I had a conversation today where my AI assistant articulated something I’d been feeling but couldn’t name. It said: “The skill to stay sharp at is having the ideas and knowing which ones matter — that’s the consulting muscle. Crafting prose is logistics.” I immediately recognised it as important. I asked it to write a post about it. Then I paused and thought: wait, the idea came from the AI, not from me. So what exactly is my contribution here?
The answer, I think, is that I picked it. The AI said many things during that conversation. Most of them I let pass. That one I grabbed. The selection was mine, even if the articulation wasn’t. And the decision to publish it — to stake a small piece of my public identity on it — that was mine too.
This feels like it might be the central skill shift of the AI era, and it’s not the one most people are talking about.
The conversation around AI and work tends to focus on execution. Can AI write code? Draft emails? Produce analysis? The answer is increasingly yes, and the implication everyone draws is that execution skills are devaluing. Learn to prompt. Learn to supervise AI. Learn to be the human in the loop.
But supervising execution is not the scarce skill either. The scarce skill is knowing which execution is worth doing. Not “can AI write this post?” but “should this post exist?” Not “can AI analyse this dataset?” but “is this the dataset that matters?” Not “can AI draft this strategy?” but “is this the right strategy to be drafting?”
This is taste. Not in the aesthetic sense — though that’s part of it — but in the deeper sense of pattern recognition applied to value. The ability to look at ten possible directions and feel, before you can fully articulate why, that one of them is the real one. Editors have always had this. Venture capitalists live and die by it. A&R people at record labels built entire industries on it. What’s new is that taste is becoming the primary skill for everyone, not just people in explicitly curatorial roles.
The uncomfortable corollary is that taste can’t be sharpened by thinking about it. You can’t read your way to better taste, or deliberate your way there, or ask five AI models to help you converge on it. I literally ran that experiment today — five frontier models agreeing on an answer gave me more confidence but not more accuracy. Taste improves only through contact with reality. You make a selection, you ship it, you see what happens. The post that resonates teaches you something. The post that lands flat teaches you more. The post you never publish teaches you nothing.
This is why the shift from “I must write everything myself” to “I select and AI drafts” isn’t laziness — it’s a reallocation of effort toward the bottleneck. If I spend two hours crafting prose for one idea, I get one data point about whether my selection was good. If AI drafts and I spend ten minutes reviewing, I get ten data points in the same time. More selections tested means faster taste calibration.
There’s a risk of this becoming an excuse for low-quality volume. “I’m training my taste” can justify publishing anything. The check is simple: would you attach your name to it? Not “is it perfect?” but “does it represent a genuine selection — something you noticed and thought mattered?” If yes, ship it. If you’re just filling a pipeline, stop.
The people who will thrive in the AI era aren’t the best prompters, or the best supervisors, or even the best strategists. They’re the people with the best taste — the ones who can look at a sea of competently-generated options and say “that one.” And then have the nerve to ship it and find out if they were right.
In the AI era, your output is your selections, not your productions.