AI Can Read COBOL. That Doesn’t Change The Mainframe Equation.

AI can read and analyze COBOL code at scale, but it doesn’t reduce the complexity of mainframe migration – making modernization the smarter path for most enterprises.

The headlines only tell part of the story. For mainframe users, it’s not the part that matters most.

Every few years, something comes along that’s supposed to finally make the mainframe obsolete.

This time, it’s AI.

Anthropic’s recent announcement around COBOL modernization has led to a predictable question: does this mean the mainframe’s clock really is ticking, and is it time to accelerate migration plans?

Short answer: No.

Longer answer: also no. These developments don’t, in fact, make it easier to leave the mainframe. They actually make it easier to stay.

A quick mainframe reality check

There’s a tendency to talk about the mainframe as if it’s something enterprises are stuck with. On the contrary, many organizations are quite deliberate about keeping it.

High‑volume transactions. I/O‑heavy batch processing. Systems where downtime is not an option. These are exactly the workloads the mainframe was built for, and it still does them exceptionally well.

Headlines may position making applications “easier to move” as inherently good news. But easier to move is only helpful if moving is the right answer. Often, it isn’t.

What AI changes (it’s not nothing)

To be clear, AI in this space is not hype. It has materially improved the ability to understand large, complex, and often under‑documented applications. Mapping code to business workflows. Finding dependencies no one remembers owning. Reducing reliance on a shrinking pool of experts, some of whom retired years ago but still get called when things break.

This is work we’re already doing with clients. What used to take months of manual analysis can now be done in a fraction of the time, with far better visibility and far fewer surprises. That alone changes how organizations approach modernization decisions.

In one recent engagement with a large insurer, we analyzed millions of lines of legacy code supporting core policy systems in a matter of weeks. The conclusion wasn’t “move it all to the cloud.” It was “this platform is doing exactly what we need – let’s modernize it to be more agile, make it easier to release new products, AND save money without breaking anything important.”

This is becoming the norm, not the exception. Rather than collapsing the platform landscape, AI is helping organizations get sharper about which platforms are best suited to which work.

What AI doesn’t magically fix

Where things tend to go off the rails is assuming that better code translation makes migration easy.

It doesn’t.

Organizations that have struggled with mainframe migrations rarely failed because COBOL was hard to convert. They ran into trouble with everything around it: data models, transaction behavior, performance characteristics, resiliency, integrations, and the operational muscle memory built over decades. Add to that the non‑functional requirements the mainframe delivers – requirements that are extraordinarily difficult to replicate on distributed infrastructure at equivalent cost and reliability – and the challenge becomes clear.

Those things matter. A lot. And they don’t come along for free just because the code lives somewhere else. AI is a powerful tool, but it still needs people who know what they’re looking at, why it matters, and what can safely be changed.

IBM leadership has been increasingly explicit about this distinction. In a recent post, IBM’s Chief Commercial Officer and SVP of Software Rob Thomas noted that while translation tools continue to improve, they “capture almost none of the actual complexity.” He points to decades of tightly integrated hardware and software design—transaction processing, I/O optimization, and operational behavior that cannot simply be replicated by moving code, underscoring his message with a particularly apt analogy: iOS alone does not an iPhone make.

Barry Baker, COO of IBM Infrastructure and GM of IBM Systems, put it more bluntly: “That’s platform design, not language semantics.”

We asked Claude to speak for itself

Out of curiosity – and a bit of due diligence – we asked Claude itself what this new capability actually changes.

The answer was refreshingly grounded. It didn’t focus on language conversion at all, but on the platform realities that make mainframe systems hard to replace: how transactions behave, how I/O works at scale, how availability is engineered, and how much operational knowledge is embedded in the environment.

In other words, even the AI didn’t claim this makes migration cheap or easy (exact quote: “Anyone who hears ‘Claude can convert COBOL’ and concludes ‘therefore we can migrate off the mainframe cheaply’ is badly misreading the situation.”) That should probably tell us something.

Modernization and migration are not the same thing

One of the quieter shifts happening right now is that organizations are increasingly choosing to modernize on the mainframe. It’s often lower risk and delivers tangible benefits faster than wholesale migration.

More and more, we see teams modernizing applications in place – cleaning up code, adopting more modern languages or patterns, and improving agility – while keeping those systems on the platform that already runs them best.

AI accelerates that approach. It doesn’t force a different one, and it shouldn’t. For many enterprises, modernization in situ isn’t a temporary stop on the way out. It’s the destination.

How organizations are thinking about this

In practice, mainframe users tend to fall into a few broad groups.

Some are core users, where the mainframe is strategic, and migration isn’t a serious discussion. Others are hybrid users, keeping the mainframe for workloads it handles best while running other applications elsewhere. A smaller group is actively migrating off entirely, with widely varying levels of complexity and risk.

AI‑assisted analysis may cause some organizations to revisit priorities. In a few cases, it might move a migration project up the list – but from a priority #17 to a #14, not from “someday” to “now.” In many others, it simply makes modernization easier and clarifies which platforms are best suited for which work.

Nor does it undermine managed mainframe models. If anything, it reinforces why they exist. AI creates opportunities both to bring intelligence closer to mainframe‑resident data and, in some cases, to run directly on the platform—making it possible to modernize and innovate in place without sacrificing the reliability and efficiency organizations depend on.

This view is increasingly reflected in the market. Research from HyperFrame describes a shift toward bringing AI to where sensitive, mission‑critical data already lives, rather than forcing that data to move. The goal isn’t to exit the mainframe. It’s to make it move as fast as the rest of the business.

Separating the signal from the noise

AI‑assisted COBOL modernization is a genuinely useful development. It gives teams better tools and better insight into systems that have historically been hard to understand.

What it doesn’t do is suddenly make the mainframe irrelevant or force immediate migration decisions.

As they have before, the headlines may suggest a turning point. The reality is more measured, and far more useful for leaders focused on outcomes rather than hype.