The state of AI in 2025: Agents, innovation, and transformation

McKinsey — The state of AI in 2025: Agents, innovation, and transformation

TL;DR

The November 2025 edition of the McKinsey & Company Global Survey on the state of AI, written by QuantumBlack, AI by McKinsey. An online survey of 1,993 participants in 105 nations, fielded 25 June to 29 July 2025 (p. 29).

1. Use is broad. 88% say their organisation regularly uses AI in at least one business function, against 78% a year earlier; 79% for gen AI, against 71% (p. 4). The 2025 question asks about regular use; 2020–2024 asked whether the organisation had adopted AI (Exhibit 1 footnote, p. 4).
2. Depth is thin. Among organisations that use AI, 32% are experimenting, 30% piloting, 31% scaling and 7% “fully scaled” (Exhibit 1, p. 4). The fully scaled share runs from 3% to 10% by company revenue, highest at $5 billion or more (Exhibit 5, p. 10).
3. Agents are early. 23% are scaling an agentic AI system somewhere in the enterprise and 39% are experimenting (p. 5). In no single function do more than 10% report scaling agents (Exhibit 2, p. 5).
4. Enterprise value is rare. 39% attribute any EBIT impact to AI, and most of those say it is under 5% of EBIT (p. 11).
5. “AI high performers” are about 6%. McKinsey’s definition: respondents who attribute 5% or more of EBIT to AI and report “significant” value (p. 14). They redesign workflows 2.8 times as often (55% vs 20%, p. 16) and set growth or innovation goals far more often (p. 15).

Key data

Measure	Value	Base	Page
Regular AI use in ≥1 function	88% (2024: 78%)	all respondents	p. 4
Regular gen AI use in ≥1 function	79% (2024: 71%)	all respondents	p. 4
Phase among AI users: experimenting / piloting / scaling / fully scaled	32 / 30 / 31 / 7%	organisations using AI	p. 4
Fully scaled by revenue: <$100M / $100–499M / $500–999M / $1–4.9B / ≥$5B	5 / 4 / 3 / 9 / 10%	all respondents, by revenue	p. 10
Scaling + fully scaled, ≥$5B vs <$100M	49% vs ~30% (text: “nearly half” vs “29 percent”)	by revenue	p. 10
Scaling an agentic AI system somewhere / experimenting with agents	23% / 39%	all respondents	p. 5
Scaling agents in any single function	≤10%	rebased to total sample	p. 5
Any EBIT impact attributed to AI	39%	all respondents	p. 11
AI improved innovation	64%	AI users, n = 1,753	p. 11
AI high performers	about 6% (n = 109)	all respondents	p. 14
Intend transformative change: high performers vs others	50% vs 14% (3.6×)	n = 109 / 1,884	p. 14
Fundamental workflow redesign: high performers vs others	55% vs 20% (2.8×)	AI users	p. 16
Defined human-in-the-loop processes: high performers vs others	65% vs 23%	AI users	p. 20
>20% of digital budget spent on AI: high performers vs others	35% vs 7%	budget-aware AI users	p. 21
Expect enterprise headcount to fall ≥3% next year / no change / rise ≥3%	32 / 43 / 13%	AI users, n = 1,753	p. 24
At least one negative consequence from AI	51%	AI users	p. 26

Method and definitions

• Sample. 1,993 online respondents “at all levels of the organization” in 105 nations; 38% work for organisations with more than $1 billion in revenue. Data are weighted by each nation’s share of global GDP (p. 29). Respondents report on their own organisation; one respondent is not one firm.
• Fieldwork. 25 June to 29 July 2025 (p. 29).
• Use. “Regular use of AI in at least 1 function”, out of 11 functions (pp. 4, 8). The wording has changed four times since 2017 (Exhibit 1 footnote, p. 4).
• Phases. Experimenting = “any use or early testing”; piloting = “implementing AI for a first use case”; scaling = “growing the deployment/adoption of AI across organization”; fully scaled = “AI has been fully deployed and integrated across organization” (Exhibit 1, p. 4).
• AI agents. “Systems based on foundation models capable of acting in the real world, planning and executing multiple steps in a workflow” (p. 5).
• AI high performers. Body text: EBIT impact “of 5 percent or more” plus “significant” value, “about 6 percent of respondents” (p. 14). Every exhibit footnote says “more than 5%” instead (pp. 14–21). The unweighted count is 109 of 1,993 (5.5%); the 6% is presumably GDP-weighted.
• What drives high performance. A relative-weights analysis of 31 variables (footnote, p. 16). Workflow redesign and human-in-the-loop validation rank among the strongest factors (pp. 16, 19). McKinsey says the practices match its Rewired framework of six dimensions (p. 19).
• Sponsor. McKinsey’s own survey and analysis; no outside sponsor named.

Scope and reliability

• Self-report by individuals. Everything is a respondent’s view of their organisation, including EBIT attribution. Nothing is audited.
• Definition change in the headline. The 78% → 88% rise compares “adopted AI” (2024) with “regular use” (2025). The report draws the line without comment; the change is visible only in the Exhibit 1 footnote (p. 4).
• “High performer” is a value threshold, not a maturity stage. It counts respondents who claim ≥5% of EBIT from AI. It is not the same instrument as the 1% “mature” figure the wiki took from the AI Index 2025, nor as the “fully scaled” phase above.
• Inconsistent threshold wording (“5 percent or more” vs “more than 5%”) and two exhibits print n = 1,933 instead of 1,993 (pp. 5, 21). Small, but they matter if the 6% is compared across editions.
• Consultancy instrument. The practices tied to high performance are mapped to McKinsey’s own Rewired framework, which the firm uses in client work (p. 19). The analysis is correlational and cannot show which way cause runs.
• Later edition. McKinsey published a 2026 edition in August 2026. It could not be reached for this ingest.

Dynamic-capabilities reading

• digital-scouting: an annual scan of how far organisations have taken AI and agents, by function, industry and company size.
• improving-digital-maturity: the experimenting → piloting → scaling → fully scaled phases measure maturity directly, and Exhibit 18 tracks hiring of AI roles (data engineers, ML engineers), more so in large firms (p. 25).
• balancing-digital-portfolios: the report is about scaling up beyond pilots, and about how much of the digital budget goes to AI (35% of high performers spend more than a fifth of it, p. 21).
• internal-enablers: senior-leader ownership (strongly agree: 48% of high performers vs 16% of others, p. 18), agile product delivery and defined human-in-the-loop processes are the practices most tied to value.

How it connects

• enterprise-ai-adoption — the breadth-versus-depth gap in one survey: 88% use, 7% of users fully scaled, 39% any EBIT impact, about 6% high performers.
• ai-maturity-measurement-comparison — this is the newest McKinsey edition the wiki holds. Its “high performer” and “fully scaled” shares are different instruments from the 1% “mature” figure in the thread.
• AI Index 2025 — the 78% for 2024 comes from the previous edition of this survey (July 2024, n = 1,491).
• AI Index 2026 — its corporate-adoption section reprints this survey’s 88%, 79% and the phases by revenue, and calls the data “directional rather than comprehensive”.
• MGI, Agents, robots, and us — cites this survey and makes workflow redesign the condition for AI’s value.
• The CEO’s agenda for the agentic age — McKinsey partners on productivity versus growth, and on redesign versus tool deployment.
• Catlin, The real AI advantage — the Rewired prescription the high-performer practices are mapped to.
• micro-productivity-trap — use-case gains (pp. 12–13) that do not reach enterprise EBIT (p. 11).
• ai-agents — the agent-adoption figures by function and industry (pp. 5–6).
• MIT CISR, August 2025: a staged maturity measure from the same year, with 18% at its top stage (N=152) against McKinsey’s 7% fully scaled; the two use different instruments and samples.
• BCG, The Widening AI Value Gap: BCG’s 5% “future-built” firms in the same year, defined by a capability assessment rather than an EBIT threshold.
• PwC Global CEO Survey 2026: CEOs asked a few months later whether AI has brought significant financial benefit; 56% say not yet.

What was actually ingested

The whole 30-page PDF, converted with pdftotext -layout. The PDF was already in the repo as raw/reports/the-state-of-ai-in-2025-vf.pdf; it was not renamed. The text conversion scrambles the rotated industry labels (pp. 6, 9) and interleaves a hidden layer on p. 20, so Exhibits 1, 5 and 14 were read from the rendered pages. The publication day (5 November 2025) comes from the MGI report’s endnote citing this survey; the PDF itself says “November 2025”.

Linked entities and concepts

• McKinsey & Company — publisher (QuantumBlack is McKinsey’s AI arm).

• Alex Singla and Alexander Sukharevsky — co-authors; QuantumBlack leaders.

• McKinsey Global Institute — Lareina Yee, a co-author, is an MGI director.

• enterprise-ai-adoption, ai-agents, micro-productivity-trap — concepts.

• Lareina Yee (senior partner, MGI director) and Michael Chui (senior fellow): co-authors; promoted to entity pages in this ingest, as their second source after the MGI report.

Dangling (single-source mention, deferred): Bryce Hall (associate partner), Tara Balakrishnan (associate partner).