TL;DR — Key Takeaways

  • AI agents are everywhere, but trust is in short supply. While 84% of surveyed organizations have deployed at least 10 operational agents, 78% still restrict what those agents can do.

  • Governance is struggling to keep pace. More than half cite safe guardrails as their biggest challenge, with security, inconsistent results and poor visibility into permissions adding to the risk.

  • Automation comes with a hefty price tag. More than two-thirds spend at least $10,000 monthly on AI agents, making cost controls and accountability essential as adoption accelerates.

A survey of 1,000 engineering and technology leaders finds that more than three-quarters (78%) maintain some restrictions on artificial intelligence (AI) agents deployed to manage IT operations by constraining them to read-only tasks, requiring human approval for changes, or limiting automated actions to noncritical workloads.

Conducted by Komodor, a provider of an IT operations platform based on AI agents, the survey also finds that 84% of respondents have deployed 10 or more operational agents, with 24% claiming some of their AI agents are fully autonomous.

Top use cases for those AI agents are automating toil (49%), support ticket triage (47%), self-remediation and site reliability engineering (SRE) (44%), incident investigation (43%) and setting up continuous integration/continuous delivery (CI/CD) pipelines (37%).

Overall, the survey identifies building safe guardrails (54%) as the top challenge IT teams are encountering in the agentic AI era, followed closely by governance and security (50%), standardization and accuracy (46%), insufficient visibility into agent permissions (43%) and lack of visibility into costs (26%) as major challenges.

More than two-thirds 68%) report spending at least $10,000 monthly on AI agents, including 38.4% spending $50,000 or more and 12% spending at least $250,000, the survey finds.

Udi Hofesh, technical product marketing and developer relations manager for Komodor, said the survey results make it apparent that while IT teams have broadly adopted AI agents, they are not yet to the point where they fully trust them enough to completely operate without human supervision.

Exactly what level of autonomy can be achieved, of course, remains to be seen. Thus far, at least, it’s become apparent that AI agents require a significant amount of supervision. In fact, in most instances, AI agents need to be managed much like a human employee, noted Hofesh.

Additionally, IT teams should also expect that they will be required by one or more compliance mandates to detail exactly how any given task was performed, especially as lawsuits involving AI agents become more commonplace, he added.

There is little doubt that AI agents will be a boon to IT but a significant amount of care should be taken before deploying them. AI agents, by definition, are probabilistic, so without the right harnesses and guardrails in place, they are not likely to perform the same tasks the same way twice. That’s a significant issue for IT operations teams where most tasks are deterministic in the sense that they need to be performed the same way each and every time. Even as confidence in the capabilities of AI agents grows, Murphy’s Law suggests that the one time an AI agent behaves in some anomalous fashion, it will be at the exact moment when the organization’s most mission-critical applications are affected.

At this juncture, it’s not so much a question of whether IT teams will deploy AI agents as it is defining the scope of the responsibilities assigned to them. However, as IT teams become more adept at managing AI agents, there is no reason to expect they won’t be relied on heavily to reduce a current level of toil that, for IT teams today, is already proving to be unsustainable without them.

Frequently Asked Questions

Why are enterprises reluctant to give IT operations agents full autonomy?
Because autonomous AI agents can behave unpredictably, particularly when handling critical infrastructure. According to Komodor's survey, 78% of organizations impose restrictions such as read-only access, human approvals or limitations on which workloads agents can modify.
What are the biggest challenges organizations face when scaling AI agents?
Building safe guardrails tops the list at 54%, followed by governance and security at 50%, standardization and accuracy at 46%, and insufficient visibility into agent permissions at 43%. These issues become increasingly difficult as organizations deploy more agents.
Will AI agents eventually replace human oversight in IT operations?
Not necessarily. AI agents can eliminate repetitive tasks, accelerate incident response and support self-remediation, but mission-critical operations require predictable outcomes. Human supervision, auditable actions and enforceable policies are likely to remain essential even as autonomy improves.