Why Has COBOL Suddenly Become So Critical in the AI Era?

Rather than replacing COBOL systems outright, banks are increasingly building artificial intelligence layers around them. AI tools are being used to interpret legacy code, document system behavior, and assist engineers in understanding complex transaction flows. This allows institutions to work more efficiently within existing infrastructure while reducing dependence on individual expertise. Large financial institutions, including JPMorgan Chase, Citigroup, and others continue to operate COBOL-based systems for core processing while simultaneously deploying AI tools for modernization support. The payments industry shows a similar pattern. Companies such as Visa and American Express continue to rely on COBOL-linked infrastructure for significant portions of transaction processing, particularly in backend settlement and authorization systems. What is emerging is a layered architecture in which COBOL remains the execution engine, while artificial intelligence becomes the interpretive and maintenance layer.

Every time you swipe an ATM card, check your deposit balance, or make a credit card payment, you are likely interacting with code written in a programming language created in 1959. That language is COBOL (Common Business-Oriented Language), and it is powering an estimated 95 percent of all ATM transactions worldwide.

Now, as U.S. banks rush to integrate artificial intelligence into their operations, COBOL has suddenly become more important, not because it is new, but because it is the bedrock holding up the entire financial system.

The Scale of COBOL’s Dominance

More than 43 percent of U.S. banks still rely on COBOL in their legacy IT systems. The financial sector processes approximately $3 trillion in daily transactions through COBOL systems. Major institutions including JPMorgan Chase, Citi Group, American Express, and Visa continue to lean on COBOL to handle their daily transactions.

Statistic	Figure
U.S. banks using COBOL	43%
Global banking systems powered by COBOL	43%
ATM operations powered by COBOL	95%
In-person credit card transactions	80%
Daily commerce generated	$3 billion+
New COBOL lines programmed yearly	1.5 billion

Source: IBM, Techchannel, Media Reports

These numbers explain why COBOL cannot simply be replaced. “The financial sector faces particularly high stakes,” noted a 2017 Reuters investigation into the aging systems. “Modern applications and tools are created using contemporary languages that must integrate seamlessly with legacy systems.”

What unites these institutions is not technological conservatism but operational constraint. COBOL systems remain embedded in environments where consistency, auditability, and transactional correctness are non-negotiable. These systems have been refined over decades, absorbing regulatory requirements, business logic, and exception handling rules that are difficult to replicate in full within newer architectures without introducing systemic risk.

As a result, modernization in the financial sector has largely taken the form of layering rather than replacement. Cloud platforms, APIs, and artificial intelligence systems are increasingly built around COBOL-based cores rather than in place of them. This creates a hybrid infrastructure in which modern interfaces depend on legacy engines to execute the underlying financial logic.

The scale of COBOL’s continued dominance therefore reflects a broader reality in global banking: the most critical systems are often not the newest, but the ones whose behavior has been validated across decades of real-world financial stress.

Why AI Is Making COBOL More Relevant

The rise of AI, particularly large language models (LLMs), has created a surprising opportunity: AI can help modernize COBOL systems instead of replacing them.

“We’re living in an AI-driven world, right? Everyone’s talking about how AI is going to sweep away all other technologies. But then, something happened that completely surprised me: I found out that a 70-year-old programming language is still thriving,” wrote tech analyst Prangya Priyadarsini in April 2025.

The real kicker? With the rise of Large Language Models (LLMs), we might actually have a way to modernize those legacy systems. Imagine AI helping to update the very system you thought you thought was outdated. It’s like discovering that your vintage car has a self-driving upgrade.

Generative AI will help engineers retool aging applications, overcoming a major modernization hurdle, according to Accenture. Banks are now deploying AI to translate COBOL code into modern languages like Java while preserving the business logic that has been tested for decades.

Large language models are increasingly able to parse legacy codebases, identify dependencies, and translate business logic into more modern programming structures. This has reframed COBOL not as an obstacle to AI adoption, but as a candidate for AI-assisted transformation.

What makes this development particularly consequential is that COBOL systems are not ordinary software applications. They encode decades of financial logic, including transaction rules, regulatory constraints, and exception handling procedures that have been refined through continuous production use. Direct replacement has historically been risky because even minor translation errors can propagate into financial discrepancies at scale.

Generative AI introduces a different approach. Rather than rewriting systems blindly, AI tools can assist engineers in mapping how legacy code functions, documenting its behavior, and generating partial translations into languages such as Java while preserving the underlying business logic. Technology consulting firms such as Accenture have described this as a shift from manual modernization to AI-assisted reengineering, where automation reduces the cognitive burden of working with deeply layered legacy systems.

In practical terms, this means banks are beginning to deploy AI systems not to eliminate COBOL, but to make it legible and maintainable at scale. Engineers can use AI tools to trace transaction flows across systems that were previously understood only by long-tenured specialists, or to simulate the effects of code changes before they are deployed into production environments.

The broader implication is that COBOL is no longer seen solely as a constraint on modernization. It is becoming part of the modernization process itself. AI does not remove the need for legacy systems; it provides new mechanisms for working within and around them.

The Talent Crisis

The renewed reliance on COBOL in global banking is being intensified by a second, less visible constraint: a shrinking pool of engineers who understand how these systems actually work.

At the center of this challenge is COBOL, a technology that continues to underpin critical banking infrastructure but is increasingly maintained by a workforce approaching retirement. The language’s longevity, once an advantage, has become a demographic liability. Many of the engineers who built and extended these systems over decades are now leaving the workforce, taking with them institutional knowledge that is often only partially documented.

This transition has been widely described in industry reporting as a generational exit from mission-critical computing environments. A Reuters analysis captured the sentiment with the headline “IT cowboys ride into sunset,” referring to veteran COBOL programmers reaching retirement age while still being called upon to stabilize legacy systems.

Among those highlighted was Bill Hinshaw, a 75-year-old COBOL specialist who, according to the report, divided his time between family life, including dozens of grandchildren and great-grandchildren, and urgent consulting work for U.S. companies seeking to prevent system failures in aging infrastructure. His profile reflected a broader reality across the financial sector, where a small cohort of experienced engineers continues to support systems that process trillions of dollars in transactions annually.

The underlying issue is not simply attrition, but the concentration of expertise in systems that were never designed to be temporary. Much of the logic embedded in COBOL-based banking platforms has evolved incrementally over decades, often without comprehensive documentation. As a result, understanding these systems requires not only familiarity with the language itself but also knowledge of historical business decisions, regulatory adjustments, and system patches layered over time.

This is where the talent gap becomes structurally significant. New generations of software engineers are typically trained in modern programming ecosystems such as cloud-native architectures, distributed systems, and microservices. COBOL, by contrast, is rarely taught in contemporary computer science curricula, leaving a gap between institutional need and available expertise.

The consequence is a growing dependence on a diminishing group of specialists who can interpret and safely modify core banking systems. In many institutions, this has turned COBOL expertise into a scarce operational asset, rather than a routine engineering skill.

That scarcity is one of the reasons artificial intelligence has become increasingly central to modernization strategies. Banks are not only using AI to accelerate code translation, but also to capture and preserve knowledge that might otherwise be lost as experienced engineers retire. AI-assisted tools are being deployed to document legacy systems, map dependencies, and simulate the effects of system changes before they are implemented in production environments.

The urgency is heightened by scale. Financial institutions continue to process trillions of dollars in daily transactions through these systems, meaning that delays in modernization or knowledge transfer carry immediate operational risk. In this context, AI is not positioned as a replacement for human expertise, but as a mechanism to extend it.

The result is a transitional moment in banking technology. As veteran COBOL engineers exit the workforce, banks are attempting to encode their knowledge into tools that can assist the next generation of developers.

Why COBOL Won’t Be Replaced Soon

The persistence of COBOL in US banking systems is often framed as technological inertia, but within financial engineering circles it is more accurately understood as a consequence of precision, stability, scalability, and transactional design choices that remain difficult to fully replicate in modern programming environments.

At the core of COBOL’s durability is its approach to numerical computation. Financial systems require arithmetic that is exact, not approximate, particularly when aggregating millions of transactions where even fractional rounding errors can accumulate into significant discrepancies. COBOL was designed with fixed-point decimal arithmetic as a native feature, allowing developers to represent and compute monetary values without the floating-point errors that can occur in many general-purpose languages. In banking environments, this property is not optional. It is foundational.

This emphasis on numerical precision is one reason banks continue to rely on COBOL for ledger systems, interest calculations, and settlement processes. As IBM explains in its mainframe documentation, COBOL provides “strong support for decimal arithmetic, which is essential for financial and business applications where accuracy is critical” 

Equally important is the language’s operational stability. COBOL systems are widely regarded as highly reliable in mission-critical environments, in part because they have been refined over decades of production use. In banking infrastructure, reliability is not measured in performance benchmarks alone but in uninterrupted uptime across regulatory cycles, market volatility, and peak transaction loads. Many of these systems have evolved incrementally rather than being replaced, resulting in platforms that are heavily tested under real-world conditions.

That stability is closely tied to how COBOL applications are deployed on mainframe environments, where transaction processing systems are engineered for continuous availability. IBM describes these systems as designed for “high-volume, mission-critical workloads with near-continuous uptime requirements,” a standard that modern distributed architectures often struggle to match at equivalent levels of predictability.

Scalability is another reason COBOL has remained embedded in financial infrastructure. Unlike many modern systems that require architectural redesign to handle large increases in transaction volume, COBOL-based applications running on mainframes can be scaled vertically and horizontally within the same ecosystem. This allows banks to process surges in demand, such as payroll cycles or market close periods, without rewriting core business logic.

Perhaps most overlooked is COBOL’s strength in file and batch processing. Large financial institutions operate on massive datasets that must be processed in structured cycles, including end-of-day reconciliation, account updates, and regulatory reporting. COBOL was designed for exactly this type of workload, with built-in mechanisms for handling sequential file processing and large-scale transaction streams efficiently and reliably.

Modern programming languages such as Python, Java, and Go excel in web services, distributed computing, and application development, but often rely on additional frameworks and infrastructure layers to replicate the deterministic batch-processing behavior that COBOL provides natively. In many banking environments, this means COBOL is not merely legacy code but a specialized tool optimized for a very specific class of financial computation.

The result is a paradox that defines much of today’s banking technology landscape. Even as institutions invest heavily in cloud computing, artificial intelligence, and microservices architectures, the core financial ledger systems that underpin global banking continue to rely on a language designed more than six decades ago.

The persistence of COBOL is therefore not a failure of modernization. It is a reflection of the fact that, in financial systems where correctness, auditability, and continuity are non-negotiable, older engineering decisions remain difficult to displace.

The Modernization Trap

The push to modernize legacy banking systems with artificial intelligence has introduced a new risk that industry analysts increasingly describe as a “translation trap,” a form of technological optimism that underestimates the complexity embedded in decades-old financial infrastructure.

The concern, highlighted in a June 2026 analysis in Retail Banker International, is that many modernization programs focus narrowly on converting legacy code written in COBOL into more contemporary languages such as Java or Python, without fully reconstructing the underlying business logic that the original systems were designed to enforce.

As the analysis noted, “When banks focus on swapping old code for newer versions in the bid to modernise their systems, they fall into the translation trap. The reality is more complex.” 

The warning reflects a growing recognition within financial technology circles that COBOL systems are not simply repositories of outdated syntax. They are, instead, dense accumulations of institutional decision-making. Over decades, these systems have encoded regulatory interpretations, exception handling rules, credit policies, and operational contingencies that are often only partially documented, and in some cases not documented at all.

In this context, a direct line-by-line translation of COBOL into a modern programming language can create a dangerous illusion of equivalence. While the resulting code may be syntactically correct, it may fail to preserve the nuanced financial logic embedded in legacy workflows. That gap can introduce subtle errors in areas such as compliance reporting, transaction reconciliation, and risk calculation.

Industry engineers caution that these failures are not always immediately visible. A system may appear to function correctly under normal conditions while diverging in edge cases that only surface under regulatory audits or rare transaction patterns. In banking environments, where precision and traceability are essential, such discrepancies can carry significant operational and legal consequences.

The broader implication is that modernization cannot be reduced to translation. Rewriting code without reconstructing intent risks breaking the implicit contracts that govern how financial institutions operate.

As one systems architect familiar with large-scale migration projects put it, “You are not translating a program. You are translating a history of financial decisions.”

The result is a growing shift in strategy. Rather than attempting full replacement, many institutions are now pairing artificial intelligence tools with legacy systems to map dependencies, document business logic, and simulate the effects of proposed changes before any code is rewritten.

How Banks Are Balancing AI with COBOL

Banks are not replacing COBOL so much as surrounding it with artificial intelligence systems designed to interpret, maintain, and gradually extend it. The result is a hybrid financial architecture in which decades-old transaction engines continue to run critical workloads while AI tools increasingly mediate how engineers interact with them.

At the center of this shift is a growing category of AI-assisted modernization tools, including systems developed by firms such as Accenture. These tools use generative AI to help engineers read, document, and partially refactor legacy COBOL codebases. Rather than rewriting systems outright, the AI acts as a translation layer, converting dense and often undocumented business logic into more readable forms, mapping dependencies, and accelerating maintenance work that previously required deep institutional memory.

In practice, this means that COBOL systems are not being retired but made more navigable. Engineers can ask AI systems to explain how a specific payment flow works or trace how a change in one module affects downstream reconciliation processes. This represents a shift from manual code archaeology to assisted system interpretation, although banks still rely on human verification for any production changes.

Large financial institutions, including JPMorgan Chase, Bank of America, and Citigroup continue to run core daily transaction processing on COBOL-based infrastructure while simultaneously investing in AI tools to support long-term modernization strategies. In these environments, COBOL remains responsible for core ledger updates and settlement logic, while AI systems are increasingly used for documentation, anomaly detection, and engineering support.

The same pattern extends across the payments industry. Companies such as Visa and American Express continue to depend on COBOL-based systems for a substantial portion of their transaction processing workloads. Industry estimates frequently cited in engineering discussions suggest that COBOL-based systems still underpin a significant share of daily payment flows, in some cases reported as high as 80 percent, particularly in backend settlement and authorization pipelines.

What is changing is not the underlying transaction layer but the interface around it. AI systems are increasingly being used to sit between engineers and legacy mainframes, translating business requirements into code changes, generating documentation for compliance audits, and assisting in impact analysis before system updates are deployed.

This approach reflects a practical constraint in banking technology. Core systems cannot be easily rewritten without introducing unacceptable operational risk, yet they must still evolve to meet regulatory, security, and performance demands. AI offers a way to reduce the friction of working within these constraints without removing them entirely.

The Bottom Line

COBOL is not becoming important because it is trendy. It is important because it is the invisible foundation of the U.S. financial system. As banks integrate AI, they are not replacing COBOL—they are using AI to understand, maintain, and gradually modernize the code that processes $3 trillion daily.

What is changing in the AI era is not the existence of COBOL, but its visibility. Artificial intelligence tools are making it easier for institutions to understand and manage codebases that were once navigable only through specialized institutional knowledge. Yet this increased visibility has also underscored how deeply these systems are embedded in financial operations, and how difficult they remain to replace without introducing unacceptable risk.

In that sense, COBOL’s importance is not diminishing. It is being redefined.

It is no longer simply a legacy language running in the background of financial systems. It is the operational foundation that modern systems must continue to interface with, even as banks build increasingly sophisticated AI layers on top of it.

Stay ahead of the stories shaping our world. Subscribe to Impact Newswire for timely, curated insights on global tech, business, and innovation all in one place.

Dive deeper into the future with the Cause Effect 4.0 Podcast, where we explore the ideas, trends, and technologies driving the global AI conversation.

Got a story to share? Pitch it to us at info@impactnews-wire.com and reach the right audience worldwide