Left to its own devices, the AI may choose to implement things efficiently and correctly, or it may choose to implement things inefficiently and incorrectly. Without clear guidance, acceptance criteria, and hand-holding, it will make assumptions that are wrong, and as the demands of the product grow, it will continue to build off of those wrong assumptions. This is my experience with Gemini, Claude, GPT -- any of the models. They need guidance or they will build crap.
Now, this may be acceptable (tons of businesses are built on broken products), but what is absolutely cope is this moronic idea that the LLM is always smarter than the average engineer and will magically produce the most efficient and correct solution.
Left to its own devices, the AI may choose to implement things efficiently and correctly, or it may choose to implement things inefficiently and incorrectly. Without clear guidance, acceptance criteria, and hand-holding, it will make assumptions that are wrong, and as the demands of the product grow, it will continue to build off of those wrong assumptions. This is my experience with Gemini, Claude, GPT -- any of the models. They need guidance or they will build crap.
Now, this may be acceptable (tons of businesses are built on broken products), but what is absolutely cope is this moronic idea that the LLM is always smarter than the average engineer and will magically produce the most efficient and correct solution.