Modernization projects are a CIO headache. Costly, cumbersome, brimming with expectation and full of pitfalls, updating a central technology such as a mainframe that functions as the central nervous system of a business is an arduous undertaking. Vendors are promising AI can change that.
They’ve rushed to deploy, for example, a menu of products that promises to update applications based on common business-oriented language. COBOL remains a widely-used but aging programming language that works across mainframe computers and legacy operating systems, supporting more than 40% of online banking systems, 80% of in-person credit card transactions and 95% of ATM transactions, according to mainframe provider IBM.
While firmly rooted in enterprises, companies have turned to more modern programming languages such as Python and JavaScript, and the often challenging task of updating COBOL code can create bottlenecks for businesses.
In February, large language model provider Anthropic sent markets into a frenzy when it said Claude Code — its AI-enabled coding assistant — could be used for COBOL modernization. The fallout sent mainframe market leader IBM’s stock into a nosedive and elicited a slew of opinions highlighting the challenges of implementing updates tied to the mainframe. But it also prompted a different kind of discussion: how AI can add to the legacy technology rather than force CIOs to leave it behind.
“I see AI as a tailwind for the mainframe, not the opposite.”
Steven Dickens
CEO, HyperFRAME Research
Anthropic’s lofty pitch for approaching COBOL modernization with Claude Code, that it can be used for “systems of any size,” might be true, but as one of the first AI frontier model competitors in the market with such a tool, the company has a long road ahead in growing enterprise trust, said Mitch Ashley, VP and practice lead of software lifecycle engineering at The Futurum Group.
“That’s why, when Anthropic says, ‘We will do all the modernization for COBOL, you don’t need to worry about it,’ it is, to me, a laughable statement,” Ashley said.
The LLM provider also faces competition from established vendors such as IBM and AWS, which operate their own AI modernization tools as well as from firms developing platforms in-house for COBOL modernization, such as Morgan Stanley.
As AI eases modernization efforts, CIOs still have to think through where applications, processes and workflows should live and whether it’s worth maintaining legacy systems. For mainframes, leaving the technology behind entirely is still often a difficult and illogical choice for CIOs, Steven Dickens, CEO of HyperFRAME Research, said in a research note. The evolution of AI tools means they might not have to decide. The technology is driving IT leaders away from rip-and-replace strategies, as they instead work to bring AI directly to mainframe data, he said.
“What I am seeing now is a shift where the platform is not just surviving but is being architected to lead in the era of AI,” Dickens wrote. “I see AI as a tailwind for the mainframe, not the opposite.”
AI is just another tool
Vendors offering AI tools for mainframe modernization often claim that AI can ease transformation as COBOL engineers retire and take their platform know-how with them.
But the reality for CIOs is that legacy system modernization remains an unsolved business problem rather than an unsolved technical problem, said Gartner VP Analyst Matt Brasier.
“How do we justify the money that it will cost? Is there value in going through a bunch of business change to do this and retrain people? Is it going to be worth it?” Brasier said, describing a few of the questions CIOs face when considering a large modernization project.
“It’s just another automation tool at the end of the day.”
Matt Brasier
VP Analyst, Gartner
Even after working through modernization strategies and roadmaps, success is anything but guaranteed. In June, Gartner issued a warning to enterprises that more than 70% of mainframe migrations started this year will fail due to tech leaders’ overestimation of generative AI’s ability to help with transforming legacy implementations. Given the high failure rate, organizations must balance mainframe exit strategies that optimize investments, limiting full platform exits to select cases, according to Gartner.
Beyond generative AI, CIOs have been turning to tools such as machine learning as a way to help modernize legacy systems for years, Brasier said. In fact, many organizations are already using the technology to help modernize COBOL applications, finding value in the technology’s ability to document and describe what an application does and building test cases.
“It’s just another automation tool at the end of the day,” Brasier said.
Leading mainframe vendors, including IBM, have added AI offerings to accelerate modernization for customers.
IBM Bob, an AI coding assistant that helps engineers modernize code, became generally available in April. The company is also heavily investing in expanding AI functionality within the mainframe, making it easier for enterprises to use the technology without having to migrate to another platform for advanced capabilities. As part of that effort, IBM released a private preview of IBM Bob Premium Package for Z for enterprise mainframe applications in April.
By the numbers
95%
The percentage of ATM transactions moving through mainframes
17 million
The lines of legacy code Morgan Stanley modernized with help from AI
7 in 10
The number of mainframe migrations started this year that will fail due to overestimating AI
Previously, IBM rolled out the Spyre Accelerator chip to IBM Z mainframes in 2025 to support AI inference. The company also launched a partnership with Arm in April to build dual-architecture hardware for IBM Z mainframes, allowing enterprises to run AI and data-intensive workloads without having to rewrite code.
Unisys, another mainframe vendor, partners with AWS to offer enterprise customers AI-augmented application modernization services.
When relying on vendors to modernize applications, processes and workflows, enterprises need to be aware that vendor tools are usually built to help migrate companies to their tech stacks, Brasier said.
“IBM’s tools are going to move you onto IBM’s newer Java-based stack. AWS tools are going to move you onto AWS’ cloud-native, serverless function stack,” he said. “The vendors provide the tools with an idea of, ‘This is the stack you’re going to be moving to.’ For many organizations, they have a destination in mind, and it’s probably not one of those vendor stacks.”
Overall, AI developments don’t make it easier for CIOs to leave the mainframe, Brian Klingbeil, EVP and chief strategy officer at IT service management company Ensono, said in a blog post.
“They actually make it easier to stay,” he said.
Homegrown tech and the hyperscaler effect
Investment banking company Morgan Stanley is taking another route to modernizing its mainframe processes. The financial services firm has modernized more than 17 million lines of COBOL, Software AG’s Natural and PERL code into modern languages, including Java and Python, with its in-house platform DevGen.AI.
The generative AI-powered modernization platform has saved developers in excess of 1 million hours of manual coding, Trevor Brosnan, global head of technology strategy, architecture and modernization at Morgan Stanley, told CIO Dive in an email.
“That means more projects getting greenlit, more productive and efficient engineers and the ability to leverage emerging technologies — and deliver more to our business partners,” Brosnan said.
While modernizing legacy COBOL is inherently complex, DevGen.AI enables developers to reverse engineer old code and translate it into a clear design so that they can migrate it into updated architectures, Brosnan said. The platform reduces coding tasks that previously required a week to half a day.
“This capability improves the entire software development lifecycle — from analysis to testing to deployment — allowing our teams to deliver change faster while maintaining full human oversight and robust guardrails,” Brosnan said.
The banking firm is also using coding agents to help plan and execute development workflows beyond simple code generation, Brosnan said.
While firms like Morgan Stanley have developed platforms in-house for modernization projects, vendors are filling the gap and providing modernization tools for companies.
The North Carolina Division of Motor Vehicles selected IT infrastructure services provider Kyndryl earlier this year to replace five legacy COBOL-based systems and implement a unified, cloud-native platform hosted on Microsoft Azure. The modernization contract, which includes platform implementation, training and data migration, totals $84.8 million.
Meanwhile, Microsoft bolstered its Azure Migrate services with agentic AI last fall, a broad migration tool aimed not just at mainframes but legacy technology in general.
“It’s going to be a journey. In the next little while, we see it’s going to be a combination where some workloads may remain on prem, on the mainframe, some will move to the cloud.”
Asa Kalavade
VP, AWS Transform
AWS Transform, which was also launched last year, uses agents for multiple points of the mainframe migration process, including the initial discovery process followed by migration waves. In one notable example, Toyota Motor North America used AWS Transform to modernize more than 40 million lines of COBOL to Java.
Going forward, successful mainframe modernization using AI hinges on both reverse engineering applications and forward engineering, according to Asa Kalavade, VP of AWS Transform.
“It’s going to be a journey,” Kalavade told CIO Dive. “In the next little while, we see it’s going to be a combination where some workloads may remain on prem, on the mainframe, some will move to the cloud.”
