Writing a new module with AI is relatively easy: you set the context and the rules. Opening a project that’s several years old, with incomplete documentation and logic grown out of dozens of compromises, is a completely different task.
In such places AI is especially useful as a guide: explaining the data flow, finding the entry points, building a dependency map. But if it confidently “understood” an implicit business rule wrongly, the price of a nice explanation can be high.
I’m curious where your line of trust is.