A global survey of 1,200 business decision-makers, IT leaders, and technical specialists finds that while 90% of organizations want to use artificial intelligence (AI) agents to drive autonomous IT operations, more than three-quarters (77%) are currently hesitant to allow AI to make operational decisions without human approval. 

Conducted by the market research firm Coleman Parkes Research on behalf of Riverbed, the survey finds 91% agree that AI investments have met or exceeded their expectations, with 76% reporting that agentic AI is central to their future IT strategy. A full 90% expect to achieve human-supervised or highly autonomous IT operations within the next two years, but for now only 24% extensively trust AI recommendations with limited human review.

Nearly all respondents (93%) specifically expect AI-powered operational intelligence to transform frontline IT and service desk capabilities within two years. By 2028, 27% predict autonomous operations will significantly reduce service desk activity, while only 22% expect agentic AI will enable teams to focus on more complex issues. Just over a third (35%) expect service desk roles to evolve from reactive support to proactive prevention.

Security and compliance concerns (55%), risk of operational disruption (45%), lack of trust in AI decisions (39%), skills/expertise gaps (37%), disconnected tools or vendors (35%), organizational resistance (34%), and poor data quality (30%) are the top obstacles identified and 34% said their IT operations teams are extremely effective at using AI to identify and resolve issues before employees notice a disruption. More than half (51%) said they are concerned about agentic performance and reliability.

A total of 92% also expect AI observability and governance will become a critical new IT domain to manage. However, while 96% said multi-domain visibility is important for AI-driven IT operations, only 17% have fully unified visibility across networks, applications, endpoints and cloud. Meanwhile, only 41% of respondents said their IT environments are fully prepared for AI. Even less (19%) said IT operations are currently automated. Less than a quarter rated their data granularity (23%) and data quality (21%) as excellent.

Jim Gargan, chief marketing officer at Riverbed, said that despite high levels of confidence, the survey suggests that many organizations might be underestimating what will clearly be a multi-year journey to achieve autonomous IT operations. In fact, only 8% report they have applied AI to operations enterprise-wide, with only 9% of AI initiatives fully deployed across their organizations.

Exactly how automated IT operations may become in the age of AI, of course, remains to be seen, but despite stated goals, the survey also finds 69% of respondents said they favor human-supervised AI. At present, only 42% of respondents said they are highly confident in their organization’s AI governance capabilities.

Most IT teams today still span multiple silos that reduce overall visibility, noted Gargan. It will be critical to eliminate that current level of friction to achieve autonomous IT operations, he added. In fact, AI agent performance and reliability (51%), employee AI adoption and utilization (49%), AI costs and resource consumption (44%), AI governance/compliance activities (38%), AI workflow execution (30%), and shadow AI usage (21%) are all monitoring challenges, the survey finds.

In all, 93% said their existing AI investments have met or exceeded their expectations, with 18% reporting that AI significantly exceeded their expectations. Nearly two thirds (63%) said they have, as a result, increased investments. Primary reasons cited for making those investments are operational productivity (86%), accelerating innovation (82%), improving employee digital experience (77%) and reducing service-desk headcount (69%).

Obviously, there is much work to be done before most of those goals can be achieved, but at this juncture it may be more of a question of time and patience than any fundamental inability to one day safely apply AI at much higher levels of scale.