AI Agent Adoption Statistics (Updated Q4 2026)

If you want citable numbers on AI agent adoption, the honest starting point is this: the curve is steep, but the figures depend heavily on how each survey defines an "agent." The baseline is now near-universal among the larger organizations these surveys tend to reach. McKinsey's State of AI 2026, fielded in May and June and published in August, found nearly nine in ten respondents reporting regular AI use in at least one business function, and 44 percent saying AI is scaling across their enterprise, up from 38 percent a year earlier. On agents the market has split by size: 40 percent of respondents at organizations above 1 billion dollars in revenue report scaling agents, up from 27 percent, while the share at smaller organizations stayed essentially flat at 22 percent (McKinsey, "The state of AI in 2026: On the road to ROI", August 2026). Autonomous, multi-step agents remain the younger, faster-moving slice of that adoption.

Gartner forecasts that 40 percent of enterprise applications will feature task-specific AI agents by the end of 2026, up from less than 5 percent in 2025 (Gartner, August 2025). It also forecasts that a third of enterprise software will include agentic AI by 2028 (Gartner, June 2025). On spend, Deloitte forecasts the agentic AI market at roughly 8.5 billion dollars in 2026, on its way to 35 billion dollars by 2030 (Deloitte TMT Predictions, November 2025).

This roundup pulls the strongest sourced figures into one place: adoption rates, spend, use cases, barriers, ROI, and the 10-20-70 rule on where AI effort should go. Every number names its source, and the what-changed section lists the newest additions in this Q4 2026 update. For a broader read on where the market sits now, see the state of AI agents in mid-2026.

The state of AI agent adoption in 2026

The 2026 picture: AI use is mainstream, agents are moving to scale at large companies, and a large share of smaller companies still report no agent use at all. McKinsey's August 2026 State of AI survey of 1,719 participants in 97 nations found nearly nine in ten respondents reporting regular AI use in at least one function.

Agent use by revenue (McKinsey 2026, percent of respondents):

No single function is far along: IT leads at 12 percent of respondents scaling agents there, followed by knowledge management at 11 percent and software engineering at 10 percent (McKinsey, "The state of AI in 2026", August 2026). For comparison, the prior edition, a survey of 1,993 participants, found 88 percent regularly using AI, up from 78 percent. On agents, 62 percent were at least experimenting, 39 percent experimenting and 23 percent scaling an agentic system in at least one function, with no function above 10 percent (McKinsey, "The state of AI in 2025", November 2025).

Why the gap between "using AI" and "running agents"? Definitions. A chatbot answering one question is not the same as an agent that plans, calls tools, and finishes a task. Capability is improving fast: the Stanford AI Index 2026 recorded agent task success on OSWorld, a benchmark of real computer tasks across operating systems, leaping from 12 percent to roughly 66 percent in a year, while global corporate AI investment more than doubled in 2025 (Stanford HAI, "2026 AI Index Report", April 2026; Stanford HAI, "Inside the AI Index: 12 Takeaways from the 2026 Report", April 2026). Note the ceiling in that same number: agents still fail roughly one attempt in three. If the terminology trips you up, AI agent vs chatbot vs assistant draws the lines clearly.

Why definitions move the numbers

Read every adoption stat with its definition attached. One survey may count any generative AI pilot; another counts only production agents with tool access. That single choice can swing a headline figure by tens of percentage points, and 2026 gave two concrete demonstrations. ServiceNow's vendor-run Enterprise AI Maturity Index found 59 percent of enterprises using agentic AI, but says few are creating autonomous or multistep workflows with it, and only 18 percent are scaling agentic processes across key functions (ServiceNow sells workflow and AI agent software).

And KPMG's quarterly pulse showed agent deployment falling from 42 percent to 26 percent in Q4 2025. It then read 54 percent in Q1 2026, 53 percent in Q2 and 62 percent in Q3. KPMG's Steve Chase said of the Q4 dip that leaders were "not pulling back" and had adopted "more sophisticated definitions of what constitutes a true agent" (KPMG, January 2026). The samples moved too, from 130 leaders in Q4 2025 to 237 in Q1 2026 and 314 in Q3 2026, and KPMG's Q3 wording is looser: building, deploying or developing. The practical takeaway: read each reading with its wording and sample attached. For the mechanics behind what makes something an agent at all, how AI agents work walks through the loop.

Enterprise adoption rates

The headline enterprise numbers for 2026, each with its definition attached: KPMG's Q3 2026 AI Pulse Survey of 314 US leaders at billion-dollar-plus organizations found 62 percent building, deploying or developing AI agents, up from 53 percent the previous quarter. Piloting fell to 25 percent from 55 percent a year earlier (KPMG, September 2026). Two quarters earlier the same pulse read 54 percent actively deploying, up from 11 percent in Q1 2025 (KPMG, March 2026). Wharton and GBK Collective's 2026 study of US enterprises found 67 percent of departments self-reporting that they use agents, but only 17 percent letting an agent act without a person approving each step (Wharton and GBK Collective, October 2026).

PwC's mid-2025 agent survey found 79 percent of senior executives saying agents are already being adopted at their companies, though 68 percent admit half or fewer of their employees actually touch one day to day (PwC, 2025). Among practitioners, LangChain's vendor-run State of Agent Engineering survey found 57.3 percent with agents in production, rising to 67 percent at organizations above 10,000 employees (LangChain, late 2025). A slide on IDC's December 2025 blog says 40 percent of US enterprises and 27 percent of Chinese enterprises have AI agents in production, and the page states no method (IDC, December 2025). Different populations and different definitions, so read them one at a time.

The shape of adoption matters as much as the rate. Two 2026 readings show how far agents remain from blanket use: McKinsey's 2026 survey finds no business function where more than 12 percent of respondents are scaling agents, and in Wharton and GBK's study the largest group of departments, 48 percent, has AI do the work and a person approve it before anything happens. For the enterprise-specific patterns in detail, see enterprise AI agent adoption trends for 2026, and for how smaller companies differ, enterprise vs SMB adoption.

Pilots versus production

A pilot is not adoption. In McKinsey's 2026 split above, many respondents are still experimenting or piloting, which is different from a production agent handling live work. The trust mechanics are visible in KPMG's trendline: 63 percent of large US organizations in its Q1 2026 pulse required human validation of agent outputs, up from 22 percent, which reads as maturity, not hesitation; companies putting agents into real workflows are adding the review gates that real workflows demand (KPMG, Q1 2026). On these surveys, the 2026 picture at large organizations is a growing, more carefully governed production core.

Spend and budget trends

The spend numbers now come in several sizes, and it helps to keep them straight.

Separately, global corporate AI investment reached 581.7 billion dollars in 2025, up 130 percent on the prior year. Private investment accounted for 344.7 billion dollars of that, growing 127.5 percent to make up about 59 percent of the total. United States private AI investment alone reached 285.9 billion dollars, 23.1 times the 12.4 billion China put in (Stanford HAI, "Inside the AI Index: 12 Takeaways from the 2026 Report", April 2026).

Budgets are bigger, and they are moving out of IT. BCG's Applied AI Index 2026 puts AI spending at 3.3 percent of revenue, up from about 1.7 percent in late 2025, with more than 80 percent of it outside the enterprise IT budget (BCG, September 2026). McKinsey finds 28 percent of respondents spending more than 10 percent of their information and communication technology budget on AI, and one in five limiting AI use because of operating costs (McKinsey, August 2026). KPMG's pulse projected average AI investment of 173 million dollars over the next 12 months in Q3 2026, up 33 percent on 130 million a year earlier (KPMG, September 2026). The same series read 207 million in Q1 and 202 million in Q2, on different samples, so we would not read the quarter-to-quarter moves as a trend.

The Deloitte, IDC and Gartner agent-market figures above are the closest thing on this page to an "agent spend" number, and their definitions differ. For how pricing models are evolving, see AI agent pricing trends for 2026; for where the venture money behind the vendors is going, the Q3 2026 funding tracker runs month by month.

A quick reference table

Most common use cases

The leading use cases cluster in a few high-volume functions. McKinsey's August 2026 survey finds respondents most often scaling AI agents in IT (12 percent), knowledge management (11 percent) and software engineering (10 percent), with service operations and product development at 9 percent each (McKinsey, August 2026). Among practitioners, LangChain's vendor-run survey names customer service (26.5 percent) and research and data analysis (24.4 percent) as the most common primary use cases (LangChain, late 2025). We read that as a fit with work that is repetitive, high-volume, and checkable. The clearest breakout category has a dollar figure attached: Menlo Ventures measured AI code-generation spend at 4 billion dollars in 2025, up from 550 million dollars in 2024, with 50 percent of developers using AI coding tools daily (Menlo Ventures, December 2025).

Why those functions first? On our read, they share a pattern: bounded tasks with clear inputs and verifiable outputs. Customer support, code assistance, content drafting, and data lookups all fit that mold. Start narrow, prove value, then widen scope.

Where agents land first

On the same read, we would start with the repetitive and the measurable: triaging tickets, drafting first-pass replies, summarizing long documents, and reconciling data across sources. These are tasks where an agent's output is easy to check and a mistake is cheap to catch. Gartner's 2026 C-suite survey adds a wrinkle: in IT, the most pursued uses were cybersecurity threat detection and service desk automation, at 54 percent each, while the top use for positive returns was intelligent IT asset and cost optimization, at 40 percent (Gartner, September 2026).

Barriers to adoption

Barrier rankings depend on the question asked. KPMG's Q1 2026 pulse found 65 percent of leaders citing difficulty scaling AI use cases as a top obstacle to ROI, up from 33 percent on the prior quarter's smaller sample, with 91 percent saying security and risk considerations shape their AI strategy (KPMG, March 2026). KPMG's Q3 2026 pulse asked about agent deployment, and data readiness and access came first, at 53 percent (KPMG, September 2026), while Wharton and GBK Collective, asking about stronger return, found the cost of tools, compute and licenses the leading barrier, at 34 percent (Wharton and GBK Collective, October 2026). Among the people who build agents, LangChain's vendor-run practitioner survey names output quality the number one blocker, ahead of cost and latency (LangChain, late 2025).

The bluntest barrier forecast comes from Gartner, which predicts that over 40 percent of agentic AI projects will be canceled by the end of 2027, on escalating costs, unclear business value, or inadequate risk controls. The same research flagged "agent washing," estimating that of thousands of vendors claiming agentic products, only about 130 offer the real thing (Gartner, June 2025).

Why reliability still matters

An agent that acts can do damage a chatbot cannot. A wrong answer is annoying; a wrong action can move money or change a record. That helps explain why practitioners in LangChain's vendor-run survey rank output quality first, and why human-in-the-loop design stays central. The practical mitigations are familiar: scope access tightly, keep a person on consequential steps, and measure outputs against expectations rather than trusting them blindly.

ROI and outcomes

ROI is reported unevenly, and the answer depends on the measure. PwC's agent survey found 66 percent of adopting companies reporting measurable value through increased productivity, and 88 percent planning to increase AI budgets in the next 12 months because of agentic AI specifically (PwC, 2025). McKinsey's 2026 survey finds 37 percent of respondents attributing any EBIT impact to AI, about the same as a year earlier, with cost reductions most often reported in supply chain management, service operations and manufacturing, and revenue gains in marketing and sales (McKinsey, August 2026).

Deloitte's October 2026 State of AI in the Enterprise, a survey of 3,500 leaders in 25 countries, found 81 percent reporting faster decisions and execution. Only 25 percent are using AI to deeply transform the business, and 28 percent report revenue gains, up from 20 percent in January (Deloitte, October 2026). To work out your own numbers, see the AI agent ROI calculator guide.

How the leaders measure it

McKinsey's 2026 high performers, about 6 percent of respondents, are twice as likely as others to report defined processes for measuring the impact of AI initiatives, and nearly three-quarters report fundamentally redesigning workflows, against one-quarter of other respondents (McKinsey, August 2026). Our practical read: pick a measurable task, set a baseline, and let the numbers decide whether you scale.

The 10-20-70 rule, and what adopters get wrong

The 10-20-70 rule is BCG's rule of thumb for where AI effort should go: 10 percent on algorithms, 20 percent on technology and data, and 70 percent on people and processes. BCG says the companies it calls pacesetters are guided by it (BCG, "The Leader's Guide to Transforming with AI", December 2024), and it repeats the model in its 2026 Applied AI Index work, calling this "an organizational shift, not just a technology upgrade" (BCG, September 2026).

The rule earns its place in an adoption-statistics roundup because it offers one way to read the statistics. Gartner's prediction that over 40 percent of agentic projects get canceled, KPMG's 65 percent citing scaling difficulty as a barrier, McKinsey's finding that no function has more than 12 percent of respondents scaling agents: we read these as portraits of the 70 percent being skipped. On that reading, organizations buy the 10 and the 20, then wonder why the agent that aced the demo dies in the workflow.

BCG's own workforce data makes the point from the other side: its June 2026 AI at Work survey of 11,749 workers found 47 percent of respondents now spend more time managing and directing AI than doing the task themselves, and 72 percent say AI has considerably changed the skills their job requires (BCG, June 2026). That is what the 70 percent looks like up close: new workflows, new review habits, new skills, none of which ship with the software. The rule got fresh attention in early 2026 (Forbes, January 2026), but BCG was already describing its pacesetters as guided by the rule in December 2024.

For a small business or solo operator, the practical translation is simpler: pick agents that arrive with the process already designed, so less of the 70 percent falls on you.

November 2026 update: same question, answers from 6 to 80 percent

Ask whether companies are getting paid for AI and five major surveys published since August answer anywhere from about 6 percent to 80 percent. They measure different things, and the word "value" hides the difference. We put them side by side because that spread is more useful to a buyer than any single figure on this page, including the deployment-versus-return gap we led with in October.

All of these are self-reported. McKinsey, KPMG, BCG and Deloitte all sell AI consulting; Wharton Human-AI Research and GBK Collective, a research firm, co-sponsor theirs. BCG sets its figure against the 5 percent it reported in 2025, but most of the new number is its wider scaling group, so we would not read it as a tenfold rise. Sources: McKinsey, August 2026; KPMG, September 2026; BCG, September 2026; Wharton and GBK Collective, October 2026; Deloitte, October 2026.

Who gets counted changes the adoption number too

At large US organizations, KPMG finds 62 percent building, deploying or developing agents. In Wharton and GBK's study of US enterprises, 67 percent of departments self-report using agents, but only 17 percent let an agent act with a person reviewing only afterwards or not at all. In the largest group, 48 percent, AI does the work and a person approves it before anything happens.

Across all US employer businesses, the Census Bureau's Business Trends and Outlook Survey reads 24.6 percent using AI in the two weeks to 18 October 2026, up from 17.3 percent on 30 November 2025, with 28.6 percent expecting to within six months. By size, that is 24.3 percent at businesses with 1 to 4 employees and 43.6 percent at those with 250 or more. The top cell is noisy: it read 36.5 percent two weeks earlier (US Census Bureau, BTOS, updated 22 October 2026). BTOS counts businesses, not people: a separate Census survey in March 2026 found about 56 percent of US workers had used AI on the job for at least one of 11 tasks (US Census Bureau, August 2026).

One note for anyone tracking that series. Census broadened the question in November 2025, now labels it "Super Intelligence or SI (formerly referred to as Artificial Intelligence or AI)", and says the wording changed again from the 19 October 2026 collection (BTOS, About the data). We treat 30 November 2025 to 18 October 2026 as the comparable run, and we will check later readings for a break.

The gap has moved to control and cost

BCG finds 42 percent of companies expect to give agents genuine autonomy by 2030, while only 5 percent have the full set of controls in place today. Deloitte finds 40 percent already allow AI to execute decisions without human approval, at least to a moderate degree, and 23 percent say accountability remains unclear when an AI system produces a negative outcome. Deloitte's 40 and Wharton and GBK's 17 can both hold: Deloitte counts moderate use across a global sample, Wharton and GBK count departments where the agent acts before any person sees it. Cost is arriving as well. McKinsey finds one in five respondents limiting AI use because of operating costs, and Deloitte finds 71 percent tracking total AI inference and API spend but only 44 percent tracking per-unit inference cost.

The spending number we declined in October is now on this page, because we read it at the source: Gartner forecasts worldwide AI spending of 2.7 trillion dollars in 2026, up 49.5 percent, while describing generative AI as "firmly in the Trough of Disillusionment" (Gartner, September 2026). More than half of that is infrastructure, so it says more about AI providers building capacity than about what enterprises are getting back.

October 2026 update: deployment versus return

When we added this survey in October, the most useful number on this page was the gap between deployment and return. Writer's 2026 enterprise AI adoption survey found 97 percent of executives saying their company deployed AI agents in the past year, while only 23 percent of organizations report significant ROI from agents and 29 percent from generative AI overall (Writer, published April 2026, read 4 October 2026). In this one vendor survey, deployment is effectively universal at the executive level and the return is rare. On that reading, the buyer's question moved from "should we adopt" to "why does adoption not pay".

The same survey's supervision numbers sit beside that gap: 36 percent of organizations have no formal plan for supervising AI agents, and 67 percent of executives believe their company has already suffered a data leak or breach through unapproved AI tools. Budget is not what this survey finds missing: 59 percent of companies report investing over $1 million a year in AI technology, and 75 percent of executives admit their company's AI strategy is "more for show" than actual internal guidance. In this one vendor survey, that is the backdrop to 97 percent deployment and 23 percent returns.

One caveat, stated plainly: this is a vendor survey of 1,200 executives and 1,200 employees, published by Writer, an enterprise AI company, and conducted with the research firm Workplace Intelligence. Vendor surveys tend to find urgency. Read it next to the studies above, rather than instead of them. The direction it points, deployment outrunning return, sits alongside KPMG's scaling-difficulty barrier (Q1 2026) and Gartner's cancellation forecast. If you are trying to land on the paying side of that gap, the AI agent ROI calculator guide sets out the before-and-after math.

What changed since the mid-2026 edition

This page was first published in June 2026, refreshed in July 2026 for Q3, and updated in October and November 2026. The material movements since late 2025, each with a date:

• Deloitte's TMT Predictions (November 2025) forecast the agentic market at roughly 8.5 billion dollars in 2026, rising to 35 billion by 2030.
• KPMG's Q1 2026 pulse (published March 2026) put active agent deployment at 54 percent of large US organizations, up from 11 percent a year earlier, alongside human-validation requirements rising to 63 percent.
• The Stanford AI Index 2026 (April 2026) quantified the capability jump: agent task success on OSWorld went from 12 percent to roughly 66 percent in a year, and global corporate AI investment more than doubled in 2025, with private investment up 127.5 percent to about 59 percent of the total.
• BCG's AI at Work 2026 (June 2026) moved the workforce numbers: 74 percent of frontline employees are now regular AI users, 30 percent say agents are integrated into their workflows, and 61 percent believe agents could handle at least half their job within three years.
• The October pass (4 October 2026) adds Writer's 2026 enterprise AI adoption survey: 97 percent of executives say their company deployed AI agents in the past year, while 23 percent of organizations see significant ROI from agents and 36 percent have no formal plan for supervising them. It is a vendor-run survey, so it is read here next to the other studies above, not in place of them. One number was deliberately left out of that pass: the widely repeated September 2026 Gartner raise to worldwide AI spending. Gartner's own site could not be read when we checked on 4 October 2026, the figure circulated only in secondary stat-aggregator posts, and a spending number we have not seen at the primary source does not belong on a page other people cite.
• This pass (3 November 2026) adds McKinsey's State of AI 2026 (August 2026), KPMG's Q3 2026 pulse (September 2026), BCG's Applied AI Index 2026 (September 2026), Wharton and GBK Collective's 2026 AI Adoption Report (October 2026), Deloitte's State of AI in the Enterprise (October 2026), the Census Bureau's BTOS reading to 18 October 2026 and two IT use-case figures from Gartner's C-suite survey (September 2026). It also corrects our own copy. McKinsey's 2025 figures now link to the 2025 edition, because McKinsey's main State of AI URL now serves the 2026 survey. We removed a ServiceNow figure of 9 percent, because it is not on the landing page we cite and the full report sits behind a form. We also label IDC's 40 percent of US enterprises with agents in production as what it is, a slide with no stated method. We corrected the base of BCG's 47 percent, the label on Menlo's 37 billion dollars, Stanford's private-investment share, and the wording of the 10-20-70 rule, which BCG states as effort, not value. The Gartner spending figure we declined in October is now included, because we read it at Gartner's own newsroom.

Outlook for the end of 2026 and 2027

The trajectory points up and toward production, on several forecasts. Gartner expects 40 percent of enterprise applications to feature task-specific agents by the end of 2026, from under 5 percent in 2025 (Gartner, August 2025). Deloitte expects as many as 75 percent of companies to invest in agentic AI during 2026 (Deloitte, November 2025). And IDC's longer arc has more than a billion agents actively deployed worldwide by 2029 (IDC, December 2025). Gartner's newest forecast has total AI spending reaching about 3.6 trillion dollars in 2027 (Gartner, September 2026), and 87 percent of the managers and above in Wharton and GBK's study expect their budgets to rise in the next 12 months (Wharton and GBK Collective, October 2026).

Two months will not settle much. We would watch three things into 2027: whether large-company agent scaling keeps pulling away from smaller companies, whether the value readings converge as more companies measure against growth as well as time saved, and whether controls catch up with autonomy, given BCG's 42 percent against 5 percent. Capability is improving fast on the Stanford benchmarks, though agents still fail roughly one attempt in three, and trust, governance, cost, and the 70 percent of the effort that is process change remain open. For the forward view in more depth, see AI agent future trends for 2026.

How Gravity fits

Most of these statistics describe a build-versus-trust gap: the technology is improving fast, but standing up reliable agents, and the process change around them, is hard. Gravity is an AI agent platform in public beta. You describe the outcome you want in plain words, and an expert-built agent runs it and hands back the finished result in about 60 seconds. The Free plan costs $0 with no card; paid plans start at $15 a month, and you can buy more usage from your account if it runs out.

Frequently asked questions

What share of organizations are using AI agents in 2026?

It depends heavily on company size. McKinsey's August 2026 State of AI found 40 percent of respondents at organizations above 1 billion dollars in revenue scaling agents, up from 27 percent, against 22 percent at smaller organizations. At those smaller organizations, 41 percent report no agent use at all. KPMG's Q3 2026 pulse put large US organizations building, deploying or developing agents at 62 percent. Across all US employer businesses, Census data shows 24.6 percent using AI of any kind. Read each figure with its definition and survey date attached.

How much are companies spending on AI and AI agents?

Gartner forecasts a total of 2.7 trillion dollars in 2026 for everything labelled AI worldwide, up 49.5 percent. More than half of it is infrastructure and it includes consumer agents, so it is not a measure of enterprise spend. Menlo Ventures measured enterprise generative AI spend at 37 billion dollars in 2025, up 3.2 times in a year. Deloitte forecasts the agentic AI market specifically at about 8.5 billion dollars in 2026, growing to 35 billion by 2030, and IDC projects agentic AI will exceed a quarter of worldwide IT spending and 1.3 trillion dollars in 2029.

What are the most common AI agent use cases?

McKinsey's August 2026 survey finds agents most often scaled in IT, knowledge management and software engineering. Among practitioners, LangChain's vendor-run survey names customer service and research and data analysis as the top primary use cases. Coding agents are the clearest breakout: Menlo Ventures measured code-generation spend at 4 billion dollars in 2025, up from 550 million in 2024, with half of developers using AI coding tools daily. We read these as bounded, high-volume, checkable tasks.

What are the biggest barriers to AI agent adoption?

It depends on the question asked. KPMG's Q3 2026 pulse names data readiness and access the top barrier to agent deployment, at 53 percent; its Q1 2026 survey found 65 percent citing difficulty scaling use cases as a top ROI barrier, and 63 percent requiring human validation of agent outputs. Wharton and GBK Collective find the cost of tools, compute and licenses the leading barrier to stronger return. Gartner predicts over 40 percent of agentic AI projects will be canceled by end of 2027 on cost, unclear value, or risk controls.

Is AI delivering measurable ROI yet?

Yes, unevenly, and the answer depends on the measure. KPMG's Q3 2026 pulse found 58 percent of leaders at large US organizations reporting measurable business value, McKinsey's 2026 survey found 37 percent attributing any EBIT impact to AI, about the same as a year earlier, and Deloitte found 28 percent reporting revenue gains. PwC's 2025 agent survey found 66 percent of adopters reporting measurable productivity value. McKinsey's high performers, about 6 percent of respondents, are far more likely than others to have fundamentally redesigned workflows.

How many companies actually see ROI from AI agents?

Few, on the one recent count we have found that isolates agents, and that count comes from a vendor. Writer's 2026 enterprise AI adoption survey, a vendor-run survey published by Writer, an enterprise AI company, found 23 percent of organizations seeing significant ROI from AI agents and 29 percent from generative AI, against 97 percent of executives who say their company deployed agents in the past year. Broader 2026 studies of AI as a whole run from about 6 percent (McKinsey's high performers) to 80 percent (Wharton and GBK Collective, any positive return), depending on what counts as a return.

What is the 10-20-70 rule for AI?

It is BCG's rule of thumb for where AI effort should go: 10 percent on algorithms, 20 percent on technology and data, and 70 percent on people and processes. BCG says the companies it calls pacesetters are guided by it, and it repeats the model in its 2026 Applied AI Index work. We read it as a warning against treating agents as a pure technology purchase.

Why do so many AI agent projects stall or get canceled?

We read it as the 70 percent getting skipped. Gartner expects over 40 percent of agentic AI projects to be canceled by end of 2027, citing costs, unclear business value, and weak risk controls. Three 2026 readings fit that view: BCG finds only 5 percent of companies have the full set of agent controls in place, KPMG's Q3 pulse names data readiness the top barrier to deployment, and BCG's AI at Work survey finds people at companies pursuing workflow redesign 24 percentage points more likely to see measurable business improvement.

Sources

• McKinsey & Company, "The state of AI in 2026: On the road to ROI" (global survey, 1,719 participants in 97 nations, fielded May 4 to June 8, 2026), August 2026, read 3 November 2026, mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai, backs the nearly nine-in-ten use figure, the 44 percent enterprise-scaling figure, the 40 and 22 percent agent-scaling split and the no-use figures, the per-function figures, the 37 percent EBIT figure, the 6 percent high-performer figure, the 28 percent budget-share figure and the one-in-five cost constraint.
• McKinsey & Company, "The state of AI in 2025: Agents, innovation, and transformation" (global survey, 1,993 participants in 105 nations, fielded June 25 to July 29, 2025), November 2025, mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai-2025, backs the 88 percent AI-use figure, the 62, 39 and 23 percent agent split, and the under-10-percent-per-function finding.
• KPMG, "AI Quarterly Pulse Survey, Q3 2026" (314 US leaders at $1B+ organizations, fielded July 24 to August 25, 2026), September 2026, read 3 November 2026, kpmg.com/us/en/media/news/q3-ai-pulse-2026.html and the Q3 2026 survey deck, backs the 62 percent, 53 percent, 25 percent piloting, 58 percent value, 53 percent data-readiness and $173 million figures. Earlier readings: Q2 2026 ($202 million), Q4 2025 (130 leaders; 26 percent, 42 percent) and Q1 2025 (11 percent).
• KPMG, "AI Quarterly Pulse Survey, Q1 2026" (237 US leaders at $1B+ organizations, fielded February-March 2026), March 2026, kpmg.com/us/en/media/news/q1-ai-pulse2026.html, backs the 54 percent agent-deployment figure, the 63 percent human-validation figure, the 65 percent scaling-difficulty barrier, the 91 percent security-and-risk figure, and the $207 million average projected AI spend.
• Gartner, "Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027" press release, June 2025, gartner.com, backs the cancellation prediction, the agent-washing estimate, and the ~33 percent of enterprise software with agentic AI by 2028.
• Gartner, "Gartner Predicts 40% of Enterprise Apps Will Feature Task-Specific AI Agents by 2026" press release, August 2025, gartner.com, backs the 40-percent-by-end-2026 forecast.
• Deloitte Global, "Technology, Media & Telecommunications Predictions 2026", November 2025, deloitte.com/global/en/about/press-room/2026-tmt-predictions.html, backs the $8.5 billion 2026 agentic market forecast, the $35 billion 2030 forecast, the up-to-$45 billion figure with orchestration, and the up-to-75-percent investment prediction.
• IDC, "Agentic AI to Exceed 26% of Worldwide IT Spending and $1.3 Trillion in 2029" press release, August 2025, idc.com, and IDC, "Agent Adoption: The IT Industry's Next Great Inflection Point", December 2025, idc.com/resource-center/blog, back the IT-spending share, the 40 percent slide figure and the billion-agents-by-2029 projection.
• Menlo Ventures, "2025: The State of Generative AI in the Enterprise" (495 US enterprise decision-makers, fielded November 2025), December 2025, menlovc.com, backs the $37 billion enterprise generative AI spend, the 16 percent true-agents share, and the $4 billion code-generation figure.
• Stanford HAI, "2026 AI Index Report", April 2026, hai.stanford.edu/ai-index/2026-ai-index-report, and "Inside the AI Index: 12 Takeaways from the 2026 Report", April 2026, hai.stanford.edu/news/inside-the-ai-index-12-takeaways-from-the-2026-report, back the $581.7 billion global corporate AI investment total for 2025 and its 130 percent rise, the $344.7 billion private-investment figure and its 127.5 percent growth, the $285.9 billion United States private-investment figure against China's $12.4 billion, and the 12-to-66-percent OSWorld agent benchmark jump.
• BCG, "AI at Work 2026" (11,749 workers across 14 markets), June 2026, bcg.com/press, backs the 74 percent frontline-use, 30 percent agents-in-workflow, 61 percent half-my-job, 47 percent managing-AI, and 72 percent skills-change figures.
• BCG, "The Leader's Guide to Transforming with AI", December 2024, bcg.com/featured-insights/the-leaders-guide-to-transforming-with-ai, backs the 10-20-70 rule; discussed in Forbes, "Why AI's 10-20-70 Principle Should Matter To CEOs", January 2026, forbes.com.
• BCG, "The Formula for Agentic AI Value" and the Applied AI Index 2026 press release (1,330 CxOs and senior leaders), September 2026, read 3 November 2026, bcg.com/publications/2026/the-formula-for-agentic-ai-value and bcg.com/press, back the 7.5 and 41 percent groups, the 3.3 percent of revenue and over-80-percent-outside-IT figures, and the 42 and 5 percent autonomy and controls figures. BCG sells AI consulting.
• PwC, "AI Agent Survey" (308 US senior executives, fielded April 2025), 2025, pwc.com/us/en/tech-effect/ai-analytics/ai-agent-survey.html, backs the 79 percent adoption, 68 percent half-or-fewer-employees, 66 percent measurable-value, and 88 percent budget-increase figures.
• LangChain, "State of Agent Engineering" (1,340 respondents, fielded November-December 2025), a vendor-run survey; LangChain sells agent engineering tools, langchain.com/state-of-agent-engineering, backs the 57.3 percent agents-in-production, 26.5 and 24.4 percent use-case and quality-as-top-blocker findings.
• ServiceNow, "Enterprise AI Maturity Index 2026" (4,500 executives, 19 countries, 12 industries), read 3 November 2026, servicenow.com/workflow/ai/enterprise-ai-maturity-index-2026.html, backs the 59 percent using-agentic-AI and 18 percent scaling figures. A vendor-run survey; ServiceNow sells workflow and AI agent software. Landing page only; the full report is behind a form.
• Writer, "Enterprise AI adoption in 2026: Why 79% face challenges despite high investment", published April 7, 2026, read 4 October 2026 and 3 November 2026, writer.com/blog/enterprise-ai-adoption-2026, backs the 97 percent agent-deployment, 52 percent employee-use, 23 percent agent-ROI, 29 percent generative-AI-ROI, 79 percent adoption-challenge, 48 percent disappointment, 39 percent no-revenue-plan, 59 percent million-dollar-investment, 75 percent strategy-for-show, 36 percent no-supervision-plan and 67 percent suspected-leak figures. This is a survey of 1,200 C-suite executives and 1,200 employees, published by Writer, an enterprise AI vendor, and conducted with the research firm Workplace Intelligence; the figures are self-reported, not independent research.
• Wharton Human-AI Research and GBK Collective, "2026 AI Adoption Report: The Adaptive Advantage" (about 1,400 respondents at US enterprises with 1,000+ employees, including 1,007 managers and above, fielded July 9 to 29, 2026), October 2026, read 3 November 2026, knowledge.wharton.upenn.edu/special-report/2026-ai-adoption-report, backs the 80 and 28 percent return figures, the 67 and 17 percent agent figures, the 48 percent approval figure, the 87 percent budget figure and the 34 percent cost-barrier figure.
• Deloitte AI Institute, "The State of AI in the Enterprise: Built Differently" (3,500 leaders, 25 countries, six industries, conducted May to June 2026), October 2026, read 3 November 2026, deloitte.com/us/en/about/press-room/enterprise-ai-enters-an-roi-reckoning.html, backs the 81, 25, 28, 40, 23, 71 and 44 percent figures. Press release only; we did not read the full report.
• US Census Bureau, Business Trends and Outlook Survey (biweekly survey of US nonfarm employer businesses), data page updated 22 October 2026, read 3 November 2026, census.gov/hfp/btos/data and BTOS, About the data, backs the 24.6, 17.3 and 28.6 percent figures and the firm-size figures. US Census Bureau, "Does Using Artificial Intelligence Save Time at Work?" (March 2026 Household Trends and Outlook Pulse Survey), August 2026, revised September 24, census.gov/library/stories/2026/08/ai-use-at-work.html, backs the 56 percent figure.
• Gartner, "Gartner Forecasts Worldwide AI Spending to Grow 49.5% in 2026" press release, September 2026, read 3 November 2026, gartner.com, backs the $2.7 trillion, 49.5 percent, $1.48 trillion infrastructure, $29.2 billion agents-and-assistants and $3.6 trillion 2027 figures.
• Gartner, C-suite AI survey press release (1,303 respondents at organizations with $50M+ revenue, fielded January to April 2026), September 2026, read 3 November 2026, gartner.com, backs the 54 and 40 percent IT use-case figures.