concluded that people, not organizations, are the unit of transformation.14 McKinsey's data show that people are transforming at scale, while enterprise results remain flat. That is not a refutation of the argument so much as a clarification of who owns which part of it. McKinsey's commentary describes the growing base of individual AI fluency as the foundation for closing the gap and places responsibility for the operating model that would close it with leadership.16 The strain data reinforce the point. Individual contributors and mid-level managers are carrying the adjustment costs of AI without controlling the design of the workflows they operate in. Individual-level agency is necessary. It is not sufficient without top-down redesign.
What an investor's lens adds
McKinsey's survey is written from the operator's perspective, and it's rigorous. An investor reads the same data with a different set of questions.
Durability, not attribution. EBIT attribution is a self-reported, lagging measure. Markets price the expected durability of earnings, which is a forward-looking judgment about whether a competitor with an AI-native cost structure can undercut an operating model. The question we posed in April, whether the managers an allocator has selected have stress-tested their holdings against an AI-native entrant, is the investor’s version of McKinsey’s workflow-redesign finding.3 A business can report zero AI contribution to EBIT and be more durable than one reporting 10 percent, if the first has redesigned around AI and the second has layered it onto existing processes.
The lagging cohort is our cohort. The survey's sample skews large: 36 percent of respondents work for organizations with more than $1 billion in revenue.1 The size finding with the greatest bearing on private markets is that agentic AI adoption is accelerating in that cohort and stalled below it.5 The private growth-stage companies and lower-middle-market borrowers that private markets managers underwrite sit almost entirely on the other side of that line. Two readings are possible. One is that smaller companies are falling behind. The other is that smaller companies carry less legacy process to redesign, and the constraint McKinsey calls an organization's "ability to absorb change,"10 is structurally lower in a 200-person business than a 20,000-person one. The data say the smaller cohort has not moved yet, not that it cannot. Private markets managers also hold something public-market investors do not: a seat at the table when operating decisions are made. Whether a manager uses that seat to shape AI adoption at the portfolio-company level is now a diligence question, and allocators can ask it directly.
Prescription by role and shape. McKinsey reports AI's effects by organizational level but does not prescribe by level.8 The Future of Organizations series does, mapping specific responsibilities and preparation for the C-suite, functional leaders, managers, and individual contributors.14 It also offers a structural forecast, the shift from pyramid to diamond, where McKinsey offers descriptive workforce statistics.13 Both the prescriptions and the forecast are hypotheses. McKinsey's 2027 survey will test them.
Two questions for advisory firm leaders
Most independent advisory firms sit in the cohort McKinsey's data describe least favorably: professional-services organizations below $1 billion in revenue, where 38 percent are already forgoing software purchases, 21 percent report cost constraints, and 61 percent expect to increase AI investment in the coming year.17 The findings in this report apply to the firm as much as to the portfolio.
The first question is about the firm itself. Which workflows have been redesigned around AI, rather than being augmented with it, and where would a competitor with an AI-native cost structure undercut the current operating model?
The second is about the managers the firm allocates to. Do they underwrite AI durability at the portfolio-company level, and can they show how?
The repricing is not over. Individual capability has arrived. The market will next price which organizations are designed to use it.