Last Updated on 5. March 2026
IBM stock dropped 13 percent in a single day, its worst loss since 2000. The trigger: a blog post. Anthropic announced that Claude Code can automate the modernization of COBOL legacy systems. What used to take years and armies of consultants is now supposed to be possible in a matter of quarters.
The market understood what that means. But have the companies sitting on these systems understood it too?
The COBOL problem is real, and it affects our clients
Hundreds of billions of lines of COBOL code are running in production worldwide. Around 43 percent of all banking systems are built on it, a quarter of German enterprises still rely on COBOL, and the insurance industry is no different. These systems process claims, manage policies, and handle payment transactions. They work. But they can barely be evolved any further.
Everyone in the industry knows the problem: the developers who built these systems are retiring. Documentation is incomplete. And the effort required to touch these systems has been so massive that no one has dared to actually do it.
The announcement is spot on. But the real question is different.
The fact that AI can help with COBOL migration is no longer a surprise. Anthropic has made a compelling case. But a tool alone doesn’t migrate a system. The real question is: Which organization can actually pull this off? Are your teams ready for it?
Code has become cheap. Architecture is expensive. A COBOL migration in an enterprise environment doesn’t just mean translating code line by line. It means understanding business logic that has grown over decades, mapping dependencies, defining a target architecture, and executing the transition in a process-driven way, with quality assurance, business stakeholder involvement, and regulatory compliance.
We’re doing it. Right now.
At mgm, we’ve been going all-in on agentic coding for the past year and a half. 250 developers are trained. Across more than 15 enterprise projects, we’ve developed the patterns needed to deploy AI-powered coding in regulated environments: structured, secure, and scalable.
Our architecture is hardened for enterprise environments. Combined with tools like Claude Code, we can put entire teams on migration projects. Business departments are involved, the process is transparent, and architectural guardrails are in place. This isn’t an experiment. It’s a proven approach.
We’re now taking on the systems that no one has dared to transform before. Because agentic coding has made us so much more efficient that projects are now feasible that would have been unthinkable just a year ago.
What does this mean for insurance, banking, and the public sector?
If you’re running COBOL landscapes and wondering what a realistic migration could look like, not as a slide deck, but as a concrete project with a timeline and a team, talk to us.
We know how to deploy agentic coding in the enterprise. We have the people. We have the experience. And we have the moment.
Got questions? Want to join the conversation?
Meet Jan at the CIONET Online Briefing #23, where he’ll share insights on AI & data trends for 2026 alongside Udo Wuertz and Karl Hausdorf from Fsas / Fujitsu, including a look at a European open-source agentic coding platform for critical infrastructure.
📅 March 16 · 5:00–6:30 PM CET · Online
Join us and bring your questions. We look forward to the exchange!
Further reading:
- Anthropic blog post: “How AI helps break the cost barrier to COBOL modernization”: https://claude.com/blog/how-ai-helps-break-cost-barrier-cobol-modernization
- eBook “The Code Modernization Playbook” by Anthropic: https://resources.anthropic.com/code-modernization-download
- Article at embedded-software-engineering: “Legacy systems in the spotlight: AI tool Claude Code for COBOL optimization sends IBM stock tumbling” (in German): https://www.embedded-software-engineering.de/ki-tool-claude-code-fuer-cobol-optimierung-laesst-ibm-aktie-abstuerzen-a-d462f8aa44b0fb78c030050f0f873708/
