Comment Re: Good riddance (Score 3) 148
Based on the explicit instructions my various coworkers give and extensive documentation I routinely give to LLMS like Claude, Gemini, and whatever else my company has decided to spin up (we are a implementation / development partner for all the big "AI" shops), that is not going to help much. Much of what we do involves integrations with external platforms and exchanging data. Each and every single god damn one will be fed the data models, examples of existing, functional code, and given the API tools along with explicit instructions to make it as concise and performance tuned as possible, and it will fuck something up majorly.
The last example was LLM building a ETL function to compliment some patching information for existing servers. It shot out individual REST calls for each server in question in a for loop to build an array and then process the information in another for loop, when the API provides a call to pull all that information back in a single payload and pass to the ETL engine; something like 150 lines of code that is slow as hell and prone to timeouts versus re-using the existing code which was like 10-15 lines and much faster.