The 'AI-generated pseudo-spec' is becoming a massive productivity killer. Non-technical managers prompt an LLM, get a highly confident 5-page document with chosen libraries, and think they've done 80% of the work. But as a developer, you now have to spend hours debugging and debunking hallucinated architectures just to prove why the AI's approach doesn't fit the existing codebase. It completely shifts the burden of actual product management into AI-babysitting for the dev team.