Big-bang rewrites fail because they require a team to rebuild years of accumulated behaviour before anything can be released. The strangler-fig pattern avoids that: new components take over one capability at a time while the old system keeps running, until nothing depends on it.
Steps
- Put a routing layer in front of the legacy system so traffic can be directed per capability.
- Pick a first slice that is valuable and well bounded.
- Extract its business rules, using AI to read the old code and summarize behaviour, then have an engineer verify each rule against reality.
- Build the replacement and run both in parallel, comparing outputs.
- Switch traffic gradually, keep a fast rollback, and retire the old slice only after a stable period.
AI is good at reading and summarizing old code. It is not a substitute for tests that prove the new component behaves like the old one.
Why it works
Each step delivers value and can be reversed. Risk is spread over many small changes instead of concentrated in a single cutover, and the team learns the system's real behaviour as it goes.