Anthropic is convinced that with Claude Code, COBOL’s days are numbered. Investors see this as bad news for IBM, even though the company itself has been using AI to rewrite COBOL for years.
Claude Code, Anthropic’s AI coding assistant, is shaking the software industry to its foundations. After Microsoft, Salesforce, and other major SaaS providers, it is now IBM’s turn. Big Blue’s share price dipped by thirteen percent because Anthropic claimed in a blog post that Claude Code can make the old programming language COBOL disappear for good.
Investors see this as bad news for IBM because mainframes—those old computer rigs that COBOL often still runs on—remain a significant source of income. But what Anthropic is claiming is actually what IBM has been saying and doing for years.
Ancient programming language
COBOL (Common Business-Oriented Language) is known today as the antonym of modern IT. Yet COBOL systems still form the heart of critical infrastructure for banks, airlines, and government services. Due to high costs and a lack of expertise, time has stood still for many organizations.
Phasing out COBOL requires more than rewriting a few lines of code. Replacing systems is a complex, expensive, and risky undertaking: a small error can damage critical infrastructure. It certainly doesn’t help that many COBOL developers have retired and their knowledge has not been passed down to younger generations.
As a result, organizations are dependent on a handful of experts, and the risk of errors during migration is high. Many organizations have therefore played it safe and kept their mainframes in place. Anthropic claims that AI addresses these very bottlenecks by automating the heavy lifting of analysis.
How AI makes COBOL disappear
In a blog post, Anthropic describes how Claude Code can accelerate COBOL modernization. The migration takes place step by step, with validation after every change. AI translates COBOL into modern programming languages, builds API wrappers for existing components, and ensures that old and new code can temporarily run side-by-side. This way, changes are tested immediately, and risk is limited to small, manageable parts.
Before the actual migration, AI creates test scenarios to guarantee that the updated code produces identical results to the original COBOL. Teams decide which scenarios require additional manual checks and establish performance standards for the new environment. By automatically documenting how data moves through a system, up-to-date descriptions and diagrams of processes emerge that no one had insight into anymore. Furthermore, AI can quickly expose risks such as tightly coupled modules or accumulated technical debt, allowing teams to make informed choices about which parts to tackle first.
After the analysis phase, AI tools put forward concrete recommendations for priorities and risk management. Subsequently, IT teams—taking business goals and regulations into account—determine which components will be modernized first. This is also the moment when the architecture, code standards, and integration requirements for new modules are established.
Every successful modernization of a component increases the team’s confidence to tackle more complex parts. This way, organizations avoid large-scale adjustments with high rollback risks, and the technical and economic threshold is significantly lowered, according to Anthropic.
Bad news for IBM
Everything Anthropic says has an immediate impact on the stock market. Today, COBOL and mainframes are still primarily associated with IBM. Consequently, the blog post single-handedly caused IBM’s share price to dive by thirteen percent on Tuesday.
That association is not unjustified: mainframes are still a major business branch for IBM. This was evident once again in the latest quarterly figures: the infrastructure division, which includes mainframes, saw annual growth of twelve percent in 2025, and 21 percent in the final quarter. IBM also continues to announce new mainframe models, such as the z17. Because few organizations have the in-house knowledge to maintain mainframes, they turn to IBM.
Or is it?
The figures and the panic reaction from investors only tell part of the story. Just because IBM is still involved with mainframes doesn’t mean it wants to keep COBOL alive at all costs. IBM has actually been saying for years that mainframes still have a future with AI. The z17 model features optimized hardware and software to run AI workloads. In that vision, the COBOL programming language no longer has a place.
The idea of putting AI to work to drive out COBOL is not unique to Anthropic. IBM already came up with watsonx in 2023, an AI assistant for rewriting COBOL into more modern programming languages. Kyndryl, which was spun off from IBM, is also focused on bringing mainframes into the AI era, as the company told us in detail three years ago.