Raw LLM Responses

"Good enough" is itself an overstatement. They can produce boilerplate, but it's only good boilerplate if you're very specific about what you want. They can integrate with systems, but it's much harder for them to provide deep insight about them. I've seen a junior dev rely on AI too much - the result was bad. Their most useful function is honestly searching through a mess of a codebase, something that they are genuinely better at than me painstakingly going through things with grep. But if you have it propose modifications, you have no idea if they're correct. I've said this before: the difficulty of coding has never been in WPM on the keyboard.