AI Agents in ITSM: Governance Is the Real Adoption Bottleneck

The headline finding

IT service management teams are no longer asking whether to use AI. They are asking how far to let it act on its own, and under what rules. A new survey of 256 ITSM professionals finds that one in four organizations already runs some level of autonomous AI in its service environment, and half plan to move further toward agentic AI within the next year 4. The most notable result is not about models or tooling. Asked what would most accelerate agentic adoption, 40% pointed to clear governance frameworks and best practices, ranking them ahead of technology improvements alone 4.

The broader picture is one of widespread but shallow use. Nearly three-quarters of respondents already use the AI features built into their ITSM platforms to some degree 1. Among those able to judge the results, 94% report efficiency gains 1. Agentic AI, meaning systems that take actions rather than just suggesting them, is still early. Data quality, governance concerns and skills gaps are the main brakes 1. On the benefits side, 27% say agentic AI lets them scale ITSM operations without a matching increase in headcount 4.

A paradox in the wider market

Those upbeat ITSM numbers sit uneasily next to a gloomier enterprise-wide narrative. Analysis of reports released in early July 2026 describes an "agent paradox." TEKsystems data shows 78% of organizations adopting AI while 74% fail to improve results 2. Gartner warns that more than 40% of agentic AI projects could be canceled by 2027 23. Forrester characterizes three-quarters of enterprises as "chasing" agents, with only a small fraction running them in production 2. That analysis lands where the ITSM survey does. The failure point is not model capability but governance, data access, and the gap between polished demos and messy real workflows 2.

Aggregated statistics show how fast the ground is moving. McKinsey reports 88% of organizations using AI regularly in at least one function. It also finds 62% experimenting with or scaling agents, though only 23% are actually scaling 3. KPMG found 54% of large US organizations actively deploying agents in early 2026, up from 11% a year before 3. IDC puts US enterprises with agents in production at 40% 3. Gartner expects task-specific agents in 40% of enterprise applications by the end of 2026, up from under 5% in 2025 3. Deloitte sizes the agentic AI market at roughly $8.5 billion in 2026, rising to about $35 billion by 2030 3.

Reconciling 94% success with 74% failure

The two sets of figures measure different things. The ITSM efficiency number is self-reported and limited to respondents able to rate their experience 1. It also mostly reflects embedded AI features rather than fully autonomous agents. The TEKsystems and Gartner figures speak to business outcomes and project survival across the whole enterprise 2. Read together, they suggest a pattern. Assistive AI inside existing tools delivers modest, real gains. Handing agents genuine authority is where projects stall.

ITSM may be better placed than most functions to cross that line. Service desks already run on ticketing systems, approval chains, change controls and audit trails. That is the governance scaffolding agents need. It may explain why governance shows up in the ITSM survey as both a barrier 1 and the leading accelerator 4. Teams understand that the rules, more than the model, determine what an agent can safely be allowed to do.

Vendors are selling the guardrails

The market is responding. Intellias has launched Agentic ServiceOps, a managed service that applies what it calls governed agentic AI across ITSM and IT operations management 5. The service adds an orchestration layer on top of systems clients already run rather than replacing them. Incoming work is enriched with identity, asset and service history at intake 5. An Intellias executive acknowledged the industry's credibility gap. Enterprises are told agentic AI will run their IT, he noted, but few have seen it in production. The company stresses that its agents operate inside guardrails the customer defines 5. Separately, a $120 million investment in Norm Ai has been read as a sign that compliance-focused agents could become core infrastructure for the wider market 2.

The takeaway

The evidence points one way. Agentic AI in IT operations is past the proof-of-concept stage, but its next phase depends on organizational controls more than on smarter models. Organizations that define what agents may touch, with what data and under whose oversight, are likely to land on the right side of Gartner's cancellation forecast. Those that deploy first and govern later risk joining the washout. For ITSM leaders, the practical priority is to treat governance frameworks as the product rather than paperwork.