when a problem is entirely self contained in a file, it's very easy to edit it with LLM.

that's not the case with a codebase, where things are littered around in tune with specific model of organisation the developer had in mind.