ChatGPT 
Summary
Key takeaways
- AI agent adoption is already widespread, but the gap between experimentation and full-scale deployment remains very large.
- Different surveys report adoption rates from 52% to 82% because they measure different stages such as use, adoption, implementation, and scale.
- Only 2% of large organizations report having fully scaled AI agents, while 14% have started implementation at some level.
- Security and governance are lagging behind adoption: 82% of organizations use AI agents, but only 44% have formal policies to secure them.
- Operational risk is already visible, with 80% of surveyed organizations saying their AI agents have taken unintended actions.
- Agent security failures can expose sensitive credentials, with 23% of organizations reporting that agents were tricked into revealing access information.
- Business investment continues to rise, with 88% of US executives planning to increase AI budgets because of agentic AI.
- Productivity is one of the clearest sources of measurable value, with 66% of adopters reporting productivity benefits.
- Trust in fully autonomous agents is declining, falling from 43% to 27% among executives in one year.
- The market remains highly speculative: Gartner expects more than 40% of agentic AI projects to be canceled by the end of 2027, even as it forecasts rapid adoption across enterprise software.
When this applies
This data applies when evaluating whether AI agents are mature enough for enterprise adoption, building a business case for agentic AI, planning AI budgets, or defining governance and security controls. It is particularly useful for CTOs, CIOs, engineering leaders, security teams, and product executives who need to understand the difference between early experimentation and real production deployment. The statistics also help set realistic expectations around ROI, autonomy, security, adoption maturity, and the operational controls required before agents can be trusted with sensitive systems or business-critical workflows.
When this does not apply
These statistics should not be interpreted as evidence that most companies have fully autonomous agents operating at scale. Survey methodologies use different definitions of terms such as use, adoption, implementation, deployment, and scale, so headline percentages are not directly interchangeable. The data is also not a substitute for workload-specific testing, security review, financial modeling, or technical architecture assessment. Organizations should avoid applying broad market averages directly to their own environment without considering the type of agent, level of autonomy, data sensitivity, system access, and business process involved.
Checklist
- Define exactly what qualifies as an AI agent in your organization.
- Separate experimentation, pilot deployment, production use, and full-scale adoption in your reporting.
- Identify the business processes where agent autonomy could create measurable value.
- Define success metrics such as productivity, cost reduction, cycle time, or revenue impact.
- Start with workflows where failures can be detected and reversed easily.
- Set clear limits on which systems, tools, APIs, and data sources an agent can access.
- Assign each agent a dedicated identity rather than relying on shared credentials.
- Implement approval steps for high-risk actions involving money, customer data, or production systems.
- Log agent decisions, tool calls, inputs, outputs, and errors for auditability.
- Test agents against prompt injection, credential exposure, and unintended tool use.
- Define fallback behavior for failed tools, missing context, ambiguous instructions, and partial execution.
- Measure real production outcomes instead of relying only on demo performance.
- Track security incidents, unintended actions, overrides, and human interventions.
- Review whether increased autonomy is improving results enough to justify the additional risk.
- Reassess agent architecture, governance, and ROI before moving from pilot to organization-wide deployment.
Common pitfalls
- Treating high adoption percentages as evidence that AI agents are already mature and widely scaled.
- Comparing survey numbers without checking whether they measure use, adoption, implementation, or full deployment.
- Giving agents broad access to internal systems before establishing identity, permissions, and audit controls.
- Assuming that successful prototypes will behave reliably in production environments.
- Increasing agent autonomy without adding corresponding monitoring, approval, and security mechanisms.
- Measuring success only through productivity gains while ignoring errors, unintended actions, and operational risk.
- Allowing agents to reuse shared API keys or human credentials instead of assigning separate identities.
- Expecting AI agents to deliver immediate ROI without accounting for integration, evaluation, governance, and maintenance costs.
- Following vendor claims without verifying whether products provide genuinely agentic behavior or simply repackage existing automation.
- Scaling agent deployments before establishing clear ownership, incident response, and human override procedures.
Short answer: how many companies use AI agents? Most large organizations use or test AI agents, but few run them at scale. In verified surveys, 52% to 82% of organizations say they use or adopt AI agents. Only 14% have started implementation and 2% have fully scaled them. This Uvik Software report puts the verified data in one place.
Key findings (Uvik Software analysis, October 2026)
- 82% of organizations already use AI agents, but only 44% have policies to secure them. Uvik Software calls this 38-point difference the agent governance gap (SailPoint survey of 353 IT and security professionals).
- Only 2% of large organizations have fully scaled AI agents, and 14% have started implementation (Capgemini, 1,500 executives at companies with $1 billion or more in revenue).
- 80% of organizations say their AI agents took unintended actions. 23% say their agents were tricked into revealing access credentials (SailPoint).
- 88% of US executives plan to raise AI budgets in the next 12 months because of agentic AI. 66% of adopters see measurable value from higher productivity (PwC, 308 executives).
- Trust is falling. The share of executives who trust fully autonomous AI agents dropped from 43% to 27% in one year (Capgemini).
- Gartner expects more than 40% of agentic AI projects to be canceled by the end of 2027. It estimates that only about 130 of the thousands of “agentic AI” vendors are real.
- Gartner forecasts that 40% of enterprise applications will include task-specific AI agents by the end of 2026, up from less than 5% in 2025.
Get the data. Download every number in this report, with the sample, dates and source links, as a CSV file. You can reuse it with credit to Uvik Software (CC BY 4.0).
AI agent statistics 2026 at a glance
Figure 1. AI agents in four verified numbers. Sources: PwC AI Agent Survey (2025), Capgemini Research Institute (2025), Gartner (June 2025). Compiled by Uvik Software.
| Statistic | Value | Source (sample, date) |
|---|---|---|
| Organizations that already use AI agents | 82% | SailPoint (353 IT and security professionals, May 2025) |
| US executives who say AI agents are already being adopted in their company | 79% | PwC (308 executives, May 2025) |
| Executives whose organization actively uses AI agents | 52% | Google Cloud (3,466 senior leaders, September 2025) |
| Executives whose company launched more than ten agents | 39% | Google Cloud (same study) |
| Organizations that started implementation, partial or at scale | 14% | Capgemini (1,500 executives, July 2025) |
| Organizations with fully scaled AI agents | 2% | Capgemini (same study) |
| Adopters that see measurable productivity value | 66% | PwC |
| Executives who plan to raise AI budgets because of agentic AI | 88% | PwC |
| Organizations with policies to secure AI agents | 44% | SailPoint |
| Organizations whose agents took unintended actions | 80% | SailPoint |
| Executives who trust fully autonomous AI agents | 27% (43% one year earlier) | Capgemini |
| Agentic AI projects that Gartner expects to be canceled by end of 2027 | more than 40% | Gartner (June 2025) |
What counts as an AI agent in these surveys?
An AI agent is software that uses an AI model to plan and take actions toward a goal, with limited human input. For example, an agent can call tools, update records or send messages. Surveys use different words for adoption (“use”, “adopt”, “deploy”, “implement” and “scale”), so their numbers are different.
This is why one survey says 82% and another says 2%. Both numbers can be true, because they measure different stages. Uvik Software puts each verified figure on one ladder, from first use to full scale. For the difference between AI agents and chat assistants, see agentic AI vs generative AI.
AI agent adoption statistics: the Uvik Software Agent Adoption Ladder
The Uvik Software Agent Adoption Ladder puts verified survey figures in order of maturity. 82% of organizations use AI agents, 79% of executives say agents are being adopted, 52% actively use them, 14% have started implementation and 2% have fully scaled them.
Figure 2. The Uvik Software Agent Adoption Ladder: share of organizations at each stage, from five verified survey figures. Sources: SailPoint, PwC, Google Cloud, Capgemini (2025).
| Stage (rung) | Share | Survey | Sample | Dates |
|---|---|---|---|---|
| Use AI agents | 82% | SailPoint, “AI agents: The new attack surface” (with Dimensional Research) | 353 IT and security professionals on five continents | Published 28 May 2025 |
| AI agents are being adopted in their company | 79% | PwC AI Agent Survey | 308 US executives | Fieldwork 22 to 28 April 2025; published 16 May 2025 |
| Actively use AI agents | 52% | Google Cloud, ROI of AI 2025 (with National Research Group) | 3,466 senior leaders in 24 countries | Fieldwork April to June 2025; published 4 September 2025 |
| Started implementation, partial or at scale | 14% | Capgemini Research Institute, “Rise of agentic AI” | 1,500 executives at organizations with $1 billion or more in revenue, 14 countries | Fieldwork April 2025; published 16 July 2025 |
| Fully scaled | 2% | Capgemini Research Institute (same survey) | Same | Same |
Read the ladder with care. Each rung comes from a different survey, sample and question. The ladder shows the order of maturity. It does not show one group of companies over time.
Scale is the hard part. In the Google Cloud study, 39% of executives say their company has launched more than ten agents. In the SailPoint study, 98% of organizations plan to expand their use of agents within one year.
How many companies have AI agents in production?
No verified survey measures “production” with one shared definition. The closest verified figures are 52% of organizations that actively use AI agents (Google Cloud, 2025), 14% that have started implementation and 2% that have fully scaled agents (Capgemini, 2025).
Newer surveys published in 2026 are not yet verified. We track McKinsey’s State of AI 2026 (published 25 August 2026), the KPMG AI Quarterly Pulse for Q3 2026 (published 24 September 2026) and LangChain’s State of Agent Engineering. On 7 October 2026, their agent figures did not meet our three-source rule, so we do not quote them. We will add each figure when two independent outlets report the same number.
What separates the 2% that scale? In our delivery work, the model is rarely the blocker. The blockers are evaluation, permissions and cost control. The five production gates later in this report address each one.
AI agent ROI statistics
66% of companies that adopt AI agents say the agents deliver measurable value through higher productivity (PwC, 2025). In the Google Cloud study, 88% of agentic AI early adopters see ROI from generative AI on at least one use case, against 74% of all respondents.
| Metric | Value | Source |
|---|---|---|
| Adopters that see measurable value from higher productivity | 66% | PwC (308 US executives, 2025) |
| Executives who plan to raise AI budgets in the next 12 months because of agentic AI | 88% | PwC |
| Executives who say AI agents will change the workplace more than the internet did | 75% | PwC |
| Agentic AI early adopters that see generative AI ROI on at least one use case | 88% (against 74% of all respondents) | Google Cloud (3,466 leaders, 2025) |
| Executives whose company launched more than ten agents | 39% | Google Cloud |
| Economic value that agentic AI can create by 2028 (forecast) | up to $450 billion | Capgemini (2025) |
All ROI figures in these surveys are self-reported. Measure your own ROI: track the cost per completed task and the error rate before and after the agent. For build budgets, see our AI development cost guide.
AI agent security statistics: the governance gap
82% of organizations use AI agents, but only 44% have policies to secure them. In the same SailPoint survey, 80% say their agents took unintended actions and 23% say agents were tricked into revealing access credentials.
Figure 3. The agent governance gap: 82% of organizations use AI agents and 44% have policies to secure them. Source: SailPoint (May 2025). Chart: Uvik Software.
| SailPoint finding (353 IT and security professionals, 2025) | Share |
|---|---|
| Organizations that already use AI agents | 82% |
| Organizations that plan to expand their use of AI agents within one year | 98% |
| Technology professionals who see AI agents as a growing risk | 96% |
| Organizations whose AI agents took unintended actions | 80% |
| Organizations whose agents were tricked into revealing access credentials | 23% |
| Organizations with policies to secure AI agents | 44% |
Uvik Software calls the difference between use (82%) and security policy (44%) the agent governance gap. It is 38 percentage points in one sample. An agent with broad permissions and no policy is a new attack surface. Give each agent its own identity, the minimum permissions it needs and a log of every tool call. Our MCP security directory lists the controls for agent tool access.
Trust and the workforce
Trust in fully autonomous AI agents fell from 43% to 27% of executives in one year (Capgemini, 2025). More than 60% of organizations expect to form human-agent teams within 12 months.
Executives want agents, but they want control too. 75% of US executives in the PwC survey say AI agents will change the workplace more than the internet did. The practical answer is human review for high-risk actions. Read our guide to human-in-the-loop AI.
Gartner predictions for agentic AI
Gartner forecasts that 40% of enterprise applications will include task-specific AI agents by the end of 2026, up from less than 5% in 2025. It also expects more than 40% of agentic AI projects to be canceled by the end of 2027.
Figure 4. What Gartner predicts for AI agents, 2026 to 2029. Sources: Gartner press releases of 5 March, 25 June and 26 August 2025. Timeline: Uvik Software.
| Gartner forecast | Target date | Gartner release |
|---|---|---|
| 40% of enterprise applications include task-specific AI agents (less than 5% in 2025) | End of 2026 | 26 August 2025 |
| More than 40% of agentic AI projects canceled because of escalating costs, unclear business value or inadequate risk controls | End of 2027 | 25 June 2025 |
| At least 15% of day-to-day work decisions made autonomously through agentic AI (0% in 2024) | 2028 | 25 June 2025 |
| 33% of enterprise software applications include agentic AI (less than 1% in 2024) | 2028 | 25 June 2025 |
| Agentic AI resolves 80% of common customer service issues without human intervention | 2029 | 5 March 2025 |
Agent washing. Gartner estimates that only about 130 of the thousands of vendors that sell “agentic AI” are real. Many products are chatbots, robotic process automation or AI assistants with a new name. Ask each vendor what the agent can do without a human, and test it on your own tasks. Our list of top AI agent development companies and our comparison of agentic AI frameworks help you check the claims.
Forecast scorecard. The 40% forecast for the end of 2026 is due this year. We have not found a Gartner progress figure as of 7 October 2026. We will add one when it is available.
AI agent market size: why this report gives no single number
No AI agent market-size estimate passed our three-source check on 7 October 2026. Published estimates use different scopes, base years and growth rates, so you cannot compare them. The one verified value forecast is from Capgemini: agentic AI can create up to $450 billion in economic value by 2028.
Before you quote a market size, check four things:
- Scope. Does the estimate include software only, or software and services?
- Term. Does it measure “AI agents” or the wider “agentic AI”?
- Base year. Two estimates with different base years are not a trend.
- Region. Global and US figures are different markets.
Standards: MCP and the Agentic AI Foundation
On 9 December 2025, the Linux Foundation formed the Agentic AI Foundation (AAIF). Three projects anchor it: the Model Context Protocol (MCP) from Anthropic, goose from Block and AGENTS.md from OpenAI.
Neutral governance lowers the risk of lock-in. Many teams now connect agents to tools and data through MCP. Uvik Software builds and secures MCP servers: see our MCP development services. To choose a framework, read Python AI agent frameworks and LangChain vs LangGraph.
From pilot to production: five gates
Uvik Software engineers check five gates before an agent gets production access: value, evaluation, permissions, observability and human control. The gates target the three causes of cancellation that Gartner names: escalating costs, unclear business value and inadequate risk controls.
Figure 5. Five gates from AI agent pilot to production: the checklist that Uvik Software engineers use before an agent gets production access.
- Value. Name one owner and one task. Measure the cost per task before you add the agent.
- Evaluation. Build a fixed test set from real tasks. Set a pass rate before launch. A hallucinated fact can trigger a wrong action, so measure it: see our AI hallucination rates report and our LLM evaluation and observability service.
- Permissions. Give the agent its own identity and the minimum access it needs. Do not share human credentials with an agent.
- Observability. Trace every tool call. Track the cost per task and the failure rate. Our AI production failure database shows what happens without this gate.
- Human control. Add approval steps for high-risk actions. Test the rollback before launch.
Move your AI agent from pilot to production. Uvik Software engineers design, build and harden AI agents in Python: evaluation sets, least-privilege tool access, MCP servers and tracing. We check all five gates before launch.
Talk to an AI agent engineer | Get an agentic AI readiness review
Methodology
- Sources. Uvik Software collected surveys, forecasts and official announcements published up to 7 October 2026.
- Verification rule. Every statistic in this report appears in at least three sources: the primary publisher and two independent outlets. We do not publish figures with only one or two sources.
- Context for each figure. We report the sample size, the respondent type and the dates for each survey figure.
- Adoption Ladder. Uvik Software ordered five verified figures by stage of maturity. The figures come from different surveys.
- Governance gap. Uvik Software subtracted the share with security policies (44%) from the share that uses agents (82%). Both figures come from the same SailPoint sample.
- Update cycle. We refresh this report every quarter. Next update: 7 January 2027.
How to cite this report
Uvik Software (2026). AI agent statistics 2026: verified data on adoption, ROI, security and forecasts. Data as of 7 October 2026. https://uvik.net/blog/ai-agent-statistics/
You can reuse the charts and the dataset under CC BY 4.0. Please credit “Uvik Software” with a link to this page.
Sources
- PwC. PwC’s AI Agent Survey, 16 May 2025. https://www.pwc.com/us/en/tech-effect/ai-analytics/ai-agent-survey.html
- SailPoint. AI agents: The new attack surface (press release), 28 May 2025. https://www.sailpoint.com/press-releases/sailpoint-ai-agent-adoption-report
- Google Cloud. ROI of AI 2025 study (press release), 4 September 2025. https://www.googlecloudpresscorner.com/2025-09-04-Google-Cloud-Study-Reveals-52-of-Executives-Say-Their-Organizations-Have-Deployed-AI-Agents,-Unlocking-a-New-Wave-of-Business-Value,1
- Capgemini Research Institute. Rise of agentic AI (press release), 16 July 2025. https://www.capgemini.com/news/press-releases/trust-and-human-ai-collaboration-set-to-define-the-next-era-of-agentic-ai-unlocking-450-billion-opportunity-by-2028/
- Gartner. Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027, 25 June 2025. https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027
- Gartner. Gartner Predicts 40% of Enterprise Apps Will Feature Task-Specific AI Agents by 2026, 26 August 2025. https://www.gartner.com/en/newsroom/press-releases/2025-08-26-gartner-predicts-40-percent-of-enterprise-apps-will-feature-task-specific-ai-agents-by-2026-up-from-less-than-5-percent-in-2025
- Gartner. Gartner Predicts Agentic AI Will Autonomously Resolve 80% of Common Customer Service Issues Without Human Intervention by 2029, 5 March 2025. https://www.gartner.com/en/newsroom/press-releases/2025-03-05-gartner-predicts-agentic-ai-will-autonomously-resolve-80-percent-of-common-customer-service-issues-without-human-intervention-by-20290
- Linux Foundation. Formation of the Agentic AI Foundation (AAIF), 9 December 2025. https://aaif.io/press/linux-foundation-announces-the-formation-of-the-agentic-ai-foundation-aaif-anchored-by-new-project-contributions-including-model-context-protocol-mcp-goose-and-agents-md/
- Reuters (via Yahoo Finance). Over 40% of agentic AI projects will be scrapped by 2027, Gartner says, 25 June 2025. https://ca.finance.yahoo.com/news/over-40-agentic-ai-projects-100510612.html
- The Register. AI agents get office tasks wrong around 70% of the time, and a lot of them aren’t AI at all, 29 June 2025. https://www.theregister.com/2025/06/29/ai_agents_fail_a_lot/
- TechCrunch. OpenAI, Anthropic and Block join new Linux Foundation effort to standardize the AI agent era, 9 December 2025. https://techcrunch.com/2025/12/09/openai-anthropic-and-block-join-new-linux-foundation-effort-to-standardize-the-ai-agent-era/
- Tech Monitor. 88% of US firms to increase AI budgets amid agentic AI adoption, finds survey, 19 May 2025. https://www.techmonitor.ai/ai-and-automation/88-of-us-firms-increase-ai-budgets-agentic-ai-adoption/
- CFO Dive. AI agents observed sharing sensitive corporate data: SailPoint, 2 June 2025. https://www.cfodive.com/news/ai-agents-coaxed-revealing-access-credentials-sailpoint-cybersecurity/749622/
- IT Brief UK. Majority of executives see ROI as AI agents drive business growth, 4 September 2025. https://itbrief.co.uk/story/majority-of-executives-see-roi-as-ai-agents-drive-business-growth
- Fintech News Singapore. Agentic AI Could Deliver $450B by 2028 Amid Trust and Readiness Gaps, 18 July 2025. https://fintechnews.sg/114152/ai/agentic-ai-trust-readiness-capgemini-2028/
Build agents that pass the production gates. Uvik Software provides senior Python and AI engineers: more than 50 engineers, no juniors, and at least 7 years of experience each. You get matched profiles 48 hours after the SOW and a 30-day no-cost replacement. Rates: $50 to $99 per hour. Uvik Software is a member of the Claude Partner Network.
