Most IT Leaders Hesitant to Deploy Fully Autonomous AI Agents Without Human Control

Gartner's latest research, released Tuesday, shows that IT application leaders harbor substantial reservations about autonomous AI agents that operate without human intervention. The firm's survey encompassed 360 IT application leaders working at companies employing at least 250 full-time staff members.

The findings paint a picture of cautious adoption. Only 15% of IT application leaders are actively piloting, deploying, or evaluating fully autonomous AI agents that eliminate the need for human oversight. By contrast, a substantially larger cohort—75%—are already piloting or deploying some iteration of AI agents, suggesting that many organizations are exploring the technology in more controlled forms.

Trust deficits emerge as a central barrier. Fewer than 20% of respondents expressed confidence that their vendors could adequately safeguard against hallucinations, while just 13% felt their organization possessed sufficient governance frameworks to oversee AI agents. Security apprehensions loom large as well, with approximately three-quarters of leaders identifying AI agents as a potential new vulnerability within their infrastructure.

Vendor Instability Fuels Skepticism

Roughly one year into the industry's push toward AI agents, most enterprises remain reluctant to venture into uncharted territory. The stakes feel substantial, and confidence remains fragile.

The lack of trust is at least in part due to the speed in which AI technology is moving with many AI vendors repeatedly changing their branding, costing models and product offerings

Max Goss, senior director analyst at Gartner

The velocity of technological change weighs heavily on technology leaders' minds. Vendors are introducing new AI tools and agents faster than the corresponding governance and security infrastructure can be developed, a dynamic that undermines confidence in the market.

The fact that many vendors are releasing new AI tools and agents before the governance and security capabilities to protect them is not helping the trust story. Organizations are concerned that the vendors are prioritizing AI sales wins in the short term over providing scalable and trustworthy enterprise-grade AI tools that they can leverage over the longer term.

Max Goss

Deployment Patterns and Alignment Challenges

According to S&P Global Market Intelligence analysis from March, enterprises are predominantly channeling AI into IT operations, with customer experience workflows and marketing processes following as secondary priorities. However, success rates have deteriorated, with more organizations reporting AI project failures in the current year than in 2024.

Misalignment between IT departments, business units, and end users frequently hampers AI initiatives. Effective AI implementations demand consensus among IT, business stakeholders, and senior leadership regarding which organizational problems AI can address. Gartner's data indicates this alignment is rare, with only 14% of survey respondents confident their organization had reached such consensus.

Internal divisions have surfaced as organizations pursue AI adoption. According to a Writer report from March, most company leaders acknowledge that AI has created friction between IT teams and other business divisions, as well as between executives and employees. Resistance from staff continues to impede progress.

These internal tensions can obscure which use cases merit prioritization and how implementation should proceed. Organizations lacking clarity on their AI strategy frequently default to office productivity and digital workplace applications, even though these may not deliver the greatest business value.

Office productivity and the digital workplace are the default for those organizations that don't have a strong grasp on what they are doing with agents, but they are not necessarily the areas that will provide organizations with the most value

Max Goss

Recommendations for Moving Forward

Gartner advises organizations to continuously refine their use case strategies and reassess the tool providers they select. Major AI vendors are increasingly offering support for multiple models and embracing interoperability standards, reflecting a growing recognition that no single model or vendor family will satisfy all organizational requirements.

Best practice is to partner with the business to understand what business pain points the organization has, what AI tool or tools can address these, and to gain alignment on how to measure success, all the while making sure you have confidence in how the AI tool is processing and securing your data

Max Goss