AI is good at means.

How to structure the file. What the error is actually saying. Which of three approaches is the ordinary one. It answers these well, and usually faster than the afternoon it would have taken to find out.

None of what follows is a complaint about that. The tool is good at the thing it is good at, and pretending otherwise would be its own kind of dishonesty.

But the questions all arrive in the same shape.

How do I write this, and whether to write it. Which framework, and whether to build the thing at all. What to put in the message, and whether to send it. One of those is a means question. The other one is not.

Nothing in the exchange marks the difference. Both get answered in the same voice, at the same length, with the same evenness. And the evenness is the part that does the work. A boundary was crossed somewhere in there and no bell rang.

A handover would be easy to catch if it felt like surrender. It doesn’t. It feels like efficiency, which is the same thing that made the tool worth opening in the first place.

And the answer that comes back is usually reasonable. That is not the flaw in this; that is the mechanism. An unreasonable answer would get argued with.

Means can be handed over without anything being lost. That is more or less what means are for. Someone who does not know how a thing is done, and finds out, has lost nothing at all. They have just skipped the part where not knowing was the obstacle.

Direction sits differently. Not because handing it over is forbidden, and not because a person is the better forecaster. Often they are not. It sits differently because there is nobody else standing in the place the question is being asked from.

Where that line actually falls is less obvious than the last few paragraphs made it sound.

Asking what the options are is a means question. Choosing between them is not. But a list of options has already done a share of the choosing, in what it left off and in the order things arrived. The framing shows up before the decision does, and it shows up from outside. By the time the choice feels like a choice, some of it has been made.

There is no clean marker here for the moment one becomes the other. That seems worth leaving stated rather than closing with a rule that would read better than it holds.

What is left is not a policy about when to ask.

It is closer to noticing the pull before following it, including the pull toward an answer that is already, helpfully, on its way.

Not to refuse the answer. Just to know which question got handed over.