AI
Where judgment still beats automation.
I’ve rebuilt a lot of my creative practice around AI in the last two years. It’s made me faster at almost everything — and clearer than ever about the parts of the work no tool can do for you.
7 min read
I’m not an AI skeptic. I’ve spent the last couple of years rebuilding how I work around these tools — exploring directions, drafting, pressure-testing ideas, getting from a thought to something I can actually look at in a fraction of the time it used to take. I lead creative, and I now do a meaningful amount of the production scaffolding with AI in the loop. So when I say there are places where human judgment still wins decisively, I’m not defending territory. I’m reporting from inside the change.
The honest picture isn’t “AI replaces creatives” or “AI can’t do real creative work.” Both of those are comfortable stories told by people who haven’t actually changed how they work. The real picture is more interesting and more demanding: the tools have gotten extraordinary at production and remain unable to do the things that were always the actual job. Knowing the difference — precisely, not vaguely — is becoming the core skill of creative leadership.
Here’s where I’ve landed on what to hand over and what to hold.
Automate the distance, not the decision
The most useful frame I’ve found is to separate the distance between an idea and seeing it from the decision about whether the idea is any good.
The distance is where AI is genuinely transformative. It used to take a week and three people to put ten real options in front of me. Now it takes an afternoon. That’s not a small efficiency — it changes the shape of the creative process, because the bottleneck moves. When generating options is cheap, the scarce resource is no longer production. It’s the judgment to know which option is actually good, and the conviction to kill the other nine.
And that decision — which one, and why, and what we’re not going to do — is exactly the thing the tools can’t make. They can give me a hundred competent directions. They cannot tell me which one is true to the brand, right for this moment, and worth betting on. That requires taste, context, and accountability, and none of those three live in the model.
So my rule of thumb is simple: automate the distance, never the decision. Use the tools to get to the choice faster. Make the choice yourself.
Taste is the bottleneck now, and that’s a feature
Here’s the thing that’s surprised people who assumed AI would commoditize creative: when everyone can produce competent, on-trend, plausible work instantly, competent work stops being worth anything. The floor came up for everyone at once. The obvious version of any idea is now free, which means the obvious version is now worthless.
What’s left — the only thing with value — is knowing which version is actually good. That was always the scarce thing. It’s just that production used to be hard enough that being able to make something was itself a moat. That moat is mostly gone. The remaining moat is judgment: the ability to look at a wall of plausible options and know, fast, which one has something the others don’t, and to articulate why so a team can move.
This is genuinely good news for the people who have that judgment and genuinely bad news for people who built their value on production speed alone. The tools didn’t kill creative skill. They killed the substitute for creative skill, which was just being faster with the software than the next person.
Where human judgment still wins, specifically
Let me be concrete, because “human judgment” can sound like a comforting abstraction. Here is where, in my actual work, the human is still decisively better — not sentimentally, but functionally.
Knowing what’s actually good. A model can rank against patterns in its training. It cannot stand behind a claim that this is the right answer for this brand at this moment, because it has no skin in the outcome and no real model of the specific human you’re trying to reach. Evaluation against taste is still ours.
Knowing what to make in the first place. The tools are answer machines. The hardest creative work is upstream of the answer — framing the right problem, asking the question nobody asked, deciding what’s worth doing at all. AI is brilliant at executing a brief and has no opinion about whether the brief is any good. The brief is where the leverage is.
Editing — the courage to subtract. Generation is additive and the tools are wildly generous with it; they’ll give you more, always more. But most good creative work is made in the cutting, and knowing what to remove — which excellent element is actually hurting the whole — is a judgment about the entire composition that the tools don’t hold. Restraint is a human act.
Context the model can’t see. The room. The politics. The thing the CEO said offhand that’s really driving the project. The cultural moment this will land in. What we tried two years ago that failed and why. Real decisions are made inside a context that mostly isn’t written down anywhere the model can reach, and that context routinely flips which answer is correct.
Accountability. This one is underrated and maybe the most important. Someone has to be answerable for the work — to defend it, to take the hit when it’s wrong, to have judgment that improves because they live with consequences. A tool cannot be accountable. The instant a decision matters, you need a human name attached to it, and that human had better have the judgment to have made it well.
The trap I watch for
The failure mode I see most often isn’t people refusing to use the tools. It’s people using them and quietly letting the tools make the decisions — accepting the first plausible output because it’s there and it’s fine and pushing back takes energy. This is the real risk, and it’s subtle, because nothing looks wrong. The work is competent. It ships. It just has no point of view, because no one actually decided anything; they curated defaults.
The discipline that protects against this is almost old-fashioned: keep the standard exactly where it was. The tools lower the cost of producing work; they do not lower the bar for good. If anything you should raise it, because you can afford to — when exploration is cheap, settling is inexcusable. The teams that win with AI aren’t the ones that produce the most. They’re the ones that use the new speed to get to a higher standard, not a lower-effort version of the old one.
What this means for how I lead
Practically, it’s changed where I spend my own attention and my team’s. Less of it goes to producing the first version of things, because that’s cheap now. Much more of it goes to the decisions — what are we actually making, is it good, is it true, what are we cutting, is it right for this moment. The work has shifted up the stack toward judgment, which is exactly where creative leadership always should have been concentrated and often wasn’t, because production was eating everyone’s day.
I tell my teams a version of this: let the tools do the distance, and spend the time you save on the decision. The companies and the people who internalize that won’t be the ones with the most automation. They’ll be the ones with the best judgment, finally freed up to use it.
The tools got astonishing. The job didn’t change. It just got more concentrated around the part that was always hard.
