20 years of code trapped in 40-year-old tech: how that chain is finally breaking

How AI is solving Public Administration's biggest bottleneck

A large portion of the software supporting critical public services is 20 years old. It is built on 40-year-old technology. Natural/Adabas, COBOL, Oracle Forms, legacy .NET. Business rules embedded directly in the code, with no documentation explaining them. Systems no one dares to touch because no one fully remembers why they do what they do.

20 years of software, 40 of technology: the gap nobody wants to inherit

This legacy code is not just a pending technical detail. It is an inheritance throttling digital transformation on three fronts:

• Cost. The talent who knows these systems is retiring. Maintenance costs are rising. Integration with modern APIs is nearly impossible.

• Gridlock. Public tenders demand what the system cannot deliver. Projects drag on for years. The result: it gets postponed, once again.

• Silent risk. Every year without modernization is another year of knowledge lost without ever being documented. Until now, the only alternative was brute force: 3-to-5-year projects, teams of hundreds of people, line-by-line code translation. Frequently, it ended in abandonment or blown budgets.

What was once insurmountable can now be done in 12 months

What has changed is not the willingness to modernize—that was always there—but the actual feasibility of doing so. AI shifts the focus from translating code to reconstructing functional specifications. It is not about converting a COBOL program line by line to Java, dragging along 20 years of patches. It is about understanding what the system does, for whom, and why—and rebuilding it with that logic on a modern framework. The question is no longer whether AI can migrate legacy systems. It is how to scale it in production.

It is not a tool. It is a modernization factory

Many AI initiatives remain isolated tools, lacking the discipline of an industrialized project. Axpe’s methodology rests on three pillars:

• Industrialized project. Five phases with formal deliverables by experts. It is not ad hoc software.

• Exportable to any Public Administration. Agnostic to both the source legacy system and the target framework.

• In production, not in pilot phase. A real-world use case since 2025 in the Government of Cantabria: software delivered across backend systems and mobile apps, written with AI under human supervision.

Five phases. Zero leaps in the dark

• F0 · Assessment — Legacy inventory and domain roadmap.

• F1 · AI Factory & Design — Target architecture and trained pilot agents.

• F2 · Production Pilot — Real-world migration of a single domain, with human validation.

• F3 · Domain-by-domain Scaling — Progressive migration, one domain at a time.

• F4 · Continuous Operation — Decommissioning of legacy systems, new platform live in production. It begins with what is self-contained and verifiable. Scaling only happens once the factory is calibrated with real data and results.

What we learned from migrating real-world systems

Four non-negotiable lessons drawn from the Cantabria experience:

• Segmentation by coupling. Start with the most decoupled modules. Payroll and accounting come last, once the factory is already calibrated.

• Functional parity, not structural parity. Success = same input/output as the current system. Internal design is not replicated—behavior is.

• Real data from day one. Without real client data, there is no reliable validation.

• Human expert at the core. AI writes, the expert decides. Without that supervision, the output is code lacking business logic.

Less risk, lower cost, faster speed — proven by results, not promises

Benefits already verified in production:

• Core legacy system modernized in 12 months

• Proven functional parity

• Lower risk and reduced costs

• Sustainable, scalable, and integrable platform

• Technological independence and accessible talent Pool