Most companies have systems that nobody dares touch. An ERP from 2009. An internal portal running PHP 5. A COBOL batch job that processes payroll every two weeks, written by someone who left the company years ago.
The typical reaction is shame. “We should have migrated ages ago.” Then one of two patterns follows: either an expensive, multi-year migration project – or nothing at all.
Both are costly.
The problem with “tear it down and rebuild”
Large migration projects fail more often than they succeed. This isn’t a controversial claim – it’s well documented. McKinsey estimated in 2020 that 70% of all digital transformation projects fail to achieve their goals. The Standish Group’s CHAOS report consistently shows that large IT projects have a success rate below 35%.
The reason is rarely technical. It’s organizational. A system that has been running for 15 years has accumulated thousands of implicit business rules – rules that nobody documented, because the system is the documentation. When you tear it down, you don’t just lose code. You lose knowledge.
The new system starts with a blank slate and spends the first two years rediscovering what the old system already knew.
AI changes the equation
Generative AI has introduced a third option: incremental modernization.
Instead of replacing an entire system, AI can be used to understand, document, and gradually transform existing code. Not theoretically – it’s already happening.
Understanding. Modern language models can read and explain legacy code at a scale that was unthinkable three years ago. COBOL, Fortran, Visual Basic 6, classic ASP – models like Claude and GPT-4 can not only parse the syntax but explain the business logic behind it. This means the knowledge buried in the code can be excavated and documented – automatically.
Translation. Code can be incrementally translated from one language to another, module by module. Not as a big-bang migration, but as a gradual process where each step can be tested and validated. Amazon did this internally: they migrated over 30,000 production systems from Java 8 to Java 17 using AI-assisted code transformation. The project saved them an estimated 4,500 developer-years.
Integration. Instead of replacing an old system, it can be encapsulated. An AI agent can function as a modern interface on top of a legacy system – accepting natural language, translating it into the API calls or database queries that the old system understands, and returning the response in a format the rest of the organization can use.
It’s not about technology
When a CTO says “we have a legacy problem,” it’s rarely a technical problem. It’s a knowledge problem. The systems work – they run payroll, handle orders, track inventory. The problem is that nobody understands how they work, and therefore nobody dares change them.
AI solves that problem directly. Not by replacing the systems, but by making them understandable again.
A Danish manufacturing company with a 20-year-old ERP system doesn’t need a new ERP system. It needs to know what the current system does, where it can be improved, and how it can be connected to the systems that have been added since.
That’s a task that takes weeks, not years. And it costs a fraction of a full migration.
Incremental over total
The most effective modernization strategy is almost never “throw it out and start over.” It’s:
- Understand what you have – let AI document the business logic in existing systems
- Prioritize what actually causes problems – not all legacy is bad
- Modernize incrementally – one module at a time, with tests ensuring nothing breaks
- Integrate instead of replace – use AI agents as bridges between old and new
The result isn’t a shiny new system. It’s a system that works, that the organization understands, and that can evolve with the business.
It’s not sexy. But it’s what actually works.
Have systems nobody dares touch? We help companies modernize incrementally – without pulling the foundation out from under them.
Read also:
