Raw LLM Responses

We were recently provided with 2 LLMs at work that are meant to be "specialized for coding" and neither of them is capable of putting out even the most basic code. Its extra painful when you ask it for code that is directly present in the Knowledge Base and it returns to you some reinvention of the wheel that looks very similar to the code from the knowledge base that achieves what you want but somehow it managed to make it entirely nonfunctional. Our top dev (34 years at this company) has given it numerous prompts and every prompt returned something that *looked* like good code. In every single instance, he would spend time fixing compilation errors before coming to figure out that the code was fundamentally flawed and had to be thrown away. AI is in the same place VR was ~8 years ago where all these investors think it is the next best thing when in reality the technology is not there yet for it to do any of the things they envision