COBOL is one of the oldest programming languages still in active use. It was created in 1959. Yet today, it quietly powers the financial systems most people use every day. On February 23, 2026, Anthropic published a blog post explaining how its AI coding tool, Claude Code, can automate the hardest parts of COBOL modernization. The announcement sent shockwaves through the tech industry — most visibly sending IBM's stock down 13% in a single day, its worst drop since October 2000.
This article explains what Claude Code actually does with COBOL, why the problem is so hard, how AI changes the economics, and what experts say about the limits of this approach.
The COBOL Problem in Numbers
To understand why this announcement matters, you first need to know how deeply COBOL is embedded in modern life.
| Statistic | Detail |
|---|---|
| ATM transactions handled by COBOL | ~95% in the United States |
| Lines of COBOL in production globally | Hundreds of billions |
| Industries relying on COBOL | Finance, airlines, government, insurance |
| Universities teaching COBOL today | Only a handful |
| Year COBOL was first released | 1959 |
| Biggest challenge | Retiring workforce; institutional knowledge lost |
The engineers who wrote these systems have mostly retired. The documentation never kept up with decades of patches and changes. The result: organizations depend on code that almost nobody fully understands anymore.
Why COBOL Modernization Has Always Been So Hard
COBOL modernization is not like refactoring a modern codebase. As Anthropic described it, you are reverse engineering business logic from systems built when Nixon was president — untangling dependencies that evolved over decades, and translating institutional knowledge that now exists only in the code itself.
The traditional path required huge consulting teams. Modernizing a COBOL system once required armies of consultants spending years mapping workflows, resulting in large timelines and high costs that few were willing to take on.
Here is what made it so expensive:
| Challenge | Why It's Hard |
|---|---|
| Hidden dependencies | Modules share data through files, databases, and global state — not visible in static analysis |
| Lost documentation | Decades of changes with no corresponding docs |
| Implicit business logic | Rules embedded in code, not in any manual |
| Talent scarcity | Fewer developers understand COBOL each year |
| Risk of breakage | One wrong migration can crash critical systems |
The fear isn't technological — it's epistemic. You cannot modernize what you cannot fully understand.
What Claude Code Does Differently
Anthropic's claim is straightforward: AI makes the economics work by automating what used to require armies of consultants, freeing your engineers to make the migration decisions that require their domain expertise.
Claude Code tackles COBOL modernization in four distinct phases:
Phase 1: Automated Exploration and Discovery
Claude Code maps dependencies across thousands of lines of code, documents workflows that no one remembers, and identifies risks that would take human analysts months to uncover. AI starts by reading the entire COBOL codebase and mapping its structure. It identifies program entry points, traces execution paths through called subroutines, and maps data flows between modules.
Critically, it goes further than simple mapping. Shared data structures, file operations that create links between modules, initialization sequences that influence runtime behavior — these types of implicit dependencies do not appear in static analysis. This is precisely what makes COBOL modernization risky. Automated discovery finds these hidden relationships before they cause problems during migration.
Phase 2: Risk Analysis and Opportunity Mapping
Once the codebase is mapped, Claude Code assesses which parts are safe to move first and which need caution. It flags modules with high coupling, surfaces isolated components as candidates for early migration, and documents areas of technical debt before they become surprises.
Phase 3: Strategic Planning (with Human Oversight)
This is where human engineers stay in control. According to Anthropic's blog, AI suggests prioritization based on risk and complexity — but the engineering team decides what to modernize first based on business value, technical risk, and regulatory requirements.
Phase 4: Incremental Implementation and Validation
Execution happens one component at a time. Each step either succeeds and gets validated, or fails and gets corrected while the scope is small. You never have massive changes in flight where failure means rolling back weeks of work.
Before and After: How the Timeline Changes
| Modernization Task | Before Claude Code | With Claude Code |
|---|---|---|
| Mapping COBOL dependencies | Months of consultant work | Days or weeks |
| Documenting legacy workflows | Often never completed | Auto-generated |
| Identifying hidden risks | Human analysts, months | Automated surfacing |
| Full codebase modernization | Years | Quarters |
| Cost model | Large consulting engagements | Smaller, focused teams |
The Market Reaction: Why IBM Fell 13%
The announcement didn't just interest developers. IBM shares had their worst day in more than 25 years on Monday after Anthropic said its Claude Code tool can help modernize COBOL. The stock plunged 13% in its biggest single-day percentage loss since October 2000. With the decline, IBM shares fell 27% in February.
IBM wasn't alone in taking a hit. Accenture and Cognizant also declined following the news — a sign that investors are looking at the entire consulting model around legacy modernization, not just IBM's mainframe hardware business.
| Company Affected | Reason | Stock Impact |
|---|---|---|
| IBM | Major COBOL consulting and mainframe revenue at risk | -13% on Feb 24 |
| Accenture | Large legacy modernization consulting practice | Declined |
| Cognizant | Significant COBOL services revenue | Declined |
IBM's Counter-Argument
IBM pushed back quickly. Rob Thomas, SVP and chief commercial officer at IBM, wrote that the engineers doing this work know the code is the starting point, not the destination. What the application runs on, how it scales, how it recovers, how it is encrypted, and how it integrates with everything around it — that is the real modernization work.
IBM also pointed to its own AI tool. IBM announced its own mainframe tool, the watsonx Code Assistant for Z, in August 2023, offering capabilities like simple-language explanations of COBOL code. A lot of clients are not entirely comfortable with opening up their code repositories to a SaaS-provided third-party LLM running out in the cloud, according to IBM's chief operating officer.
Evercore ISI analyst Amit Daryanani added that clients already had the option to migrate from the mainframe, yet they are sticking with the platform, suggesting the fear of displacement may be outrunning the reality.
What Claude Code Can and Cannot Do
It's important to be clear about the scope of Claude Code's capabilities here.
| Claude Code Can | Claude Code Cannot |
|---|---|
| Map code dependencies automatically | Replace human judgment on architecture decisions |
| Document forgotten workflows | Understand regulatory requirements on its own |
| Identify migration risks faster than humans | Handle hardware-software integration challenges |
| Translate COBOL logic to Java or Python | Guarantee zero defects in migrated code |
| Compress discovery from months to weeks | Manage change management across an organization |
| Generate test suites for validation | Certify compliance with financial regulations |
As one industry expert noted, translating code itself isn't modernization. The full scope includes security, scalability, encryption, and integration — all areas that still require deep human expertise.
Anthropic's Code Modernization Playbook
Alongside the February 23 announcement, Anthropic published a companion resource. The Code Modernization Playbook shares how agentic coding solutions like Claude Code can accelerate modernization without delaying more pressing projects, highlighting how agentic coding can analyze complex codebases, generate test suites, and preserve business logic during migrations.
The playbook covers:
- How to migrate from COBOL to modern languages like Java or Python
- Prompt engineering techniques for code migration
- How to choose the right agentic coding solution
It is available at resources.anthropic.com/code-modernization-playbook.
What This Means for Organizations Running COBOL Today
The practical path forward, according to Anthropic, is to start small and validate carefully.
- Pick one component with clear boundaries and moderate complexity.
- Use Claude Code to analyze and document it fully before touching anything.
- Plan with your engineers — let AI suggest priorities, humans decide strategy.
- Implement incrementally — one module at a time with testing at every step.
- Validate carefully — run old and new code in parallel during transition.
- Build confidence — early wins surface the adjustments needed for larger migrations.
This approach applies to COBOL systems of any size, from a small regional bank to a national government treasury system.
The Broader Pattern: AI Disrupting Consulting Economics
The COBOL story is part of a larger shift. The pattern is becoming familiar: each new AI capability announcement triggers a reassessment of which existing revenue streams might be compressed, and the market prices in fear.
Just one week before the COBOL announcement, cybersecurity stocks sold off sharply after Anthropic introduced Claude Code Security, a tool that scans codebases for vulnerabilities. The concern across the market is the same: AI is compressing workflows that entire consulting practices were built to deliver manually.
The phrase "AI will replace 10 developers" is misleading. The more accurate way to put it is: a developer using AI effectively can deliver the output of 10 senior engineers on specific tasks. AI handles the repetitive scaffolding; humans handle architecture, risk, compliance, and design.
Conclusion
Anthropic's February 2026 announcement about Claude Code and COBOL modernization is one of the most consequential AI-in-enterprise stories of the year so far. The core claim — that AI can compress a multi-year, multi-million-dollar consulting engagement into a process measurable in quarters — is credible and backed by specific technical capabilities. Dependency mapping, workflow documentation, risk surfacing, and incremental translation are all tasks where AI genuinely accelerates human work.
But the critics are also right. Moving code is not the same as modernizing a system. Hardware integration, security, compliance, and institutional change management still require human expertise. The honest picture is one of significant acceleration, not full automation.
For any organization still running COBOL — and there are thousands of them — the smart move is to treat Claude Code as a powerful discovery and analysis engine, not a replacement for experienced engineers. Start with one component, validate thoroughly, and build from there.
The economics of COBOL modernization have shifted. The question now is who moves first.