Comments are the cheap version of it and they leave a trail the next reader has to wade through. If the goal is to waste an agent's time, the better placement is in the boring middle of the code it has to reason about, where a wrong comment about an invariant costs it the same parse as a wrong line. The problem is that the same wrong comment costs a human the same parse, so the poisoning only pays if you can scope it to something a person would never read and an agent always will.