Work / Software modernisation
Old system out, new system in, same answers.
Legacy code rebuilt on modern technology with AI assistance, and switched over only when its outputs matched the original.
The situation
A business-critical system had been running for years on technology that was reaching the end of its supported life. It worked, which was precisely the difficulty: years of small decisions were buried in code that nobody could safely change.
The business set one condition for the move: the new system had to behave exactly like the old one, and that had to be demonstrated, module by module.
What we built
We used AI to read and document what each part of the legacy code really did, then rebuilt it module by module on a current stack. Alongside the new code we built a comparison harness that feeds identical inputs to both systems and reports every difference in their outputs.
Each difference was traced to its cause and resolved, and every undocumented rule the business relied on was carried into the new system.
What changed
Modules went live one at a time, each only after its outputs matched the original. The business moved to a maintainable, supported platform and kept the comparison harness as a permanent regression test.
- Starting point
- Long-lived legacy codebase
- Method
- AI-assisted rebuild with side-by-side output matching
- Switch-over
- Module by module
- Acceptance test
- Identical outputs on identical inputs
- Left behind
- A permanent regression suite
The pattern behind it
Legacy Modernisation: Ageing code rebuilt on a modern stack, with every output checked against the original until they match.
The workflow
Rebuilt in pieces, verified against the original.
- Legacy systemThe code that works, that nobody wants to touch.
- AI step: Read and mapThe layer documents what each part actually does, including the surprises.
- AI step: RebuildModules are rewritten on a current, supported stack.
- AI step: Match outputsOld and new run on the same inputs; every difference is investigated.
- Switch overThe new module goes live only when its results are identical.
Amber steps are the AI layer. Everything else is the system you already have.
Next step
Want the same result?
Tell us what your version of this problem looks like, and we will map the route to the same result.