AI-assisted legacy modernisation
What is AI-assisted legacy modernisation?
Using AI to make sense of old systems and move them forward: reading decades-old COBOL or Java that nobody fully understands, recovering the business logic buried in it, then helping rewrite or migrate it. The hard part was never typing the new code. It was understanding the old. Specialist tools now pair code discovery with large language models for exactly this.
Why it matters
When AI reads an ancient system and hands back a clean account of what it does, the temptation is to believe it. That account is a hypothesis, not a transcript, and old systems are full of behaviour nobody meant. Decades of patches leave quirks that downstream processes have come to depend on, so a “bug” the model helpfully offers to fix may be load-bearing. The value is genuine. Recovering intent from code that has outlived everyone who understood it is hard, and the technology is good at a first draft of that understanding. But the draft has to be checked by people who know the business, not only the code, before anything is rewritten. Modernisation that trusts the machine’s reading wholesale swaps a system nobody understands for one nobody validated.
In practice
A model reads a mainframe module and produces a readable summary of its rules, the first anyone has had in years. Before a line is rewritten, an analyst who knows the business walks through it and finds that one “obvious error” is actually a deliberate rounding rule three other systems rely on. The summary was the breakthrough. Trusting it unchecked would have been the disaster.
