Hand someone an AI agent and watch what happens. Inside a week they are good at handing off work. That skill is easy and everybody picks it up. Then watch them the next Tuesday, typing the same correction they typed the Tuesday before, to an agent that has no memory of the first conversation.
I have a line I have been repeating since long before any of this existed.
Don't be worried about making mistakes. Be worried about making them again.
That rule works on agents too. Almost nobody uses it on them.
Watch how it happens. You ask for something. What comes back is wrong in one specific way. You explain the way it is wrong, it fixes it, you get your result, and you close the tab. Everything you just taught it dies with the session. Next week, same correction. The week after that, same correction. You are not building anything. You are paying tuition again for a class you already passed.
I am not writing this from moral high ground. I made the same correction four times in a week before I noticed I was the problem.
The fix is not clever, which is probably why a lot of people skip it. When you correct your agent, write the correction down where the agent reads it every single time. A rules file. A skill. Whatever your tool calls that. The bar is embarrassingly low. If you had to say it twice, it should have been written down after the first time.
I have an assistant I call Mike. He knows my voice, my writing rules, the programs I run, and the history of what I am working on. None of that came from a clever prompt. It came from me correcting him and telling him to write down every correction so I would never have to make it again. He is not smarter than your agent. He has just been taught to remember, and yours has not.
So the people pulling ahead right now are not the ones with the best prompts. They are not the ones on the newest model either. They are the ones whose agents know something this month that they did not know last month.
You already know which correction you keep making. Now go write the rule.
Originally posted on LinkedIn: "Everybody Delegates. Nobody Teaches"