The Day the Market Woke Up
On February 23, 2026, IBM shares plunged 13.2% in a single trading session — their worst single-day drop in over 25 years — wiping more than $31 billion in market capitalization off the table. The catalyst was Anthropic’s announcement that Claude Code could modernize legacy COBOL systems in quarters instead of years, threatening what had been IBM’s most protected franchise: the complex, expensive consulting work required to maintain and migrate systems that enterprises can’t simply turn off. Wall Street read the announcement as an existential threat to mainframe consulting. And they weren’t wrong about the threat. But they were wrong about one thing — and it’s the thing that matters most for the hundreds of enterprises sitting on top of these systems right now.
The Mainframe Isn’t Going Anywhere
Here’s what the stock market reaction missed. The problem isn’t that mainframes are hard to work with. The problem is that they are structurally impossible to simply replace — and no AI tool, however powerful, changes that fundamental reality. Decades of regulatory change, business rule amendments, and edge-case handling live in the code and in the memory of the people who maintain it. The system knows things that no one has written down. The business logic baked into COBOL programs running at major banks, insurance companies, and government agencies wasn’t written once and left alone. It was modified thousands of times over forty years by people who are now retired — or gone entirely. Nobody documented it. Nobody drew the dependency maps. The code is the documentation. Every new compliance requirement becomes an integration project. Every integration project adds cost, fragility, and another layer of undocumented dependency. The stacks that run on top of mainframes — the modern Java services, the APIs, the reporting layers, the digital banking interfaces — were all built assuming the mainframe would be there. They call it. They depend on it. They were never designed to work without it. 95% of ATM transactions in the United States run on COBOL. $3 trillion in daily commerce flows through mainframe systems. 45 of the top 50 banks in the world run on IBM mainframes. These systems aren’t legacy in the sense of "old and replaceable." They are foundational infrastructure that cannot be switched off without catastrophic operational and regulatory consequences. The mainframe isn’t going anywhere. Which means the people who can work with it are more valuable than ever — and the shortage of those people is more urgent than it’s ever been.
The Real Crisis Nobody Is Solving
The COBOL developer shortage is not a new problem. It has been quietly worsening for two decades. The average COBOL programmer is in their late fifties. Nearly one in three will retire by 2030. Younger developers don’t learn COBOL — not because it’s impossible, but because there’s been no compelling reason to, and no good path to do it. The traditional answer has been to hire the remaining specialists at enormous cost, or to engage consulting firms at even greater cost to manage migrations that take years and frequently fail. Gartner estimates that 70% of COBOL migrations fail or are abandoned. The IT Modernization market is projected to reach $18 billion by the end of the decade — most of it spent on projects that underdeliver. What nobody has built — until now — is a program that answers the question enterprises actually need answered: how do we equip the modern developers we already have to work safely and effectively with the legacy systems we cannot replace?
Introducing Claude for Mainframe — the Industry’s First Program of Its Kind
bILTup has designed and built Claude for Mainframe: Augmenting Modern Developers to Work with Legacy Systems. It is, to our knowledge, the first enterprise training program in the industry specifically designed to address the mainframe skills gap using Claude as the core comprehension and workflow tool. This is not a COBOL bootcamp. It does not attempt to turn Java and Python developers into career mainframe engineers. Its explicit goal is far more achievable and far more valuable: to raise a team of modern developers to a level of legacy-system competence sufficient to work independently on comprehension, documentation, and routine modifications — with mainframe SME review reserved for high-risk changes and architectural decisions. The program runs 5 to 8 months depending on the track selected, with a weekly cadence of one 4-hour instructor-led session, 2 to 3 hours of self-paced O’Reilly content, and scheduled office hours. Every session is built around the client’s actual codebase and processes — not sample code, not synthetic exercises.
What the Program Actually Teaches
The program is structured across five paths, each building on the last. Path 1 establishes the mainframe ecosystem for modern developers — z/OS fundamentals, dataset types, JCL essentials, DB2 and IMS, and the security and change management culture that makes mainframe work different from modern software delivery. Participants come out of Path 1 able to navigate the environment they’ve been afraid to touch. Path 2 establishes the Claude workflow for legacy code work — when to use Claude chat versus Claude Code, how to manage context across multi-thousand-line programs, the prompt patterns that work for legacy code, and critically, the verification habits that catch what Claude gets wrong. This is the foundation that makes everything else safe. Path 3 is where the biggest immediate business value lives — reading, documenting, and understanding legacy systems. Section-by-section COBOL analysis, dependency mapping across entire codebases, business rule extraction and validation with business stakeholders, JCL and batch flow interpretation, and structured knowledge capture from retiring SMEs before that knowledge walks out the door. For most enterprises, this path alone justifies the entire program. Path 4 teaches safe modification — the draft-review-test cycle that keeps Claude in the loop without letting Claude make unchecked changes to production systems. Generating COBOL modifications, writing and modifying JCL, test generation and regression, cross-language interop with Java and Python, and working with mainframe data from modern tooling. Path 5 is the optional modernization extension — for teams ready to move beyond augmentation into selective modernization. API wrapping, strangler fig patterns, selective refactoring, and full COBOL-to-Java conversion — with an honest final module on when not to modernize. Every path concludes with a capstone: real system, real cohort teams, real deliverable presented to organizational leadership.
Why This Is Different From Everything Else in the Market
The market is full of COBOL tutorials, mainframe certification programs, and AI-assisted modernization tools. None of them solve the actual problem. COBOL tutorials assume you want to become a COBOL developer. Most modern engineers don’t — and enterprises don’t need them to. What enterprises need is a Java developer who can read COBOL well enough to understand what it’s doing, document it accurately, and make routine modifications safely. That’s a completely different skill set than becoming a mainframe specialist. Modernization tools — including IBM’s watsonx Code Assistant, Amazon Q, and yes, Claude Code used independently — address the code translation problem but not the human capability problem. You still need people who understand what the code is supposed to do before you can trust an automated translation of it. The organizations that are failing at COBOL migration aren’t failing because they lack tools. They’re failing because they lack the human judgment to use those tools correctly. bILTup’s Claude for Mainframe program builds that judgment. It’s not a tool. It’s a structured, practitioner-led program that produces developers who can work with legacy systems safely — and a workforce that multiplies the reach of the SMEs you have left before they retire.
The Window Is Closing
The mainframe SMEs who hold the institutional knowledge your systems depend on are retiring. Every year that passes is another year of undocumented business logic that gets harder to recover. The enterprises that act now — that invest in building the human capability to work with, document, and selectively modernize their legacy systems — will be in a fundamentally different position than the ones that wait for a tool to solve it. Claude for Mainframe is available now as a custom enterprise engagement. Programs are scoped individually based on team size, existing skill level, target mainframe environment, and business objectives.
Talk to us about building a Claude for Mainframe program for your team → View the full program outline →