Why AI Alone Won’t Modernize the Mainframe. And What Actually Works

The recent announcement about Anthropic’s Claude being able to refactor COBOL into modern languages has sparked a lot of discussion in the mainframe community.

And rightly so.

The ability of AI coding assistant to analyze legacy code, identify relationships, and generate modern equivalents represents a real step forward in how we approach modernization.

But it is important to put these developments into context.

Tools are part of the process, not the process

Tools like Claude are excellent at helping with two areas that have historically slowed down mainframe modernization projects: application discovery and code transformation assistance. These capabilities are valuable because understanding legacy systems is often the hardest part of modernization. Decades of embedded logic, undocumented dependencies, and tightly coupled components can make change risky.

AI can accelerate this discovery process and help developers work more efficiently with legacy code. 

But these are just a few pieces of the modernization puzzle.

Enterprise mainframe environments typically include much more than COBOL programs: applications in other (exotic) languages like Natural, Assembler, and Easytrieve, batch workflows and scheduling, online transaction systems, multiple databases and data models, integrations with hundreds of downstream applications, and operational processes built around those systems in decades.

Modernization involves systems, not just programs

Modernization requires transforming entire application ecosystems, not just translating individual programs. This is why the most successful modernization initiatives combine several elements:
automation platforms that can transform large application portfolios

• structured modernization frameworks
• automated testing and validation
• cloud architecture redesign
• and increasingly, AI-assisted transformation

From this perspective, new AI capabilities are enhancing modernization platforms, not replacing them.

We see technologies like Claude as part of a broader evolution in the modernization landscape where AI is improving our ability to analyze legacy systems, understand dependencies, and accelerate parts of the transformation process.

But delivering complete, enterprise-scale modernization still requires a mix of approaches that combine automation, AI, and deep domain expertise.

The secret sauce of enterprise-scale modernization

At Amdocs, we address these challenges through the Amdocs Agentic Services Platform, an agent-driven modernization framework designed to orchestrate the full transformation lifecycle. Within the platform, specialized agents perform key modernization tasks such as application discovery, documentation, disposition analysis, refactoring, rearchitecting (reimagining), and testing.

Some of these agents leverage LLMs to accelerate processes, while tasks that require precision and scale rely on Amdocs’ deterministic, rules-based modernization methodologies and IP, a technology refined through more than 30 years of experience and over 300 successful modernization projects.

Conclusion: The real opportunity here is not AI. It is AI integrated into modernization platforms. This is what finally makes large-scale transformation of legacy systems faster, safer, and more achievable. The conversation around AI and mainframe modernization is just beginning, and it’s an exciting one to be part of.

Curious about how agentic mainframe modernization works in real-life projects? Get in touch with our experts.