AI Agent Adoption Statistics (2026): Forecasts, Failure Rates and Security

AI agent adoption is moving fast on paper and unevenly in practice. Gartner forecasts that up to 40% of enterprise applications will include task-specific AI agents by 2026, up from less than 5% in 2025, and in the same year warned that over 40% of agentic AI projects will be cancelled by the end of 2027. This page collects the figures we could trace to a source, and says where we could not.

AI agent adoption statistics at a glance

The enterprise forecasts

Gartner expects up to 40% of enterprise applications to include integrated task-specific agents by 2026, up from less than 5% in 2025. The same release describes a longer-range best case in which agentic AI drives about 30% of enterprise application software revenue by 2035.

Note the “up to”: some outlets drop the qualifier, but it is part of the forecast.

The failure forecast

Gartner’s other headline number is the counterweight. In June 2025 it predicted that over 40% of agentic AI projects will be cancelled by the end of 2027, citing escalating costs, unclear business value and inadequate risk controls.

Coverage of that release also describes “agent washing”, where existing assistants, RPA tools and chatbots are rebranded as agentic. Gartner estimated that only about 130 of the thousands of agentic AI vendors are real.

Read together, the two forecasts describe a technology that is being embedded quickly and abandoned frequently. We would treat any single adoption percentage with that in mind.

What developers are choosing: GitHub stars

Enterprise forecasts describe procurement. GitHub shows what individual developers bookmark. We checked three popular self-hosted projects on 6 November 2026:

GitHub rounds these to 0.1k. A star is a bookmark, not an installation, so this shows where developer attention sits rather than how many people run each tool. It is still a useful leading indicator of documentation, community answers and ecosystem, which affect whether self-hosting works out.

For a second angle on why one metric can mislead, see our self-hosting statistics, where stars and Docker pulls rank projects very differently. If you are choosing between projects, our OpenClaw alternatives comparison covers the published requirements side by side.

The security statistics

Adoption figures rarely include failures, and in this category the failures are concrete.

CVE-2026-25253

CVE-2026-25253 is a one-click remote code execution flaw in OpenClaw rated CVSS 8.8, fixed in version 2026.1.29 on 30 January 2026. The Hacker News reported that a single visit to a malicious web page could trigger code execution, even on instances listening only on loopback.

Exposure counts vary by researcher. SecurityScorecard identified 28,663 unique IP addresses with exposed OpenClaw control panels, of which 12,812 were vulnerable to remote code execution. Censys reported 21,639 publicly exposed instances in a 31 January scan.

ClawHavoc

Koi Security audited all 2,857 skills on ClawHub and found 341 malicious, 335 of them tied to a single campaign it named ClawHavoc. The campaign delivered the Atomic macOS Stealer. Koi later reported its count rising to 824 as the marketplace grew past 10,700 skills, and Antiy CERT counts at least 1,184 over time.

The pattern is a category whose adoption is outrunning operational practice: popular, privileged software, installed by many people for the first time.

What we could not verify

We removed several figures from earlier versions of this page because we could not trace them to a primary source: enterprise-in-production percentages, sector-by-sector adoption rates and market-size estimates. We also removed our hosting cost estimates, which move too often to quote without re-checking each provider.

Limitations of what remains:

• The Gartner figures come from its press releases and reporting of them. We did not read the full research notes, which are paywalled.
• The June 2025 cancellation forecast is cited through news coverage of the release.
• Exposure and malicious-skill counts differ between researchers because they scanned at different times and by different methods.
• GitHub stars measure attention, not use.

FAQ

How fast is AI agent adoption growing?

Gartner forecasts up to 40% of enterprise applications will include task-specific agents by 2026, up from less than 5% in 2025. It also warns that over 40% of agentic AI projects may be cancelled by the end of 2027.

What percentage of AI agent projects fail?

Gartner predicts more than 40% of agentic AI projects will be cancelled by the end of 2027, mainly due to escalating costs, unclear business value or inadequate risk controls.

Which self-hosted AI agent is most popular?

By GitHub stars on 6 November 2026, OpenClaw leads the projects we checked with 394.0k, ahead of n8n (209.7k) and Dify (160.6k).

Do GitHub stars mean people use these tools?

No. A star is a free bookmark and implies no installation. It indicates where developer attention is concentrated.

How many OpenClaw instances were exposed?

Counts differ by researcher: SecurityScorecard found 28,663 exposed IP addresses (12,812 vulnerable to remote code execution) and Censys found 21,639 in a 31 January 2026 scan.

What is agent washing?

Agent washing is rebranding existing assistants, RPA tools or chatbots as agentic without real agentic capabilities. Gartner estimated only about 130 of the thousands of agentic AI vendors are genuine.