*What AI engineers think they do:* - design a clear prompt for the LLM - connect the LLM to the frontend - sit back while the LLM does its magic *What AI engineers actually do* : - change the prompt because for some reason the llm is giving you the recipe to create pancakes. - turn down the content filtering because the llm is acting more prudent than the pope. - ask the llm to be more concise because you don’t want to read a 2-page monologue. - deploy a second llm because you’ve already reached the token limit (was the token limit always that low?) - put the most important rules in UPPERCASE so the llm knows you’re being serious - bump up the API version (because you heard the newest API is much better) - add more rules to the prompt because god forbid you get prompt hacked - Go back to the original prompt because now your LLM is confused