Agentic AI Statistics 2026
By Axis Intelligence Research
Co-authors: Sarah Mitchell & Simplice Fosso | Last updated: October 30, 2026 | License: CC BY 4.0
Among large US enterprises in KPMG’s panel, 62% now report agents in deployment and 25% still pilot, which gives an Axis Pilot Graduation Index (APGI) of 71.3 out of 100 in Q3 2026, up 28.0 points from 43.3 a year earlier. The same wave shows the gap between sectors: Technology scores 93.9, Banking 65.3, Asset Management and Private Equity 46.9. Adoption figures for “agentic AI” range from 17% to 62% depending on how a source defines deployment, so the definition matters as much as the number.
Quick Answer
Agentic AI adoption among large US enterprises reached 62% deployed and 25% piloting in KPMG’s Q3 2026 wave (314 leaders, fielded July 24 to August 25, 2026). Gartner’s 2026 CIO survey puts deployment at 17% on a stricter definition, and McKinsey finds 40% of large-company respondents scaling agents in at least one function. Gartner also forecasts that over 40% of agentic AI projects will be canceled by the end of 2027 and that agents-and-assistants spending will reach $65.5 billion in 2027 (from $29.2 billion in 2026).
Key Findings (Axis Intelligence Research)
- The APGI reached 71.3 in Q3 2026, from 43.3 in Q3 2025 (+28.0 points), computed from KPMG’s deployed and piloting shares in the same wave.
- Pilot share fell 30 points in a year (55% to 25%) while deployed share rose 20 points (42% to 62%) in KPMG’s US large-enterprise panel.
- Technology graduates pilots about twice as fast as Asset Management and Private Equity: Q3 2026 APGI of 93.9 against 46.9 (a ratio of 2.0).
- Published “deployed” figures span 17% to 62% (a 3.6x spread), because sources count different rungs: production use, scaling in a function, or building and developing.
- Controls trail deployment: the average of five harness-layer controls in KPMG’s Q3 wave is 48.2%, and the share of leaders with controls built into agents alongside monitoring fell to 30% from 43% two quarters earlier.
What is the Axis Pilot Graduation Index (APGI)?
The Axis Pilot Graduation Index (APGI) measures how much of an organization’s agent activity has left the pilot stage.
Formula: APGI = deployed ÷ (deployed + piloting) × 100, using the deployed and piloting shares from the same survey and the same wave.
Why one survey: both inputs come from KPMG’s AI Quarterly Pulse Survey, so definitions, respondents and field dates match. An APGI of 100 means no piloting organizations remain; 50 means deployment and piloting are equal.
| Wave | Deployed | Piloting | APGI | Source |
|---|---|---|---|---|
| Q3 2025 | 42% | 55% | 43.3 | KPMG Q3 2026 deck (prior-year comparison) |
| Q1 2026 | 55% | 30% | 64.7 | KPMG Q2 2026 deck (restated Q1 column) |
| Q2 2026 | 53% | 35% | 60.2 | KPMG Q2 2026 deck |
| Q3 2026 | 62% | 25% | 71.3 | KPMG Q3 2026 deck |
Snapshot date: October 30, 2026. The Q4 2025 wave (26% deployed) is excluded because KPMG attributed that drop to leaders adopting a stricter definition of a true agent, and the piloting share is not carried in the deck we read.
How to read it. The index fell 4.5 points from Q1 to Q2 2026 (64.7 to 60.2) and rose 11.0 points from Q2 to Q3 (60.2 to 71.3). Quarterly moves are noisy because each wave surveys a different panel of 204 to 314 leaders; the year-over-year reading (+28.0) is the steadier signal.
How many companies use agentic AI in 2026?
It depends on the rung of the ladder a source measures. Axis Intelligence Research lines the published figures up by definition:
| Rung | Figure | Population | Source |
|---|---|---|---|
| Deployed to date | 17% | CIOs and technology executives, 2026 survey | Gartner |
| Using agentic AI at least moderately | 23% | 3,235 leaders, fielded Aug to Sep 2025 | Deloitte |
| Scaling agents in at least one function | 40% | Large-company respondents (revenue above $1 billion), 1,719 participants | McKinsey |
| Building, deploying or developing | 62% | 314 US leaders at $1 billion+ companies | KPMG |
| Plan to develop or deploy | 83% | 8,039 leaders, fielded Aug 2025 | Cisco |
The spread between the highest deployment-rung figure (KPMG, 62%) and the lowest (Gartner, 17%) is 3.6x. Populations and dates differ by source, so the ratio describes how far published figures diverge, not a like-for-like gap. Analyst commentary: when a vendor deck says “most enterprises run agents,” check which rung it is quoting before repeating it. Sources: Gartner, McKinsey, KPMG, Cisco.
Which industries move agents from pilot to production fastest?
KPMG’s Q3 2026 industry panels (100 US leaders each, revenue above $1 billion, fielded August 10 to September 14, 2026) allow a sector APGI that no other source publishes.
| Sector | Q2 2026 deployed / piloting | Q3 2026 deployed / piloting | APGI Q2 | APGI Q3 | Change |
|---|---|---|---|---|---|
| Technology | 67% / 31% | 93% / 6% | 68.4 | 93.9 | +25.6 |
| Banking | 39% / 51% | 49% / 26% | 43.3 | 65.3 | +22.0 |
| Asset Management and PE | 19% / 52% | 30% / 34% | 26.8 | 46.9 | +20.1 |
Source: KPMG industry decks (Technology, Banking) and release text (Asset Management and PE). Axis calculation for APGI and change.
Technology’s Q3 APGI is 2.0 times that of Asset Management and Private Equity. All three sectors gained 20 points or more in one quarter, but the order did not change.
Are enterprises scaling agents or stuck in pilots?
McKinsey’s State of AI 2026 (fielded May 4 to June 8, 2026) finds 40% of large-company respondents scaling AI agents in at least one function, up from 27% a year earlier, while respondents at smaller organizations stayed flat at 22%. About 31% of large-company respondents scale software and coding agents. Only 6% qualify as AI high performers (at least 5% of EBIT attributed to AI), unchanged from 2025, and 37% attribute at least some EBIT impact to AI. Source: McKinsey.
Deloitte’s October 29, 2026 release (3,500 director-to-C-suite leaders, 25 countries, fielded May to June 2026) reports 28% seeing revenue gains from AI, against 20% in its January report, and 81% reporting faster decision-making. Source: Deloitte.
Analyst commentary: the APGI says pilots are being cleared; McKinsey and Deloitte say financial results lag. Those facts fit together: graduation to production is the first gate, measurable P&L impact the second.
Multi-agent systems and workforce adoption
KPMG reports 25% of organizations actively developing or implementing multi-agent systems in Q3 2026, against 6% in both Q1 and Q2. Orchestration of multiple agents stood at 9% in Q1 and 18% in Q2. Significant workforce adoption reached 44% in Q3 2026, from 23% in Q2 and 10% in Q3 2025. Source: KPMG Q3 2026.
Gartner expects one-third of agentic AI implementations to combine agents with different skills by 2027. Source: Gartner.
How much do AI agents cost to run, and who is watching the bill?
Inference and token spend now sits in the approval process. KPMG finds cost review inside AI approval at 74% in Q3 2026 (61% in Q2) and usage or token budgets at 43% (36% in Q2); in Q2 only 26% of leaders said AI operating costs were fully visible. Deloitte reports 71% of respondents track total inference and API spend, 51% token volume and 44% per-unit inference cost. McKinsey finds about one in five respondents say operating cost constrains AI use. Sources: KPMG, Deloitte, McKinsey.
Cisco’s 2025 index lists high compute costs as the top hurdle to ROI for 54% of respondents. That 54% is a compute-cost figure, not an infrastructure-scalability one; the scalability figure is 34% of respondents with fully adaptable IT infrastructure. Source: Cisco.
Governance and security: do controls keep pace with autonomy?
| Control (KPMG, Q3 2026, share of leaders) | Share |
|---|---|
| Data access controls in the AI harness layer | 59% |
| Security and identity controls | 53% |
| Monitoring of AI outputs | 47% |
| Policy and guardrail enforcement | 42% |
| Model access management | 40% |
| Mean of the five (Axis calculation) | 48.2% |
Other KPMG Q3 2026 figures: 49% of leaders define high-risk use cases where autonomous agent decisions are not allowed (28% a year earlier); 70% run AI monitoring dashboards; 73% are confident in their governance capability (57% in Q2). Source: KPMG.
Deloitte finds 40% of respondents allow AI to execute without human approval and 23% say accountability is unclear when outcomes are negative. Cisco’s 2025 index shows 24% of all respondents have guardrails and live monitoring for agent actions, against 84% of its top-tier “Pacesetters” (3.5 times higher). Sources: Deloitte, Cisco.
Gartner’s May 26, 2026 release predicts 40% of enterprises will demote or decommission autonomous agents by 2027 because governance gaps surface only after production incidents. Source: Gartner.
What will agentic AI look like in 2027?
- Cancellations: Gartner (June 2025) predicts over 40% of agentic AI projects will be canceled by the end of 2027, citing escalating costs, unclear business value and inadequate risk controls. Gartner
- Spending: Gartner’s September 16, 2026 forecast puts worldwide AI spending at $1.79 trillion in 2025, $2.67 trillion in 2026 (49.5% growth, headline rounded to $2.7 trillion) and $3.64 trillion in 2027. AI agents and assistants spending: $16.5 billion (2025), $29.2 billion (2026), $65.5 billion (2027). Axis calculation: agents and assistants are 1.1% of 2026 AI spending, while AI software is 17.3%. Gartner
- Agent software scope: Gartner’s May 5, 2026 release forecasts AI agent software spending of $206.5 billion in 2026 and $376.3 billion in 2027, from $86.4 billion in 2025. This is a wider category than the “AI agents and assistants” line ($29.2 billion in 2026) in the September table, so the two figures are not interchangeable. Gartner
- Application software: Gartner estimates up to $234 billion of enterprise application software spend is exposed to agentic arbitrage through 2030, roughly 20% of enterprise application SaaS spend in 2030. Gartner
- Embedded agents: 40% of enterprise applications will feature task-specific agents by the end of 2026, from under 5% in 2025. Gartner
- CEO view: IBM’s 2026 CEO study (2,000 CEOs, fielded February to April 2026) finds 10% say agentic AI is primarily driving growth today and 72% expect it to by 2030; CEOs estimate 25% of operational decisions are made by AI without human intervention now and 48% by 2030. IBM
- Regulation: under Regulation (EU) 2026/1744 (the Digital Omnibus on AI, in force July 27, 2026), stand-alone high-risk AI obligations under Annex III apply from December 2, 2027, and those for AI in Annex I products from August 2, 2028. EUR-Lex
Analyst commentary: Gartner’s 40% cancellation call and the 71.3 APGI describe different things. One is project survival through 2027; the other is today’s pilot clearance. A high APGI does not guarantee survival, which is why we will track both.
Is there an agent-washing problem?
Gartner estimated in June 2025 that about 130 of the thousands of vendors claiming agentic capabilities offer genuine agentic AI. Gartner’s April 2026 Hype Cycle places agentic AI at the Peak of Inflated Expectations. Sources: Gartner, June 2025, Gartner, Hype Cycle.
Methodology
Sources. Eight primary organizations: KPMG, McKinsey & Company, Deloitte, IBM Institute for Business Value, Cisco, Gartner and the European Parliament and Council. Every figure in this article is a row in the CSV with source organization, document, URL and retrieval date (October 30, 2026).
Axis calculations. APGI, its quarterly and year-over-year changes, the sector APGI, the Technology-to-Asset-Management ratio (2.0), the definition spread (62 ÷ 17 = 3.6), the harness mean (48.2%) and the Gartner spend shares (1.1% and 17.3%) are computed by Axis Intelligence Research from published inputs. Percentages are reported as the sources print them; Axis calculations use unrounded ratios for changes.
Comparability. Survey populations, years and definitions differ across sources. We therefore never mix inputs from different surveys inside one index, and we label every cross-source comparison as a spread, not a gap.
Verification. Each prose figure was checked against its CSV row and each source was retrieved on October 30, 2026. The Gartner releases of September 16, 2026 and June 25, 2025 and the authentic text of Regulation (EU) 2026/1744 were read in full from saved copies of the source pages.
About this dataset
agentic-ai-statistics.csv (229 rows, UTF-8, tidy long format, CC BY 4.0). Columns: row_id, metric, value, unit, period_label, as_of_date, geography, segment, source_org, source_document, source_url, retrieved_date, is_primary, axis_calculated, method_note. Missing values are NA; dates follow ISO 8601.
How to cite
- APA: Axis Intelligence Research. (2026, October 30). Agentic AI statistics 2026: How many pilots reach production? Axis Intelligence. https://axis-intelligence.com/agentic-ai-statistics/
- MLA: Axis Intelligence Research. “Agentic AI Statistics 2026: How Many Pilots Reach Production?” Axis Intelligence, 30 Oct. 2026, axis-intelligence.com/agentic-ai-statistics/.
- Chicago: Axis Intelligence Research. “Agentic AI Statistics 2026: How Many Pilots Reach Production?” Axis Intelligence. October 30, 2026. https://axis-intelligence.com/agentic-ai-statistics/.
To cite the index, name it “Axis Pilot Graduation Index (APGI)” and give the wave and date.
Frequently Asked Questions
What share of large US companies have moved AI agents out of pilot?
In KPMG’s Q3 2026 wave, 62% report deployment and 25% still pilot, an APGI of 71.3. A year earlier the split was 42% and 55% (APGI 43.3).
Why do agentic AI adoption numbers range from 17% to 62%?
Each source counts a different rung: Gartner counts agents deployed to date, McKinsey counts scaling in at least one function, KPMG counts building, deploying or developing. Compare figures only within one definition.
Which sector is furthest ahead on production agents?
Technology, with a Q3 2026 APGI of 93.9 (93% deployed, 6% piloting), ahead of Banking (65.3) and Asset Management and Private Equity (46.9).
Are agentic AI projects likely to be canceled by 2027?
Gartner predicts over 40% will be canceled by the end of 2027, citing cost, unclear business value and weak risk controls. The same firm expects 40% of enterprise applications to embed task-specific agents by the end of 2026.
How much will companies spend on AI agents in 2027?
Gartner forecasts $65.5 billion on AI agents and assistants in 2027, up from $29.2 billion in 2026, inside a $3.64 trillion worldwide AI spending forecast.
Do enterprises track what an agent costs per task?
Partly. Deloitte finds 71% track total inference and API spend but 44% track per-unit inference cost; KPMG finds 43% run usage or token budgets.
Which controls do enterprises build around agents first?
Data access controls (59%) and security and identity controls (53%) lead KPMG’s harness-layer list; policy and guardrail enforcement (42%) and model access management (40%) trail.
Do autonomous agents need human approval under the EU AI Act?
The Act ties obligations to risk class, not autonomy alone. Under the Digital Omnibus on AI, Annex III high-risk obligations apply from December 2, 2027 and Annex I product obligations from August 2, 2028.
How does the APGI differ from a plain adoption rate?
An adoption rate counts organizations at one rung. The APGI measures the conversion between two rungs, deployed versus piloting, within one survey wave.
When is the next APGI reading?
When KPMG publishes the Q4 2026 AI Quarterly Pulse, we add the wave to the CSV and update this page.
Correction note
Correction, October 30, 2026: This page was rebuilt. The earlier Agentic Deployment Readiness Index (ADRI) is retired because it combined inputs from different surveys; the Axis Pilot Graduation Index (APGI) replaces it using a single survey. Outdated McKinsey, IBM and Gartner figures were replaced with their 2026 releases. A Cisco figure of 54% previously described as infrastructure that cannot scale is a high-compute-cost figure and is now labeled correctly. An unsupported agent-washing percentage was removed; the Gartner estimate of about 130 genuine vendors is cited instead.