Most people hear reassurance in this. Relax — it only gets better. But Mollick is not consoling you. He is telling you where you are standing: on the floor, not the ceiling. And the floor is the worst place to form a permanent opinion.

Here is what the line catches. You tried the tool once — for a memo, a model, a block of code in your own domain — and it handed you something generic. So you filed a verdict: it can't do our work. The verdict felt final. The tool it was about shipped three versions ago.

The trap is not that you were wrong. It probably was weak at that task in March. The trap is that the verdict outlived the version. You stopped testing the edge, so you never noticed it move. Meanwhile someone with less domain skill and more patience is now doing in twenty minutes what still costs you an afternoon — not because they are smarter, but because they kept their judgment current.

This is why the skill that compounds is not prompting. It is re-checking. The model is a moving target; the durable asset is the habit of pointing it, again, at the thing you last decided it couldn't do.

So today: take one task you wrote off months ago — the one where you concluded AI can't handle this — and give it twenty real minutes. Not to be impressed. To re-date your verdict. Because the only opinion worth holding about this tool is one you formed this week.