After the Basics
There is a stage where being bad at using AI is completely normal. You do not yet understand context windows, message structure, tool calls, or where the model is strong and where it is brittle. In that phase, poor results are mostly a literacy problem.
But once you get past the mechanics, the excuses start sounding familiar. The model does not get what I mean. It keeps going in the wrong direction. It makes avoidable mistakes. I have to repeat myself. I asked for something simple and got something useless.
Sometimes the model is the problem. A lot of the time, the person using it is vague, impatient, unclear, or incapable of giving good direction. They want the output without doing the management work that makes good output possible.
This Is Delegation
Working well with AI uses many of the same muscles as working well with people. You need to provide context instead of assuming shared understanding. You need to define outcomes instead of throwing over half-formed requests. You need to know where ambiguity will create failure. You need to review the work at the right level and correct course without rewriting everything yourself out of frustration.
That is delegation. People like to pretend they are evaluating a machine when they are often revealing something about their own operating style.
The same pattern shows up with junior hires and cross-functional teams. Bad managers give thin instructions and expect mind-reading. They confuse tasks with outcomes. They fail to separate what is essential from what is optional. They give feedback too late or too vaguely, then conclude the other side is incompetent. AI exposes the same failure pattern faster because the loop is shorter.
AI Removes the Human Buffer
A good employee will often compensate for bad direction. They infer intent, read the room, ask clarifying questions, and use social judgment to fill in gaps. That can hide the fact that the original instruction was weak.
AI does much less of that. It reflects your ambiguity back at you. If your request is underspecified, your priorities are muddy, or your mental model is weak, the model will often make that painfully visible.
This is why some people find AI so frustrating. It removes the human buffer that used to cover for sloppy thinking.
Good Users Are Often Good Managers
The people who become good with AI are rarely good only at prompts. Usually they are already good at working through other people. They know how to frame a problem, decide what information matters, break work into pieces without losing the goal, and inspect a draft without overreacting to every imperfection.
Those are not AI-native skills. They are management, communication, and judgment skills. The interface changed, but the underlying discipline did not.
This does not mean every frustration with AI is your fault. Models hallucinate. Tooling breaks. Context gets stale. There are technical limits that no amount of good direction will solve. But after a certain point, blaming the tool for every bad result becomes a way of avoiding a more uncomfortable conclusion: maybe you are not as good at directing work as you thought.
The Uncomfortable Part
Once you understand the basics, being good at AI is less about magic phrasing and more about whether you can lead work clearly.
That is why I think AI is basically a people skill. If you are consistently bad at working with AI after the mechanics are no longer the issue, there is a decent chance you are also bad at working with people. Not because humans and models are the same. Because the same weaknesses show up in both places.