Much of my work day is spent immersed in various artificial intelligence and Large Language Model (LLM) tools. Some are internal and task-specific, some are approved commercial tools like Gemini and Claude Code.
I found myself thinking: prompting is like dog training.
I provide the AI tool information, and prompt it for what, and refine the information and prompts in order to get what I want, and continually proofing the responses.
Dog training works in a similar manner.
You provide the dog information, prompt it, refine, and reward the dog when it performs the action you want. And you are continually proofing the behaviors.
The AI tool rewards you, while you reward the dog, but getting to the reward is similar. The AI tool, like the dog, doesn’t know what I want until I tell it, and sometimes even then it doesn’t understand. There have been times when I’ve been fighting with an AI tool, and it has felt like trying to recall my dog as an adolescent when he suddenly developed selective hearing.
Dogs and AI also never grow up, so to speak.
My dog will never go to college, get a job, move out, and live on his own. Neither will AI. They won’t become full fledged, independent adults. They may always need supervision, and continual proofing in various environments. Sometimes they will excel. Sometimes you find their “thing”. They may have a tantrum, or a meltdown. Sometimes something outside your control rubs it the wrong way and you have to figure out how to deal with it, and then reconstruct what happened to figure out what training to implement to be better prepared.
Dog training has taught me many things, given me a variety of new skills. Having those skills also be beneficial for AI prompting, and working with AI tools, is an unexpected benefit.