For decades, the global financial system has quietly depended on aging code written in a programming language many assume belongs in a museum. That language is COBOL — and it still powers vast swathes of banking, insurance, government, and transaction infrastructure.
Now, artificial intelligence is reshaping how enterprises approach COBOL modernisation. But the market’s response suggests investors are still grappling with what this shift truly means.
Earlier this week, shares of IBM fell 13% in their worst single-day drop in more than 25 years. The catalyst? A blog post from AI startup Anthropic claiming its Claude Code tool can dramatically accelerate COBOL modernisation — a segment of enterprise transformation long associated with high-value consulting work.
The reaction was swift. But beneath the headlines lies a deeper story about legacy systems, AI disruption, enterprise trust, and the future of mainframes.
Why COBOL Still Matters More Than You Think
COBOL (Common Business-Oriented Language) may have been introduced in 1959, but it remains deeply embedded in today’s digital economy.
Hundreds of billions of lines of COBOL code run in production globally. In the United States alone, an estimated 95% of ATM transactions depend on COBOL-based systems.
These systems:
- Process financial transactions
- Manage government benefits
- Power insurance claims
- Support payroll infrastructure
- Operate airline reservation systems
COBOL isn’t “legacy” in the sense of being obsolete. It’s legacy in the sense of being foundational.
The real problem isn’t that COBOL is unreliable. It’s that it’s difficult to replace — and even harder to understand.
The Talent Gap Driving Modernisation Costs
The generation of developers who originally built these systems is largely retired. Universities rarely teach COBOL. Younger engineers typically work in Java, Python, or modern cloud-native stacks.
This shrinking talent pool has made COBOL modernisation expensive and complex.
Traditionally, large consulting teams would:
- Map system dependencies
- Document undocumented workflows
- Identify hidden logic
- De-risk migrations
- Gradually rewrite or refactor systems
These projects often took years — and generated significant consulting revenue for firms like:
- IBM
- Accenture
- Cognizant
The economics of this work relied heavily on human expertise and long timelines.
AI now threatens to compress that model.
How Anthropic’s Claude Code Changes the Equation
Anthropic’s Claude Code claims to accelerate COBOL modernisation by automating the most labor-intensive phases of the process.
Instead of relying on teams of analysts to manually examine thousands of lines of code, Claude Code can:
- Map code dependencies automatically
- Identify workflow structures
- Detect risk areas
- Document system architecture
- Provide modernisation insights
Anthropic argues that what once required years of consulting work can now be completed in quarters.
That claim alone was enough to rattle investors.
The concern: if AI reduces the need for prolonged consulting engagements, revenue tied to legacy modernisation could shrink significantly.
But IBM Was Already Working on AI for COBOL
What the market reaction may have overlooked is that IBM has been pursuing AI-driven COBOL transformation for years.
The company launched watsonx Code Assistant for Z, an AI tool designed to help enterprises understand and modernise COBOL applications on IBM Z mainframes.
IBM CEO Arvind Krishna has previously stated that adoption of the tool has been strong, particularly among clients seeking to analyze their codebases before deciding on modernisation paths.
IBM’s strategy has centered on:
- Using AI to interpret legacy COBOL
- Assisting in rewriting code to Java
- Helping enterprises decide what to modernise
- Preserving core mainframe value
In other words, IBM was already promoting AI-assisted COBOL transformation before Anthropic’s announcement.
Translation vs. Transformation: The Critical Distinction
IBM’s response to the selloff focused on a key nuance: translating code is not the same as modernising a platform.
Rob Thomas, IBM’s Senior Vice President and Chief Commercial Officer, emphasized this distinction.
Rewriting COBOL into Java may help modernise syntax — but it does not automatically replace the underlying infrastructure.
IBM Z mainframes run on:
- z/OS operating systems
- Deeply optimised transaction processing engines
- Hardware-level performance tuning
- Integrated security controls
- Quantum-safe encryption protocols
These elements represent decades of vertical integration between hardware and software.
A code translation tool does not replace that ecosystem.
The question investors are wrestling with is whether AI-assisted translation reduces reliance on IBM’s broader platform — or increases demand for modernization services.
Not All COBOL Runs on Mainframes
Another overlooked factor complicates the narrative.
Roughly 40% of COBOL runs on distributed systems — including Windows and Linux — rather than traditional mainframes.
That means part of the disruption story isn’t strictly about IBM’s mainframe dominance. It also affects distributed enterprise environments.
COBOL modernisation is a broader industry challenge, not just a mainframe issue.
Market Pattern: AI Announcements Trigger Selloffs
The reaction to Anthropic’s blog post fits a broader pattern emerging in financial markets.
Recent examples include:
- Cybersecurity stocks falling after Claude Code Security was announced
- Consulting firms declining on AI automation fears
- Enterprise software vendors seeing volatility after generative AI launches
Each new AI capability prompts investors to reassess which existing revenue streams may be compressed.
Fear gets priced in immediately.
Reality often unfolds more gradually.
Consulting Firms Also Felt the Pressure
IBM wasn’t alone in taking a hit.
Shares of Accenture and Cognizant declined as well, signaling that investors are evaluating the entire consulting ecosystem built around legacy system modernisation.
For decades, consulting firms thrived on:
- Manual code audits
- Multi-year transformation programs
- Complex system integration projects
If AI significantly shortens project timelines, billing models may need to adapt.
But that doesn’t necessarily mean demand disappears.
It could mean:
- More projects become economically viable
- Smaller enterprises attempt modernisation
- AI augments consultants rather than replaces them
Real-World Case Studies Tell a More Nuanced Story
IBM’s customers offer insight into how AI-assisted modernisation is actually unfolding.
Royal Bank of Canada has used IBM’s watsonx Code Assistant for Z to map dependencies and build modernisation blueprints.
Meanwhile, National Organisation for Social Insurance reported a 94% reduction in time needed to analyze legacy COBOL code — cutting an eight-hour task to roughly 30 minutes.
These results suggest AI accelerates analysis.
But analysis is only one phase of enterprise transformation.
Deployment, compliance validation, regulatory approvals, performance testing, and operational migration remain complex.
Why Enterprises Don’t Move Off Mainframes Easily
Even when migration options exist, many enterprises choose to stay on mainframes.
Evercore ISI analyst Amit Daryanani noted that clients have long had the option to migrate away from mainframes — yet many remain.
Reasons include:
- Reliability
- Performance under extreme transaction loads
- Security certifications
- Regulatory compliance
- Decades of operational stability
Mainframes process enormous transaction volumes with minimal downtime.
Replacing that infrastructure involves risk that CFOs and CIOs weigh carefully.
AI may reduce technical barriers, but strategic decisions remain conservative.
AI Makes Modernisation Economically Viable
What is undeniably true is that AI changes the cost curve.
Historically, many enterprises postponed COBOL modernisation because:
- It was too expensive
- It was too risky
- It required scarce talent
If AI reduces:
- Time-to-analysis
- Human labor requirements
- Dependency mapping complexity
Then projects that once seemed prohibitive may now become feasible.
This could expand the total addressable market rather than shrink it.
Threat or Acceleration?
The central question facing IBM and the broader consulting ecosystem is this:
Does AI-driven COBOL modernisation threaten existing revenue streams — or accelerate transformation work already underway?
There are two competing narratives:
1. Disruption Narrative
AI compresses consulting timelines, reducing billable hours and eroding margins.
2. Acceleration Narrative
AI lowers barriers to entry, increasing demand for structured modernisation programs and enterprise-grade platforms.
History suggests technological accelerators often increase overall market activity.
When cloud computing reduced infrastructure complexity, enterprise IT spending did not disappear — it expanded.
AI may play a similar role in legacy system transformation.
The Bigger Shift: AI as a Systems Interpreter
COBOL modernisation is part of a broader AI trend: using machine intelligence to interpret complex legacy systems.
AI can now:
- Reverse engineer undocumented code
- Identify cross-system dependencies
- Surface hidden logic
- Suggest refactoring strategies
This capability extends beyond COBOL to other legacy languages and enterprise systems.
The real disruption may not be about COBOL alone — but about AI’s ability to unlock institutional knowledge trapped inside aging codebases.
Investor Anxiety vs. Enterprise Reality
Markets move faster than enterprise IT.
A 13% single-day stock drop reflects investor anxiety.
Enterprise transformation unfolds over years.
Banks, governments, and insurers do not rip out mission-critical systems overnight.
Even if AI shortens modernisation timelines, procurement cycles, compliance reviews, and integration planning remain lengthy processes.
The tension between market volatility and enterprise conservatism often creates overshooting reactions.
The Future of COBOL in an AI Era
COBOL is unlikely to disappear overnight.
Instead, AI may usher in a hybrid future where:
- Some systems are rewritten
- Others are encapsulated behind APIs
- Some remain untouched but better understood
- AI continuously monitors and documents legacy logic
In this scenario, AI becomes a co-pilot for legacy infrastructure rather than its executioner.
Conclusion: A Shortcut, Not a Silver Bullet
AI has undeniably created a shortcut in COBOL modernisation.
Tools like Claude Code and IBM’s watsonx Code Assistant reduce the time and labor required to analyze and refactor legacy systems.
But modernisation is more than translation.
It involves infrastructure decisions, security considerations, regulatory compliance, and operational risk management.
The market’s sharp reaction reflects fear of revenue compression.
The longer-term outcome may be more nuanced: AI could expand modernisation demand while reshaping how consulting firms deliver value.
COBOL may be old, but it still powers the global economy.
AI is not erasing that reality — it is finally making it economically manageable.
Whether that becomes a threat or an opportunity depends less on the code — and more on who leads the transformation.