The instinct, and why it is a detour
People who want to get good at AI almost always start in the same place: looking for the right course.
It is the responsible instinct. It is also mostly a detour.
The people who get genuinely fluent did not take a better class. They took one real problem they already had — a report they dread, a decision they keep circling, a task they redo every week — and worked it with AI until the output was actually useful. Then they did the next one.
Two kinds of knowledge, and only one compounds
Codified knowledge is the teachable kind: prompt frameworks, "ten prompts every professional should know," a tidy pattern list. Easy to publish, easy to forget, precisely because it is not attached to anything you do.
Contextual skill is the other kind. Knowing how to brief AI on your situation. Knowing what its answers get wrong on your tasks specifically. Knowing where it saves you an hour and where it quietly costs you one.
Only the second kind compounds. And it only builds in contact with real work.
Which raises the question the advice usually skips: which task?
Three things make one worth two weeks of your attention.
You do it often enough that improvement pays back. Once a quarter is not practice, it is an event.
You can tell good from bad without asking anyone. If you need three people to agree the output is right, you will never get enough reps to learn anything.
And it has a shape you can describe. "Write better" is not a task. "Turn a messy client thread into three decisions and an owner for each" is.
What the two weeks actually looks like
Less than people expect. You are not blocking out time. You are refusing to do that one task the old way. Each time it comes up it goes through AI first, and you keep a two-line note: what you asked for, and what you had to fix. Nothing more formal than that.
By the end of the fortnight the notes are the useful artifact. The same correction turns up four or five times, and that repeated correction is the thing you now know that no course could have told you — because it is about your work, not about the tool.
The failure mode is picking the wrong task, and it is always the same one: something important but rare, where you cannot judge the output on your own. That task will teach you nothing in a fortnight. Pick the thing you redo most and least enjoy.
Your own AI, your real problem, and someone in the room with you: See the sessions →
Keep reading: Codified vs Contextual · The Briefing Pattern
This is the thinking behind a Trusted AI Culture.