LLMs excel at simple coding tasks but are still too unreliable to use without extensive human supervision on complex tasks where mistakes are expensive.