Legacy is context maturity, not code age

Legacy is not defined by how old the code is. It is defined by how little the organisation has externalised about what the code does.

The teams know. The knowledge lives in people. Tribal conventions, implicit ownership, undocumented rules. That context works for human-driven delivery. It does not work for AI-assisted transformation.

A twenty-year COBOL estate has the same problem as a three-year microservices platform whose team rotated out. The variable is context maturity.

The current conversation about AI and code focuses on creating context forward. Better specs, better rules files, better ways to declare intent so AI can execute. For new code, that is the right focus.

Most enterprise code already exists. For those estates, the problem reverses. Context must be extracted from the code, not declared alongside it. Structure, ownership, coupling, business rules. All inferred. If the interpretation does not match what the code actually does, the interpretation is wrong.

Almost no one is building serious processes for this backward direction. Controlled AI processes that analyse code, dependencies, runtime behaviour, and documentation to produce attested, evidence-backed context.

That is where enterprise-scale AI leverage sits.