On this page
Ask how many companies run AI agents and you can find a published answer anywhere from 7% to 97%. Both numbers are real. One comes from 105 US federal decision-makers asked about production systems; the other from executives in a survey run by a company that sells agents, asked whether they deployed any in the past year. Articles quote them side by side as if they measured the same thing.
So I stopped collecting the numbers and started grading the surveys. This page logs every survey I could find and verify, published since that reports a share of organizations, workers or developers using AI agents, with the sample, who paid, whether the respondents were screened, how the survey defines an agent, and a transparency grade from A to C. As of it holds 35 surveys.
If you want the statistics themselves, they live on my AI agent adoption statistics page, the broader AI agent statistics for 2026, and the small business AI adoption numbers. This page is the layer underneath: which of those numbers to trust, and how far. A disclosure, since it is the point of the page: I run Gravity, an AI agent platform in private alpha, so I have a stake in agent adoption too. None of the surveys here is ours.
What the index shows
- 35 AI agent adoption surveys indexed and graded, published between 26 February 2025 and 29 September 2026 by 26 publishers. Each row records the sample, the sponsor, the definition of "agent" and a method grade; 8 earn an A, 16 a B and 11 a C.
- "In production" figures run from 7% to 97%, with a median of 52.5%, across 14 surveys (Booz Allen Hamilton low, Writer high; open samples only, latest edition of each series). Counting any stage from pilot to production, the range is 35% to 99%.
- Surveys from companies that sell agents report higher adoption: a median of 57.2% across 10 vendor surveys, against 48% across 9 from consultancies and independent publishers, a gap of 9.2 points. Consultancies alone: 35%; analysts, media and others: 55%.
- Only 23% define what they mean by "agent" (8 of 35); 10 define it or print the question wording. When the bar is set high, the number collapses: VentureBeat Research finds 10% (of organizations where true agents are the majority of what they run); Gartner finds 15% (of IT application leaders working with fully autonomous agents); Menlo Ventures finds 16% (of enterprise deployments that qualify as true agents).
- 89% state a sample size (31 of 35) and 54% give fieldwork dates (19 of 35). 3 surveys only admitted respondents already working with agents, which inflates any adoption share they report.
Updated . Medians use one figure per survey, the latest edition of repeated series, and exclude screened samples. Percentages are the publishers' own; nothing is re-weighted.
Cite this page
Cite this page: "AI agent adoption survey index", Gravity, updated , https://gravity.fast/data/ai-agent-adoption-survey-index/. The index (grades, checks, classifications) is licensed CC BY 4.0: copy it, chart it, quote the numbers, and link back here. The survey figures belong to their publishers, so cite the original survey too. Every row as a spreadsheet: data.csv.
The surveys
Newest first. Each row links the publisher's own page, which is where every figure was read. "Headline figure" is the publisher's number, labelled with what it actually counts.
| Survey | Sample | Sponsor | Definition of "agent" | Headline figure | Grade |
|---|---|---|---|---|---|
| Agoda Agoda AI Developer Report 2026 Published 29 September 2026 | Not disclosed Developers; Indonesia, Malaysia, Thailand, Philippines, Singapore, Vietnam, India. Fieldwork: not disclosed. | Agoda other company | Not defined | 53% Deployed or in production 53% of developers use AI agents in selected production workflows or broadly across their organizations. Conducted with Macramé Consulting; read from the release as published by Hotel News Resource. Only 38% say their codebase is mostly or fully ready for autonomous agents. | C 2 of 5 |
| KPMG KPMG AI Quarterly Pulse Survey, Q3 2026 Published 24 September 2026 | n = 314 314 US C-suite and business leaders. Enterprise; United States. Fieldwork: 24 July to 25 August 2026. | KPMG consultancy | Not defined | 62% Any stage, piloting to production 62% of organizations are building, deploying or developing AI agents, up from 53%; 25% are actively developing or implementing multi-agent systems. The measure widened this quarter from deployment to building, deploying or developing, so 62% is not directly comparable with the earlier deployment figures. | B 3 of 5 |
| Cisco (study by Omdia) The Impact of Agentic AI on Network Operations Published 23 September 2026 | n = 1,000 1,000 IT and network operations leaders at organizations with 500 or more employees. Enterprise; Not stated. Fieldwork: not disclosed. | Cisco agent vendor, sells agents or agent tooling | Not defined | 51% Deployed or in production 51% run agentic AI that acts in production network operations today; 75% have deployed AI for network operations. Conducted independently by Omdia for Cisco. Covers network operations only, not agents across the business. | B 3 of 5 |
| Temporal The State of Development 2026 Published 25 August 2026 | n = 554 554 engineers and engineering leaders who currently use AI agents. Developers; United States and United Kingdom. Fieldwork: 29 April to 25 May 2026. Screened sample. | Temporal agent vendor, sells agents or agent tooling | "An LLM instance capable of taking multi-step actions on its own." | 80.8% Other agent measure 80.8% of engineers who use AI agents use them daily, up from 47.3% a year earlier. Respondents had to be using AI agents to qualify (Qualtrics panel), so the figure is frequency of use among users, not adoption. | A 5 of 5 |
| The Modern Data Company Annual survey of enterprise data leaders (interim findings) Published 21 August 2026 | n = 540 More than 540 enterprise data leaders and practitioners in 66 countries. Mixed; 66 countries. Fieldwork: not disclosed. | The Modern Data Company agent vendor, sells agents or agent tooling | Not defined | 23.5% Deployed or in production 23.5% already use AI agents in production in data and analytics workflows; 33.8% are piloting. Interim findings, read via IT Brief. Covers data and analytics workflows only. The sponsor sells a data platform marketed for AI agents. | C 2 of 5 |
| Deloitte AI agents are only the beginning: The path to agentic transformation Published 12 August 2026 | n = 501 501 US senior managers to C-suite at organizations at least piloting agentic AI. Enterprise; United States. Fieldwork: April to June 2026. Screened sample. | Deloitte consultancy | Not defined | 58% Deployed or in production 43% are expanding AI agent deployments across functions and 15% have reached scaled, orchestrated multi-agent adoption; 42% are still testing. 58% adds the two stages beyond testing (43% plus 15%). Every organization had to be at least piloting agentic AI to take part, so this is not a population adoption rate. | B 4 of 5 |
| VentureBeat Research VB Pulse: Where enterprise AI agent governance hasn't caught up Published 24 July 2026 | n = 573 573 qualified respondents across five parallel surveys, organizations with 100 or more employees. Enterprise; Not stated. Fieldwork: June 2026. | VentureBeat media | true agents: deployed agents that "can complete multi-step work on their own" | 10% Other agent measure 71% say a quarter or fewer of their deployed "agents" can complete multi-step work on their own; only 10% say true agents are the majority of what they run. Self-selected, not a probability sample, as VentureBeat states. Measures how many deployed agents are real agents, not adoption. | A 5 of 5 |
| Booz Allen Hamilton Booz Allen survey of federal agencies on agentic AI Published 21 July 2026 | n = 105 105 federal government IT and cybersecurity decision-makers and influencers. Mixed; United States (federal government). Fieldwork: April 2026. | Booz Allen Hamilton consultancy | Not defined | 7% Deployed or in production Only 7% of federal agencies have agentic AI in production; 51% are piloting or testing and 22% plan to deploy within 12 months. Fielded online with Market Connections. Read via GovInfoSecurity's report of the release; the any-use figure (58%) adds piloting and production. | B 4 of 5 |
| KPMG KPMG AI Quarterly Pulse Survey, Q2 2026 Published 24 June 2026 | n = 204 204 US C-suite and business leaders. Enterprise; United States. Fieldwork: 28 April to 25 May 2026. | KPMG consultancy | Not defined | 53% Deployed or in production 53% agent deployment, on par with the prior quarter. This release gives the prior quarter as 55%; the Q1 2026 release itself said 54%. | B 3 of 5 |
| Box 2026 State of AI in the Enterprise Published 11 June 2026 | n = 1,640 1,640 IT decision-makers in the US, UK, France and Japan. Mixed; United States, United Kingdom, France, Japan. Fieldwork: not disclosed. | Box agent vendor, sells agents or agent tooling | Not defined | 83% Deployed or in production 83% of organizations are running AI agents; 19% run them autonomously at scale. Respondents self-selected into four AI-maturity tiers. Box sells AI agents for enterprise content. | B 3 of 5 |
| Boston Consulting Group AI at Work 2026 Published 3 June 2026 | n = 11,749 11,749 workers in 14 markets. Workers; 14 markets. Fieldwork: not disclosed. | Boston Consulting Group consultancy | Not defined | 30% Workers or developers using agents 30% of respondents say AI agents are already integrated into their workflows, more than double the 13% a year earlier. 52% of respondents say they have a limited understanding of what agents are. | C 2 of 5 |
| OutSystems 2026 State of AI Development Published 13 April 2026 | n = 1,900 Nearly 1,900 IT leaders worldwide, including 527 in Asia Pacific. Mixed; Global. Fieldwork: not disclosed. | OutSystems agent vendor, sells agents or agent tooling | Not defined | 96% Any stage, piloting to production 96% of organizations are already using AI agents in some capacity; 94% worry about AI sprawl. Fielded by an unnamed third party. "In some capacity" is the loosest use measure in the index. | B 3 of 5 |
| Writer Enterprise AI adoption in 2026 Published 7 April 2026 | n = 2,400 1,200 C-suite executives and 1,200 non-technical employees who use AI at work. Mixed; Global (countries not listed). Fieldwork: not disclosed. | Writer agent vendor, sells agents or agent tooling | Not defined | 97% Deployed or in production 97% of executives say their company deployed AI agents in the past year; 52% of employees already use them. Fielded with Workplace Intelligence. The highest deployment figure in the index; the same survey finds only 23% see significant ROI from agents. | B 3 of 5 |
| KPMG KPMG AI Quarterly Pulse Survey, Q1 2026 Published 31 March 2026 | n = 237 237 US C-suite and business leaders. Enterprise; United States. Fieldwork: 17 February to 17 March 2026. | KPMG consultancy | Not defined | 54% Deployed or in production 54% of organizations are actively deploying AI agents; 63% now require human validation of agent outputs. Organizations with $1 billion or more in annual revenue; more than a third above $10 billion. | B 3 of 5 |
| Deloitte The State of AI in the Enterprise, 2026 Published 21 January 2026 | n = 3,235 3,235 senior business and IT leaders directly involved in AI programs. Enterprise; 24 countries. Fieldwork: August to September 2025. | Deloitte consultancy | Not defined | 23% Deployed or in production 23% of companies use agentic AI at least moderately today; 74% expect to within two years; 21% have mature agent governance. The report prints its question wording ("To what extent is your organization utilizing agentic AI in its operations?"). | A 5 of 5 |
| KPMG KPMG AI Quarterly Pulse Survey, Q4 2025 Published 15 January 2026 | n = 130 130 US C-suite and business leaders. Enterprise; United States. Fieldwork: not disclosed. | KPMG consultancy | Not defined | 26% Deployed or in production 26% of organizations report agent deployment, down from 42% in Q3; KPMG attributes part of the drop to leaders adopting stricter definitions of a true agent. The release says leaders "adopt more sophisticated definitions of what constitutes a true agent", the clearest case in this index of a definition moving a headline number. | C 2 of 5 |
| G2 G2 AI agents survey (August 2025), cited in G2's Enterprise AI Agents Report Published 15 December 2025 | Not disclosed Mixed; Not stated. Fieldwork: August 2025. | G2 software marketplace | Not defined | 57% Deployed or in production 57% of companies already have AI agents in production; 22% are piloting. G2 cites its own August 2025 survey without sample or respondent details on the page read. | C 1 of 5 |
| Menlo Ventures 2025: The State of Generative AI in the Enterprise Published 9 December 2025 | n = 495 495 US enterprise AI decision-makers. Enterprise; United States. Fieldwork: 7 to 25 November 2025. | Menlo Ventures investor | "systems where an LLM plans and executes actions, observes feedback, and adapts its behavior" | 16% Other agent measure Only 16% of enterprise and 27% of startup AI deployments qualify as true agents. A share of deployments, not of organizations, so it sits outside the adoption ranges. The strictest definition in the index and one of the lowest numbers. Menlo invests in AI companies. | A 5 of 5 |
| LangChain State of Agent Engineering Published December 2025 | n = 1,340 1,340 responses to a public survey. Mixed; Global. Fieldwork: 18 November to 2 December 2025. | LangChain agent vendor, sells agents or agent tooling | Not defined | 57.3% Deployed or in production 57.3% of respondents have agents in production; another 30.4% are developing agents with concrete plans to deploy. Open, self-selected survey of a framework vendor's audience; 49% of respondents work at companies under 100 people. The prior edition reported 51%. | B 4 of 5 |
| MIT Sloan Management Review and BCG The Emerging Agentic Enterprise: How Leaders Must Navigate a New Age of AI Published 18 November 2025 | Not disclosed On the report page read. Mixed; Global. Fieldwork: not disclosed. | MIT Sloan Management Review and Boston Consulting Group consultancy | Not defined | 35% Any stage, piloting to production 35% of organizations have adopted agentic AI within two years; another 44% plan to deploy it soon. Global executive survey; the web report describes agentic AI as systems that "can plan, act, and learn on their own" but the page read gives no sample size. | C 1 of 5 |
| McKinsey & Company The state of AI in 2025: Agents, innovation, and transformation Published 5 November 2025 (archived copy) | n = 1,993 1,993 participants in 105 nations, all regions, industries and company sizes. Mixed; 105 nations. Fieldwork: 25 June to 29 July 2025. | McKinsey & Company consultancy | "systems based on foundation models capable of acting in the real world, planning and executing multiple steps in a workflow" | 23% Deployed or in production 23% are scaling an agentic AI system somewhere in the enterprise; 62% are at least experimenting with AI agents. Responses weighted by each nation's contribution to global GDP. McKinsey has since replaced this page with its 2026 edition; the 2025 figures were read from the Internet Archive capture of 2 January 2026. | A 5 of 5 |
| Gartner Gartner Survey Finds Just 15% of IT Application Leaders Are Considering, Piloting, or Deploying Fully Autonomous AI Agents Published 30 September 2025 | n = 360 360 IT application leaders at organizations with 250 or more employees. Enterprise; North America, Europe and Asia/Pacific. Fieldwork: May and June 2025. | Gartner analyst firm | fully autonomous AI agents: "goal driven AI tools that do not require human oversight" | 15% Other agent measure 15% of IT application leaders are considering, piloting or deploying fully autonomous AI agents. Counts only fully autonomous agents, a far stricter bar than most surveys, so it sits outside the adoption ranges. It shows what happens to the number when "agent" means no human in the loop. | B 4 of 5 |
| KPMG KPMG AI Quarterly Pulse Survey, Q3 2025 Published 18 September 2025 | n = 130 130 US C-suite and business leaders. Enterprise; United States. Fieldwork: not disclosed. | KPMG consultancy | Not defined | 42% Deployed or in production 42% of organizations have deployed at least some AI agents, up from 11% two quarters earlier. Organizations with $1 billion or more in annual revenue. | C 2 of 5 |
| Google Cloud The ROI of AI 2025 Published 4 September 2025 | Not disclosed On the page read (the full report is gated). Mixed; Not stated on the page read. Fieldwork: not disclosed. | Google Cloud agent vendor, sells agents or agent tooling | Not defined | 52% Deployed or in production 52% of executives report their organizations are deploying AI agents in production; 39% have deployed more than 10 agents. The public summary gives no sample size or dates; method details sit behind a form. | C 1 of 5 |
| Capgemini Research Institute Rise of agentic AI: How trust is the key to human-AI collaboration Published 16 July 2025 | n = 1,500 1,500 executives (director level and above) at organizations with over $1 billion in revenue. Enterprise; 14 countries. Fieldwork: April 2025. Screened sample. | Capgemini consultancy | "programs/platforms/software that are connected to the business environment with a defined boundary, make decisions autonomously, and act to achieve specific goals with or without human intervention" | 14% Deployed or in production 14% have deployed AI agents (2% at scale, 12% at partial scale); 23% have launched pilots and 61% are exploring. Every respondent organization had already started exploring agentic AI, so the sample is screened toward interest. The any-use figure (37%) adds deployed and piloting. Capgemini reconfirmed 900 of 1,500 responses. | A 5 of 5 |
| Stack Overflow 2025 Stack Overflow Developer Survey: AI Published July 2025 | n = 31,877 31,877 answered the AI agents question (49,009 survey responses from 177 countries). Developers; 177 countries. Fieldwork: 29 May to 23 June 2025. | Stack Overflow media | "autonomous software entities that can operate with minimal to no direct human intervention using artificial intelligence techniques" | 30.9% Workers or developers using agents 30.9% of developers use AI agents at work (14.1% daily, 9% weekly, 7.8% monthly or less); 37.9% do not and do not plan to. The 30.9% sums the three published yes answers. Respondents were recruited mainly through Stack Overflow's own channels. | A 5 of 5 |
| Boston Consulting Group AI at Work 2025: Momentum Builds, but Gaps Remain Published 26 June 2025 | n = 10,600 More than 10,600 leaders, managers and frontline employees in 11 countries and regions. Workers; 11 countries and regions. Fieldwork: not disclosed. | Boston Consulting Group consultancy | "smart digital assistants capable of learning, reasoning, and handling complex tasks independently" | 13% Workers or developers using agents 13% of employees see AI agents deeply integrated into their daily workflows. Only one third of employees said they understand how agents work. | B 3 of 5 |
| KPMG KPMG AI Quarterly Pulse Survey, Q2 2025 Published 26 June 2025 | n = 130 130 US C-suite and business leaders. Enterprise; United States. Fieldwork: not disclosed. | KPMG consultancy | Not defined | 33% Deployed or in production 33% of organizations have deployed at least some AI agents, up threefold after two quarters at 11%. Organizations with $1 billion or more in annual revenue. | C 2 of 5 |
| Gartner Gartner poll of webinar attendees on agentic AI investment (in the June 2025 agentic AI cancellation forecast) Published 25 June 2025 | n = 3,412 3,412 webinar attendees. Mixed; Not stated. Fieldwork: January 2025. | Gartner analyst firm | Not defined | 19% Other agent measure 19% say their organization has made significant investments in agentic AI and 42% conservative ones; 31% are waiting or unsure. Measures investment, not use, so it sits outside the adoption ranges. Webinar attendees are a self-selected, interested audience. The same release estimates only about 130 of thousands of agentic AI vendors are real. | B 3 of 5 |
| PwC PwC's AI Agent Survey Published 15 May 2025 | n = 308 308 US business executives (33% C-suite, 13% VP, 54% director). Mixed; United States. Fieldwork: 22 to 28 April 2025. | PwC consultancy | Not defined | 79% Any stage, piloting to production 79% of executives say AI agents are already being adopted in their companies; of adopters, 66% report measurable productivity value. "Adopted" is not defined; PwC itself notes that reports of full adoption often reflect excitement rather than evidence of transformation, and 68% say half or fewer of their employees interact with agents. | B 3 of 5 |
| EY EY Technology Pulse Poll: technology companies set the pace of agentic AI Published 14 May 2025 | n = 504 504 US business leaders in the technology industry at organizations with 5,000 or more employees. Enterprise; United States. Fieldwork: 9 to 21 April 2025. | EY consultancy | Not defined | 48% Any stage, piloting to production 48% of technology leaders are already adopting or fully deploying agentic AI. Fielded online by Atomik Research; margin of error plus or minus 4 points. Technology companies only. | B 4 of 5 |
| Microsoft 2025 Work Trend Index: The year the Frontier Firm is born Published 23 April 2025 | n = 31,000 31,000 knowledge workers in 31 markets, including 9,037 leaders. Workers; 31 markets. Fieldwork: 6 February to 24 March 2025. | Microsoft agent vendor, sells agents or agent tooling | Not defined | 46% Deployed or in production 46% of leaders say their companies are using agents to fully automate workflows or processes. Fielded by Edelman Data x Intelligence; the report prints the survey question under each chart. | A 5 of 5 |
| Cloudera The Future of Enterprise AI Agents Published 16 April 2025 | n = 1,500 Nearly 1,500 enterprise IT leaders in 14 countries. Enterprise; 14 countries. Fieldwork: not disclosed. | Cloudera agent vendor, sells agents or agent tooling | Not defined | 57% Deployed or in production 57% of enterprise IT leaders say they implemented AI agents in the past two years; 96% plan to expand their use. Sample size given as approximately 1,500. | C 2 of 5 |
| KPMG KPMG AI Quarterly Pulse Survey, Q1 2025 Published 16 April 2025 | n = 130 130 US C-suite and business leaders. Enterprise; United States. Fieldwork: not disclosed. | KPMG consultancy | Not defined | 11% Deployed or in production 11% of organizations are deploying AI agents (flat on the prior quarter); 65% are piloting them, up from 37%. Organizations with $1 billion or more in annual revenue. | C 2 of 5 |
| IBM AI agents in 2025: Expectations vs. reality (IBM and Morning Consult developer survey) Published 26 February 2025 | n = 1,000 1,000 developers building AI applications for enterprise. Developers; Not stated. Fieldwork: not disclosed. | IBM agent vendor, sells agents or agent tooling | Not defined | 99% Any stage, piloting to production 99% of developers building enterprise AI applications say they are exploring or developing AI agents. Reported in a quote on IBM's own page; "exploring" sets a very low bar, and the sample is developers already building AI applications. | B 3 of 5 |
Grade: A = all five transparency checks, B = three or four, C = two or fewer (rubric below). "Screened sample" means respondents had to be using, piloting or exploring agents to take part. Download every field as data.csv.
Why the numbers disagree
Here is every organization-level adoption figure in the index on one scale. The spread is not noise. Four things move it, and you can see each of them in the rows.
| Figure | Survey | What it counts | Sponsor | Agent defined | |
|---|---|---|---|---|---|
| 7% | Booz Allen Hamilton 21 July 2026 | Deployed or in production | consultancy | No | |
| 11% | KPMG 16 April 2025 | Deployed or in production | consultancy | No | |
| 14% | Capgemini Research Institute 16 July 2025 | Deployed or in production; screened sample | consultancy | Yes | |
| 23% | Deloitte 21 January 2026 | Deployed or in production | consultancy | No | |
| 23% | McKinsey & Company 5 November 2025 | Deployed or in production | consultancy | Yes | |
| 23.5% | The Modern Data Company 21 August 2026 | Deployed or in production | Sells agents | No | |
| 26% | KPMG 15 January 2026 | Deployed or in production | consultancy | No | |
| 33% | KPMG 26 June 2025 | Deployed or in production | consultancy | No | |
| 35% | MIT Sloan Management Review and BCG 18 November 2025 | Any stage, piloting to production | consultancy | No | |
| 42% | KPMG 18 September 2025 | Deployed or in production | consultancy | No | |
| 46% | Microsoft 23 April 2025 | Deployed or in production | Sells agents | No | |
| 48% | EY 14 May 2025 | Any stage, piloting to production | consultancy | No | |
| 51% | Cisco (study by Omdia) 23 September 2026 | Deployed or in production | Sells agents | No | |
| 52% | Google Cloud 4 September 2025 | Deployed or in production | Sells agents | No | |
| 53% | Agoda 29 September 2026 | Deployed or in production | other company | No | |
| 53% | KPMG 24 June 2026 | Deployed or in production | consultancy | No | |
| 54% | KPMG 31 March 2026 | Deployed or in production | consultancy | No | |
| 57% | G2 15 December 2025 | Deployed or in production | software marketplace | No | |
| 57% | Cloudera 16 April 2025 | Deployed or in production | Sells agents | No | |
| 57.3% | LangChain December 2025 | Deployed or in production | Sells agents | No | |
| 58% | Deloitte 12 August 2026 | Deployed or in production; screened sample | consultancy | No | |
| 62% | KPMG 24 September 2026 | Any stage, piloting to production | consultancy | No | |
| 79% | PwC 15 May 2025 | Any stage, piloting to production | consultancy | No | |
| 83% | Box 11 June 2026 | Deployed or in production | Sells agents | No | |
| 96% | OutSystems 13 April 2026 | Any stage, piloting to production | Sells agents | No | |
| 97% | Writer 7 April 2026 | Deployed or in production | Sells agents | No | |
| 99% | IBM 26 February 2025 | Any stage, piloting to production | Sells agents | No |
Gold bars: sponsor sells agents or agent tooling. Grey bars: consultancies, analysts, media and others. Rows in italics are screened samples. Each survey appears once, with its production figure when it reports one, else its any-stage figure.
What is counted. "Exploring or developing" agents (IBM's developer survey, 99%) and "using agents in some capacity" (OutSystems, 96%) set a bar almost everyone clears. "Scaling an agentic AI system" (McKinsey, 23%) or "agentic AI in production" at a federal agency (Booz Allen, 7%) set a bar most do not. A headline that says "use" can mean either.
Who is asked. KPMG asks leaders at companies with more than $1 billion in revenue. LangChain's survey is open to anyone in its developer community, and nearly half the respondents work at companies under 100 people. Capgemini, Deloitte's agentic survey and Temporal only admitted organizations or people already exploring, piloting or using agents, so their adoption shares describe the converted, not the market.
Who paid. Surveys from companies that sell agents or agent tooling report higher figures than surveys from consultancies, as the table below shows. That alone does not prove bias, since the samples differ too, but it is the pattern writers suspect and the data supports it.
Whether "agent" means anything. Most surveys never say. When one does and sets the bar high, the number falls hard: Gartner finds 15% of IT application leaders working with fully autonomous agents, Menlo finds 16% of enterprise deployments qualify as true agents, and VentureBeat's respondents say true agents make up the majority of what they run in only 10% of organizations. KPMG's own series shows the effect inside one survey: reported deployment fell from 42% to 26% in a quarter, and KPMG attributed part of the drop to leaders adopting stricter definitions of a true agent. For the vocabulary, see what an AI agent is and how an agent differs from a chatbot.
By sponsor and transparency
| Sponsor group | Surveys | Median adoption figure | Define "agent" | Grade A | State sample size |
|---|---|---|---|---|---|
| Sponsor sells agents or agent tooling | 11 | 57.2% (10 figures) | 1 of 11 | 2 | 10 of 11 |
| Consultancies | 17 | 35% (7 figures) | 3 of 17 | 3 | 16 of 17 |
| Analysts, media, investors and others | 7 | 55% (2 figures) | 4 of 7 | 3 | 5 of 7 |
| Check | Surveys passing | Share |
|---|---|---|
| Sample size stated | 31 of 35 | 89% |
| Fieldwork dates stated | 19 of 35 | 54% |
| Who was surveyed (role, company size or geography) | 34 of 35 | 97% |
| Defines "agent" or prints the question wording | 10 of 35 | 29% |
| Names the fieldwork firm or describes the method or its limits | 18 of 35 | 51% |
Worker and developer figures are kept apart from organization figures: they run from 30% to 52% across 3 surveys (Boston Consulting Group 30%, Writer 52%, Stack Overflow 30.9%).
Repeated surveys
A series that asks the same people the same question every quarter is the closest thing to a trend line this market has. Read it inside its own method, and watch for changes in the question.
| Published | Sample | Figure | What it counts |
|---|---|---|---|
| 16 April 2025 | n = 130 | 11% | Deployed or in production |
| 26 June 2025 | n = 130 | 33% | Deployed or in production |
| 18 September 2025 | n = 130 | 42% | Deployed or in production |
| 15 January 2026 | n = 130 | 26% | Deployed or in production |
| 31 March 2026 | n = 237 | 54% | Deployed or in production |
| 24 June 2026 | n = 204 | 53% | Deployed or in production |
| 24 September 2026 | n = 314 | 62% | Any stage, piloting to production |
| Published | Sample | Figure | What it counts |
|---|---|---|---|
| 26 June 2025 | n = 10,600 | 13% | Workers or developers using agents |
| 3 June 2026 | n = 11,749 | 30% | Workers or developers using agents |
KPMG's figures move fast because each wave is small (130 leaders a quarter in 2025, 204 to 314 in 2026) and the question changed in the third quarter of 2026, from deployment to "building, deploying or developing". The 62% is not comparable with the 54% before it.
How to quote a survey number
If you write about agent adoption, these six habits keep a number honest. Each takes one clause.
- Name the publisher and the year. "McKinsey's 2025 State of AI survey", not "a recent survey".
- Say what was counted. Production, piloting and exploring are different numbers; the table labels each one.
- Say who was asked. Sample size, role, company size, country. "308 US executives" tells a reader more than "79% of companies".
- Say who paid when the sponsor sells agents. One clause is enough: "a survey by Writer, which sells AI agents".
- Prefer surveys that define the term. Eight in this index do; their numbers mean something specific.
- Do not stitch series together. A 97% from one survey and a 23% from another are not a trend.
The same discipline applies to agent prices, which also hide in definitions; the AI agent pricing index and the pricing change log handle that side, and the AI agent incidents database logs what happens when agents in production go wrong.
Change log
One dated line for each batch of surveys added or corrected. Corrections name the row and what changed.
- : Index opened with 35 surveys published 26 February 2025 to 29 September 2026, each read on the publisher's own page (or the named report of it) and graded against the five checks.
Methodology
What counts. A survey or study published from 1 January 2025 onward that reports a share of organizations, workers or developers using, piloting, building or deploying AI agents or agentic AI. A few closely related measures, such as the share of deployments that are true agents, are logged as "other" rows: they appear in the table but not in the adoption ranges.
Sources. Every row was read on the publisher's own page or report. Where that page could not be opened, I used the release as reported by a named outlet and the row says so (Booz Allen via GovInfoSecurity, The Modern Data Company via IT Brief, Agoda via Hotel News Resource). McKinsey has replaced its 2025 page with the 2026 edition, so the 2025 figures were read from the Internet Archive copy, which the row links. Aggregator statistics pages were used only to find surveys; no figure was copied from them.
Classifications. "Sells agents" means the sponsor sells AI agents, an agent-building platform, or infrastructure marketed for agents. Consultancies sell agent implementation services, so they get their own group rather than counting as independent. "Screened" means respondents had to be using, piloting or exploring agents to take part. Measures: production covers deployed, in production, scaling or running; any stage covers using, adopting, building or piloting; worker figures are shares of workers or developers. Where a survey publishes exclusive stages (for example 7% in production and 51% piloting), the any-stage figure adds them, and the row says so.
Headline numbers. Ranges and medians use one figure per survey, the latest edition of each repeated series, and leave out screened samples, so a series such as KPMG's counts once. Nothing is re-weighted or adjusted; every percentage is the publisher's own.
The grading rubric
Each survey gets one point for each check it passes on the page I read. The grade is computed from the points, never assigned by hand.
- Sample size stated. A number of respondents, even an approximate one ("nearly 1,500").
- Fieldwork dates stated. At least the month the survey ran.
- Who was surveyed. Role, company size or geography.
- Defines "agent" or prints the question wording. A reader can tell what a yes meant.
- Method detail. Names the fieldwork firm, describes weighting or recruitment, or states the limits of the sample.
A = all five. B = three or four. C = two or fewer. A low grade means the public page leaves out what a writer needs to judge the number; the underlying survey may be sound, and a gated full report may hold the detail. When a publisher makes the detail public, the grade changes and the change log says so.
Not indexed, and why
- Forecasts. Gartner's prediction that 40% of enterprise apps will feature task-specific agents by the end of 2026, and similar projections, measure expectations, not use.
- Product telemetry. Salesforce's Agentic Enterprise Index tracks Agentforce customers' activity (agents per organization, work units), and the Anthropic Economic Index reports how Claude is used. Both are useful, neither is a share of organizations.
- Compilations. The Stanford AI Index gathers other organizations' surveys; the underlying surveys are indexed where they qualify.
- Surveys without an adoption share. Microsoft's 2026 Work Trend Index surveyed only workers who already use AI and reports no single agent-use share; IBM's June 2026 study of 2,000 technology executives reports expected growth in agents, not current adoption; SoundHound's June 2026 study sampled only organizations already running agents and reports return on investment.
- Pending verification. McKinsey's State of AI 2026, published in August 2026, will be added once its page can be read in full.
Refresh. Weekly. A script pulls candidate news stories every week; each candidate is checked against these rules on the publisher's own page before a row is added, and every number on this page is recomputed from the rows. Nothing in the table is edited by hand.
Known gaps. Coverage is English-language and skews to the US and to large enterprises, because that is where most published surveys come from. Small business surveys rarely ask about agents specifically yet. Gated reports are graded on their public pages.
FAQ
What percentage of companies use AI agents?
It depends on whose survey you read and what it counts as use. As of 4 October 2026, the 14 surveys in this index that report a deployed or in-production figure from an open sample run from 7% (Booz Allen, US federal agencies) to 97% (Writer, executives), with a median of 52.5%. The two largest surveys that publish a definition or their question wording both land at 23%: McKinsey found 23% of organizations scaling an agentic AI system in 2025, and Deloitte found 23% using agentic AI at least moderately.
Why do AI agent adoption statistics disagree so much?
Four things move the number: what is counted (exploring, piloting, or running in production), who is asked (large companies, developers, a vendor's own audience, or only people already using agents), who paid for the survey, and whether the survey defines an agent at all. Only 8 of the 35 surveys in this index define the term, and KPMG reported a fall from 42% to 26% between two quarters partly because leaders adopted stricter definitions of a true agent.
Is it true that 79% of companies use AI agents?
That figure comes from PwC's May 2025 survey of 308 US executives, 79% of whom said AI agents are already being adopted in their companies. The survey does not define adoption or agents, and PwC itself notes that 68% say half or fewer of their employees interact with agents. It is a fair measure of executive sentiment, not of agents running in production.
Are vendor surveys about AI agents biased?
They report higher numbers. Across the surveys in this index, those sponsored by companies that sell agents or agent tooling report a median adoption figure of 57.2%, against 35% for surveys by consultancies. That gap is not proof of bias, because the samples differ too, but only 1 of 11 vendor surveys defines what an agent is, so treat vendor figures as an upper bound and check who was asked.
Which AI agent adoption survey is most reliable?
For organization-level adoption, the surveys with large open samples, published fieldwork dates and a stated definition or question wording are McKinsey's State of AI 2025 (1,993 respondents, 23% scaling agentic AI) and Deloitte's State of AI in the Enterprise 2026 (3,235 leaders, 23% using agentic AI at least moderately). For developers, the Stack Overflow 2025 survey prints its definition and found 30.9% using agents at work.
What counts as an AI agent in these surveys?
Usually nothing is stated. Of the 35 surveys indexed, 8 define the term. McKinsey uses systems based on foundation models capable of acting in the real world, planning and executing multiple steps in a workflow; Stack Overflow uses autonomous software entities that can operate with minimal to no direct human intervention; Gartner's strictest version is goal driven AI tools that do not require human oversight, and only 15% of IT application leaders were working with those.
Can I use this data?
Yes. The table and the CSV are licensed CC BY 4.0: copy them, chart them, quote the numbers, and credit "Gravity AI agent adoption survey index" with a link to this page. The survey figures themselves belong to their publishers, so cite the original survey alongside this index.
How do I add a survey to the index?
Email support@gravity.fast with a link to the publisher's own page or report. I read every survey at its source and grade it against the same five checks before it goes in, and corrections are noted in the change log.
Sources
- Per-survey sources: linked in each table row and listed in full in data.csv, with publication and fieldwork dates. Each was read on the date in the row's added field.
- McKinsey & Company, "The state of AI in 2025: Agents, innovation, and transformation", 5 November 2025, read from the Internet Archive capture of 2 January 2026: web.archive.org
- Deloitte, "The State of AI in the Enterprise, 2026", 21 January 2026: deloitte.com
- KPMG, "AI at Scale: How 2025 Set the Stage for Agent-Driven Enterprise Reinvention in 2026" (Q4 2025 AI Pulse), 15 January 2026: kpmg.com
- Gartner, "Gartner Survey Finds Just 15% of IT Application Leaders Are Considering, Piloting, or Deploying Fully Autonomous AI Agents", 30 September 2025: gartner.com
- VentureBeat Research, "Where enterprise AI agent governance hasn't caught up", 24 July 2026: venturebeat.com
- Menlo Ventures, "2025: The State of Generative AI in the Enterprise", 9 December 2025: menlovc.com