Unisys survey finds AI agents stuck in cloud management pilots

Unisys said more than four in five senior IT and business leaders expect AI agents to help manage expanding cloud application portfolios, while only 23% have deployed the technology at scale. The gap matters for cloud and AI vendors because enterprise buyers are signaling demand for autonomous management tools, but security, governance and trust are still slowing production rollouts.

The findings come from a Unisys survey of 1,000 senior IT and business decision-makers in North America, Europe and Asia-Pacific. The company did not disclose spending levels, vendor preferences or how respondents defined scaled deployment, which limits how far the results can be read as a market forecast.

Unisys CEO and President Mike Thomson told CIO Dive that organizations adding cloud applications, platforms and services without a strategic AI roadmap face rising management complexity. He said agentic AI could help address that problem by improving visibility across cloud environments, identifying inefficiencies and taking on routine management work that would otherwise add pressure to IT teams.

Security is shaping how much autonomy buyers will allow

The survey points to a familiar constraint on enterprise AI adoption: buyers want automation, but they are cautious about letting systems act independently in infrastructure they do not fully understand. Unisys said 93% of respondents reported that cloud security capabilities affect how much autonomy they are willing to grant AI systems.

Nearly half of respondents said they had experienced a cybersecurity breach in the past year, up from 17% in the prior Unisys survey. That increase gives context to why governance is becoming a gating issue for AI agents in cloud operations, particularly when the systems are expected to observe, recommend or potentially execute changes across distributed environments.

Thomson told CIO Dive that visibility, security and trust are the main obstacles. He said many organizations still do not have a full view of where their data and AI workloads are located, while others are dealing with regulatory requirements and concerns about managing autonomous systems broadly.

Foundations are improving, outcomes are not keeping pace

Unisys also found that nearly all respondents said they have the architecture needed to support large-scale, data-driven decision-making, compared with 72% a year earlier. That suggests enterprises have spent on the technical base for AI, even if the returns remain uneven.

One sign of that mismatch: 65% said operational efficiency exceeded expectations, down from 80% in 2025. The survey does not identify why that measure declined, but it undercuts the easy vendor narrative that better AI architecture automatically produces better operating results.

Adrian Clamp, global head of consulting strategy and investment at KPMG International, told CIO Dive that the challenge has shifted from access to AI tools toward the organizational work needed to scale them. He said companies can often show results in individual projects but struggle to extend those benefits across the broader business because they must coordinate people, processes, technology, data, governance and investment decisions.

Clamp said organizations trying to move beyond pilots should strengthen data and cloud foundations before broadening AI deployments. He also pointed to governance, accountability, workforce readiness and measuring AI against business outcomes such as growth and customer experience, rather than productivity alone.

For infrastructure software companies, cloud service providers and consultancies, the Unisys data is useful but not surprising. Buyers are interested in AI agents for cloud operations because app sprawl creates real administrative load. The revenue opportunity will depend less on agent branding and more on whether vendors can prove control, auditability and security in environments where customers are already reporting more breaches.

This story draws on original reporting from CIO Dive.