Legacy systems often run on rules nobody has documented. The code is one source; the people who know why it behaves as it does are another, and they are not permanent.
Approach
- List systems and the specific people who understand each.
- Interview with real scenarios: 'what happens when this batch fails?'
- Use AI to summarize code and job schedules, then have the expert correct the summary.
- Record decisions and workarounds, with examples.
- Store the result where the team will find it and keep it updated.
An expert correcting a draft is faster and more accurate than an expert writing from a blank page. Use AI for the draft, people for the truth.