Tech chiefs turn to AI agents to rein in cloud sprawl

AI agents could become a practical answer to cloud sprawl

Cloud environments have become harder to manage. Large enterprises now operate applications, data, and infrastructure across multiple cloud platforms, regions, and business units. That complexity creates a basic management problem: IT teams need more visibility and control, but continually adding people is expensive and difficult to scale.

AI agents offer another option. These systems can monitor cloud environments, identify inefficiencies, and perform routine management tasks with less human intervention. According to Thomson, AI agents could help IT teams oversee increasingly complex cloud estates without requiring headcount to grow at the same rate. The objective is not simply automation. It is giving IT teams the capacity to manage more infrastructure while keeping costs and operational complexity under control.

Security determines how far companies can take this approach. The Unisys survey found that 93% of respondents said cloud security capabilities influence how much autonomy they are willing to give AI systems. That is an important signal for executives. Enterprises may be interested in autonomous AI, but autonomy depends on confidence in the systems surrounding it.

This changes the executive conversation. The relevant question is no longer just whether AI agents can perform cloud-management tasks. Leaders need to decide which actions agents can take independently, which require human approval, and how every consequential action will be monitored and audited. Strong cloud architecture, accurate data, clear access controls, and reliable monitoring become prerequisites for greater autonomy.

The opportunity is substantial. Companies that establish these foundations can potentially manage larger and more dynamic cloud environments without matching every increase in complexity with additional staffing. But deploying agents before achieving sufficient visibility and control can create new operational risks instead of removing existing ones.

Security, visibility, and trust will determine the speed of agentic AI adoption

The strongest constraint on agentic AI is not access to the technology. It is whether organizations can trust autonomous systems to operate inside critical business environments. Thomson identified three major barriers: visibility, security, and trust. Many organizations still lack a complete view of where their data and AI workloads reside, while regulatory requirements add another layer of difficulty.

This matters because AI agents can do more than generate information. Depending on their permissions, they can interact with systems and take actions on behalf of users or applications. Increasing that autonomy can increase business value, but it also raises the importance of identity management, access permissions, monitoring, audit records, and clear limits on what an agent is allowed to do.

The security environment makes caution rational. Nearly half of respondents in the cited survey said their organization had experienced a cybersecurity breach during the previous year, compared with 17% in the prior survey. At the same time, 93% said cloud security capabilities affect how much autonomy they are prepared to grant AI systems. Taken together, the findings show that AI autonomy and cybersecurity strategy are becoming closely connected.

For C-suite leaders, governance therefore needs to arrive before broad deployment. Management should know where agents operate, what data they can access, which decisions they can make, and who is accountable when something goes wrong. Higher-risk actions may require human approval, while lower-risk and reversible tasks can support greater automation.

The measured deployment approach makes sense in this context. Organizations are concentrating first on governance, operational readiness, and high-value use cases before expanding agentic AI across the enterprise. This is a way to establish the controls required to increase autonomy with confidence.

Companies that solve these issues early will have more room to move. Better visibility creates stronger control. Stronger security supports greater autonomy. And greater autonomy, deployed within clear limits, can expand the economic value of AI across increasingly complex cloud operations.

The hard part is moving AI from successful pilots to enterprise scale

AI access is no longer the main constraint for many companies. The bigger challenge is turning successful experiments into systems that work reliably across the enterprise. A pilot can demonstrate technical capability with a limited team, controlled data, and a narrow objective. Enterprise deployment must work across departments, processes, technology platforms, and governance structures.

Adrian Clamp, global head of consulting strategy and investment at KPMG International, described this problem clearly: “Many organizations can demonstrate success in individual projects, but struggle to extend those benefits across the wider business.” He said the larger challenge is coordinating people, processes, technology, data, governance, and investment decisions throughout the organization.

This explains why companies are taking a measured approach to agentic AI. According to Thomson, the adoption gap reflects a shift from experimentation toward implementation. Organizations are prioritizing governance, operational readiness, and high-value use cases before expanding AI agents across the enterprise.

For executives, this means AI strategy needs to move beyond the number of pilots launched or tools purchased. A successful deployment needs defined business ownership, dependable data, appropriate controls, integration with existing processes, and clear accountability. It also needs enough economic value to justify continued investment.

The choice of use case becomes especially important with AI agents because these systems can take actions rather than simply provide information. Companies can start with processes where outcomes are measurable and risks are understood. They can then increase deployment and autonomy as performance, security, and governance improve.

The next stage of enterprise AI will therefore depend less on proving that the technology works and more on proving that organizations can operate it consistently at scale. Companies that coordinate business strategy, technology, governance, and investment effectively will be in a stronger position to convert AI capabilities into repeatable business results.

Better AI infrastructure does not guarantee better business results

Enterprise technology foundations are improving quickly. The Unisys survey found that nearly all respondents said they now have the architecture required for large-scale, data-driven decision-making, up from 72% a year earlier. That is significant progress in technical readiness.

Business outcomes tell a more complicated story. Only 65% of respondents said operational efficiency exceeded expectations, down from 80% in 2025. Companies have strengthened their technical foundations, yet reported results have not improved at the same pace.

This gap matters. Infrastructure gives an organization the ability to deploy AI, but capability alone does not create business value. Companies still need to decide where AI should be used, redesign processes where necessary, establish responsibility for outcomes, and measure whether deployment actually improves performance.

Adrian Clamp of KPMG International emphasized this distinction. “The greatest benefits come when organizations rethink workflows and decision-making processes around AI,” he said. In practice, this means executives should avoid treating AI primarily as another technology installation. Existing processes may need to change if the company expects AI to materially affect costs, revenue, customer experience, or decision quality.

Measurement also needs to evolve. Productivity is useful, but it captures only part of AI’s potential impact. Clamp recommended assessing AI against broader business outcomes, including growth and customer experience. Executives can also consider measures such as operating cost, process time, revenue impact, service quality, and risk reduction where relevant to the specific deployment.

The survey results therefore point to an important next phase. Many enterprises are becoming technically ready for large-scale AI. The management challenge is converting that readiness into measurable economic outcomes.

That requires discipline. Technology investments need defined objectives, accountable owners, suitable metrics, and processes designed to take advantage of AI capabilities. For C-suite teams, the question is increasingly not whether the company has enough AI infrastructure. It is whether that infrastructure is producing results that matter to the business.

Scaling AI requires strong data, governance, and workforce readiness

Moving AI beyond pilots requires more than adding new tools. Organizations need reliable data, secure cloud infrastructure, clear accountability, and employees who understand how to use AI in their daily work. Without these foundations, companies can expand deployment while still struggling to generate measurable business value.

Adrian Clamp, global head of consulting strategy and investment at KPMG International, said organizations should strengthen their data and cloud foundations before expanding AI deployments. Governance and accountability should also be established early. This means defining who owns AI systems, what data they may access, which decisions they may influence or make, and how their performance and risks will be monitored.

Workforce readiness is just as important. AI can change how employees analyze information, make decisions, and execute processes. Companies therefore need to prepare teams for new responsibilities, including reviewing AI outputs, supervising automated actions, and knowing when human intervention is required. Training should be connected to specific roles and workflows rather than treated as a general technology exercise.

Executives should also broaden how they measure returns. Productivity gains can demonstrate immediate value, but they do not capture the full business impact. Clamp recommended evaluating AI against outcomes such as growth and customer experience. Depending on the use case, leadership teams may also track revenue, operating costs, service quality, decision speed, risk, and other measures directly tied to corporate priorities.

Clamp summarized the opportunity by saying, “The greatest benefits come when organizations rethink workflows and decision-making processes around AI.” That requires more than automating existing tasks. Leaders need to determine whether processes should change because AI can perform certain activities faster, continuously, or with a different level of human involvement.

For the C-suite, the priority is execution. Data, infrastructure, governance, people, and business metrics must develop together. Strong performance across these areas gives organizations a better foundation for moving from limited deployments to AI systems that contribute consistently across the enterprise.

Agentic AI needs to be part of broader business transformation

Agentic AI should not be managed as an isolated technology initiative. Because AI agents can increasingly perform tasks and make certain operational decisions, their deployment can affect workflows, accountability, cybersecurity, data management, and the way employees work. The implications extend beyond the IT organization.

Thomson said organizations realizing the greatest value from agentic AI maintain holistic oversight of its use and treat the technology as part of broader business transformation. This approach connects AI investments to operating models and strategic objectives instead of measuring success primarily by technical deployment.

The distinction matters at the executive level. A company can deploy advanced AI and still generate limited value if its data is unreliable, controls are unclear, or business outcomes have not been defined. Thomson captured this directly: “The technology is important, but success ultimately depends on having the right data, the right controls and a clear understanding of the outcomes you’re trying to achieve.”

Holistic oversight also becomes more important as AI systems gain autonomy. Leadership teams need clear rules covering what agents can access, what actions they can perform independently, when human approval is necessary, and who remains accountable. Those controls should reflect the importance and risk of each use case rather than applying the same level of autonomy everywhere.

For business leaders, this also requires cross-functional ownership. CIOs and technology teams may operate the underlying systems, but security leaders, legal teams, business-unit executives, data leaders, finance, and HR can all have responsibilities in enterprise AI deployment. Decisions about investment, risk, workforce changes, and expected returns cannot remain confined to a single function.

The opportunity is to build AI into how the company operates rather than simply adding another technology platform. Organizations that combine suitable data, effective controls, redesigned processes, and clearly defined business outcomes will be better positioned to increase agent autonomy over time.

The technology will continue to advance. The more important executive task is ensuring the organization advances with it. Companies that establish the necessary governance, operational discipline, and business alignment can move beyond experimentation and make agentic AI a practical source of long-term enterprise value.

Key executive takeaways

• Scale cloud operations with AI agents: AI agents can improve cloud visibility, automate routine management, and help contain staffing growth as complexity rises. Leaders should link greater agent autonomy to proven security and oversight capabilities.
• Build trust before increasing autonomy: Security, visibility, and governance remain major barriers to agentic AI. With nearly half of surveyed organizations reporting a cybersecurity breach in the past year, executives should establish access controls, monitoring, and accountability before expanding agent permissions.
• Turn successful pilots into enterprise systems: Access to AI is no longer the main challenge; scaling it across people, processes, data, and technology is. Prioritize high-value use cases with clear ownership and governance before pursuing broader deployment.
• Connect technical readiness to business results: Nearly all surveyed organizations report architecture capable of large-scale data-driven decision-making, yet only 65% said operational efficiency exceeded expectations. Leaders should measure AI investments against concrete business outcomes rather than technical deployment alone.
• Strengthen the foundations for AI scale: Reliable data, secure cloud infrastructure, governance, and workforce readiness are essential for moving beyond pilots. Measure success through growth, customer experience, costs, risk, and other outcomes tied directly to corporate priorities.
• Make agentic AI a business transformation priority: AI agents should be integrated into workflows, decision-making, and operating models rather than managed as isolated technology projects. Cross-functional leadership, clear controls, and defined outcomes will determine whether greater AI autonomy creates durable enterprise value.