- The user tells the agent to add feature X

- The agent adds the feature

- User: "Yeah that's good but I don't think we need Y that you added, get rid of it"

- The agent removes Y, adds comments that Y was removed, adds backwards-compatibility for saves which used Y (obviously there aren't any, since Y only existed for 5 minutes and was never deployed - the project itself might never have been deployed either), adds a test for Y's non-existence, adds text to the UI saying "Note: Y has been removed. There are no traces of Y in the project"

I think that GPT-5.5 does it the most.

I've heard the argument that leaving these comments in, helps support future AI code generation. In this example, it was important to you at the time you made the change that create_* functions have defaults, so having this in the code captures that knowledge for later agents. It's similar to a more general argument that you should leave (correct) AI-generated code alone because it will be easier for the AI to "understand" code it generated that your modified version of it.

I see the validity in these arguments, but I don't think we should be so deferential to these models. The first part of this comment ("Now create_background params have default values") is completely redundant with the function signature and as no place in the code. The second part ("the same as create_screen") is genuine knowledge that is worth capturing for the agent, but it should be captured at a higher level doc about the codebase, rather than tacked onto some arbitrary function as a comment.

Another idea was Chain of Thought leakage. It's like as if a chef saying "Now I am adding salt because the customer requested so"