Only 10% of IT Chiefs Use Agentic AI for Modernization, Study Finds

Only one in ten IT leaders currently uses agentic AI for modernization, according to research from Kyndryl — and nearly one in five companies that modernized say they saw limited or unclear value from the effort. The numbers frame a hard question for the industry: can AI finally fix what decades of IT modernization programs have not?

The stakes are considerable. Four in ten enterprises still run mainframes, by the report's own count, and almost half of modernization respondents admit they are behind schedule and over budget. AI vendors are pitching their technology as the answer. A former IBM strategist says the economics don't add up.

Why modernization keeps failing

IT modernization is not a new problem. CIOs have chased it for decades through web services, microservices, virtualization, cloud, and x86-based Windows and Linux servers. The goal was always the same: more advanced, more flexible systems.

The results have been underwhelming. Kyndryl's researchers found that modernization often turns into just another expensive, time-consuming project.

• "Only about half of those modernizing report achieving better IT operations, and less than half are seeing innovation gains."
• "Nearly one in five, 18%, say their firms saw limited or unclear value from modernization efforts."
• Almost half of respondents are behind schedule and suffering cost overruns.

Modernization has to become more than swapping mainframes for something sleeker. The Kyndryl study points to a shift in motivation: "Cost reduction and legacy escape are no longer the leading motivations to modernize."

Why the change? AI is increasing demand on nearly every element of IT, and sovereignty considerations are becoming more acute. Leaders now want their mixed estates of old and new systems to simply work better together.

What can agentic AI actually do here?

Agentic AI — software agents that operate autonomously — is shaping up as the next wave of modernization. Its concrete value, according to the Kyndryl report, is visibility.

"Fewer than one in 10 organizations are fully confident they know those dependencies today, a significant technical blind spot," the report stated. Agents can map dependencies between legacy and new systems, something almost nobody can currently do with confidence.

The early use cases are specific and unglamorous:

• Mapping hidden dependencies between systems
• Generating code
• Creating documentation

These are traditionally time-consuming, error-prone processes. "Agentic AI is showing early promise to address longstanding challenges in modernization and speed progress," the Kyndryl authors wrote. "Organizations not using agentic AI for modernization are much more likely to be behind schedule than early adopters."

The takeaway in plain words: agents won't replace your mainframe, but they can chart the terrain around it faster than humans can.

Why the economics 'don't hold up'

Andy Thurai, founder of The Field CTO and former chief strategist with IBM, read the study and offered a blunt assessment. "IT modernization increasingly looks like a chronic condition," he said. "According to this report, four in ten enterprises are still running mainframes. We keep hearing modernization is the number-one mandate, yet the finish line is always 5 to 10 years out."

His diagnosis has two parts. First, procurement habits: "organizations are addicted to buying software. The legacy systems underneath it almost never get decommissioned. AI is now the latest shiny distraction from the grind of untangling that technical debt."

Second, cost opacity. "Vendors are pitching AI to IT leaders as the cure-all," Thurai said. "The economics don't hold up. When AI drives infrastructure sprawl, the monthly bill becomes a black box. The report found that three in 10 leaders already worry AI-driven modernization will end in wildly unpredictable compute costs."

Three in 10 leaders bracing for unpredictable compute bills is a real budget risk, not a hypothetical one. The takeaway: AI can accelerate modernization work while quietly inflating the cost of running what you've built.

Where AI modernization is working

There is a genuine bright spot, and it is narrower than the vendor pitch. Thurai pointed to organizations "that went past the bolt-on chatbot" as the ones seeing results.

"They're using agentic frameworks as an orchestration layer to map out hidden technical dependencies, and then generating the missing documentation as they go," he said. "Agentic AI is speeding up initiatives that used to eat months of error-prone manual labor."

The mechanism matters here. These companies are not using agents to rewrite their estates. They are using agents as a coordination layer — finding what connects to what, then documenting it — which makes every later decision cheaper and safer.

Today's AI stack is tomorrow's legacy

Thurai closed with the structural point that keeps this story honest. "Sooner or later, today's AI-driven architectures are going to be the legacy environments somebody else has to modernize," he said.

That is the mechanism of the whole cycle in one sentence. Every generation of technology — cloud, microservices, and now AI — solves one problem and creates the next decade's technical debt. For CIOs, the practical read of the Kyndryl data and Thurai's analysis is to use agents where they demonstrably help, in mapping, code generation, and documentation, while budgeting hard for the compute costs they bring with them.

By Sophie Lindqvist