I don't think of it as prompting anymore. I think of it as transferring the actual weight I'm carrying onto whatever I'm working with, including AI.
Here's the pattern I keep noticing. Ask AI to do something specific — draft this, summarise that, produce a workpaper in the right shape — and you'll usually get something that looks right. It fills the shape you asked for. But something in your gut tells you it's off. Not wrong, exactly. Just missing whatever actually makes the thing matter.
I think the missing piece isn't more instruction. It's the tension. The part where you're not fully sure about something, or you know exactly which line someone's going to push back on, or you understand precisely what's riding on one detail. That's not information you'd normally think to hand over — it feels like your problem, sitting in your head, not an instruction you'd write down. But it's the one thing that actually changes what comes back.
It's worked well enough that a reviewer once described my AI-assisted documentation as "impressive use of Copilot" — which I'll take, though I think the credit belongs less to any particular prompt and more to whatever I chose to carry into it that day.
I'm still figuring out how far this generalises — this is just where my own thinking has landed, working inside a field where the bar for "good enough" doesn't move for anyone. But the shift, for me, has been from specifying tasks to transferring stress: not what needs to get done, but what needs to be held onto while it's getting done. A good associate works out the stakes on their own, eventually. Right now, AI mostly needs you to hand them over directly.
Maybe that changes. Maybe in a year this looks obvious, or wrong. But if you're getting technically correct output that still doesn't sit right, that's usually where I'd look first — not at the instructions you gave, but at the weight you kept to yourself.
Find me on LinkedIn.