The call went out: ten thousand product descriptions needed rewriting, and the deadline wasn't moving.
I started the only way I knew how — by hand, one at a time. About three minutes each once you're warmed up and you've stopped rereading your own sentences. Ten thousand of those is five hundred hours. Twelve weeks of somebody doing nothing else.
I ran that math somewhere in the first afternoon. It was obvious we were going to blow past the date.
So I went and found a colleague on the data side and asked him straight: couldn't AI get us there faster? I'd been doing prompt engineering on the side. He'd been doing more of it than I had.
He said yes — with the right prompts. Feed it the format. Teach it what good looks like. And the moment he said that, the other half landed for me: if we could generate them at that quality, I could import the results straight into the client's store instead of anybody re-keying ten thousand of anything.
It took a few cycles to get right. This was early days, back when most of the department wanted nothing to do with any of it.
It worked. We cut the time roughly in half and hit a deadline that, an hour earlier, had looked impossible.