Anthropic just launched Claude Code, an AI tool presented as able to accelerate COBOL application modernization: dependency mapping, workflow documentation, risk identification… with a promise that makes CTOs salivate: move a project from years to quarters. The market reacted: on February 24, 2026 IBM shares plunged 13% (roughly ? billion of market value wiped out, by circulated estimates). Behind the noise there’s a real business question for SMEs and mid‑market enterprises still running COBOL in finance, healthcare, insurance or quasi‑public environments: is this finally the right time to exit legacy without hiring near‑unobtainable COBOL talent?
The Opportunity for SMEs
If you run a COBOL core that handles billing, contract management, accounting flows or batch processing, the upside isn’t “AI writes code.” The upside is the ROI of clarity.
- Speed up a true scope audit: many projects fail because nobody really knows what the application does or what depends on what. A tool like Claude Code can reveal the topology (programs, calls, dependencies) and produce actionable documentation.
- Reduce reliance on scarce expertise: COBOL specialists are expensive, rare, and usually fully committed. AI can ease pressure on those resources by automating parts of analysis and transformation.
- Create a path to more hireable technologies: modernization isn’t just translation. But if the tool helps prepare a trajectory (API wrappers, migration to Java, service exposure), you regain agility: faster integrations, better observability, and less technical debt on aging infrastructure.
Key point: Claude Code’s pricing is not publicly disclosed yet. The pragmatic approach is to cost a pilot first and compare it to the bill for “consultant‑years.”
Where to Be Cautious
IBM is right on one point: code translation ≠ modernization. That’s where budgets hide.
- Underestimated complexity: your system is more than a set of source files. It runs with middleware, transaction processing, job schedulers, DR/BCP, and security controls accumulated over 30–50 years. None of that “disappears” because programs were refactored.
- Extensive functional validation: in finance and healthcare you don’t sign off on “probably correct.” An LLM remains probabilistic. UAT, regression testing and audits can take months and often cost more than the rewrite itself.
- Interconnections: modernizing a single system “in isolation” is sometimes impossible. COBOL frequently sits at the center of an application network, data flows and legacy formats.
- Lock‑in risk: depending on a single tool plus potentially significant API costs over a long project is a governance issue, not just a technical detail.
Compliance Considerations
If your COBOL applications process personal data (common in healthcare, finance, insurance and the public sector), modernization becomes a GDPR/nLPD compliance project before you even introduce AI.
- GDPR (EU) / nLPD (Switzerland): a clean migration requires a data map, a strategy for pseudonymization/encryption, and an end‑to‑end audit trail. COBOL histories often contain uncontrolled legacy fields.
- AI Act: if the tool participates in processes that support decisions affecting people, you may cross into “high‑risk” obligations (impact assessments, controls, traceability).
- Data sovereignty & hosting: a US provider forces verification of data flows, legal bases and target architecture. Post‑modernization hosting can be moved to more sovereign options depending on context (e.g. Infomaniak, OVHcloud, Scaleway) or to regionally appropriate cloud regions (e.g. AWS Zurich/Paris), but only after a proper audit.
Conclusion & Cohesium Support
Claude Code highlights a critical truth: COBOL can be modernized, but the value is not in literal “translation.” It’s in reducing uncertainty (scope, risks, dependencies) and in controlling the execution (testing, compliance, target architecture).
Rather than patching things together, Cohesium AI can support you with a package: 'COBOL Modernization Audit + Data Compliance' — a feasibility audit (Claude Code applicability, real cost estimates, timeline, hidden dependencies), personal data mapping and GDPR/nLPD plan, followed by custom development post‑migration (API wrappers, functional validation, and RAG to reconstruct forgotten business logic).
Contact us