LLMs: Great Code Generators, Terrible Software Engineers

Years of interviewing software engineers reveals that building and maintaining clear mental models is key. While LLMs are good at generating and modifying code, they lack the crucial ability to maintain these models. They easily get confused, suffer from context omission and recency bias, and hallucinate details, preventing iterative problem-solving for complex tasks. The author concludes that LLMs are helpful tools for software engineers but cannot yet replace them for anything beyond simple projects.