Ilana MachadoLead, AI-driven digital experience
Pauline BertryCo-founder, Polar Bear · ex-McKinseyWhere creative judgment actually comes from, how to hand work to a model without handing over the thinking, and the systems that keep your people noticing things.
Look at a film from the 1950s and you know the decade in seconds. The 60s, the 70s, the 80s, the 90s: each era of the last century carried a look you could identify from a single frame. Then something changed. Set aside the phones and the laptops, and 2002 is hard to tell apart from 2025. The visual signature of an era, the thing decades used to have by default, has faded.
Part of the story is risk. Twenty years ago, agencies and brands took more of them. Websites were strange, campaigns were strange, and a lot of it failed in interesting ways. The work that survives from that period survives because someone tried something nobody had tried.
AI arrives into this already-flattened world with a specific danger. Generative tools produce the statistical average of everything they have seen, so a team that leans on them for the thinking itself will converge on the same output as every other team leaning the same way. The technology did not create the sameness problem, but it industrialized it.
The conclusion some creative leaders draw is to keep AI out. That conclusion has been drawn about every technology since the typewriter, which was accused of killing the penmanship of literature, and it has never once held. The tools arrive whether you boo them or not. The real question is what separates the teams whose creativity survives the tools from the teams whose creativity dissolves into them.
Creativity has a supply chain, and it starts outside the building.
It is exposure. Exposure to art, to music, to literature, to travel, to talented people whose thinking rubs off on you, to the awe of standing in front of something bigger than your discipline. And exposure to problems in the wild: one of us spent years sitting in concerts watching musicians stop mid-performance to flip sheet music, and could not stop redesigning the experience in her head. Why can't a screen listen to the music and turn the page itself? Nobody assigns that observation. You have to be out in the world for it to happen to you.
The best creative work in advertising proves the point. The campaigns with decades of life in them are built on a human insight someone noticed before anyone articulated it. You are not yourself when you are hungry: an observation about human behavior that carried Snickers through fifteen years of work across every channel. The media pushes women toward a manufactured ideal of beauty: an observation that gave a soap brand a purpose it still holds. Strategists dig for months to reach an insight like that, and the digging happens in research, in conversations, in culture, in noticing.
The raw material of creative work is lived attention, and lived attention is the one input you cannot generate.
No model produces that noticing. A model can tell you what people have already said about hunger and mood. It cannot sit in the concert hall, feel the interruption, and mind it.
Here is a working relationship with AI that protects the muscle instead of replacing it: treat the model as a junior assistant you are teaching to be better. A capable junior with infinite patience and a very large library, working for a senior who owns the judgment.
What do you hand a junior like that?
You have an idea. Before it goes near a client, throw it at the model the way you would throw it at a focus group or a sharp colleague. Test it against research, against analytics, against the obvious objections. Where are the holes? Where does it break? The idea stays yours; the stress test gets faster and cheaper.
One strong creative idea needs thirty-five executions to find out which one carries. Producing variation thirty-five used to eat a week of studio time. Now it eats an afternoon, and your team spends the recovered week on the next idea instead of the next resize.
Copy adaptation, versioning, formatting, the long tail of tasks that require hands but not much brain. Offload it. This is the same trade farmers made when machines took over harvesting: the ones who prospered used the freed hours to grow more, not to stop farming.
Notice what stays on the senior's desk in every case: the insight, the idea, the judgment about which of the thirty-five versions is alive. The moment you hand over the thinking rather than the tasks around the thinking, you have crossed from delegation into substitution, and substitution is where the muscle starts to go.
Creativity behaves like a muscle, and the analogy is uncomfortably literal. Strap a machine to your body to do the moving for you, and the muscle does not stay the same. It shrinks. The brain treats unused capacity the same way.
A creative professional who lets the model do the thinking is running that experiment on their own brain. Every idea you outsource is thinking your brain no longer practices. The decline is invisible for months, because the outputs keep arriving and the outputs look fine. Fine is the problem. Fine is the statistical average. The distance between your work and the average is the entire commercial value of a creative team, and it erodes a little more with every piece of thinking you hand over.
The counter-program is the same one that built the muscle: exposure, deliberately maintained. Go to the museum. Travel. Read outside your field. Watch what the influencers are doing, because the generation producing entertainment natively has lessons for seasoned professionals, not just the reverse. Sit in the concert hall and let the problems find you. None of this shows up on a utilization report, and all of it is load-bearing.
Whether AI becomes your enabler or your crutch depends on which parts of the work you keep for yourself.
Every technology in history has been either an enabler or a crutch, and never by its own choice. The typewriter did not decide what it would do to writing; writers did.
Individual discipline does not survive organizational pressure. If your studio runs at 95 percent utilization and rewards only shipped work, your people will use AI as a crutch, because the crutch is what the incentive structure pays for. Keeping creativity alive is a systems job. Four places to build:
The agencies that excel will not be the ones with the best tools. Everyone will have the same tools within a quarter of release. They will be the ones whose people still notice things.
Most of what this guide describes requires your people growing on purpose, consistently. Growth goals that ladder, mentorship that happens, check-ins that catch the drift before it becomes a resignation. That is the work that always gets scheduled and never gets done.
Ron does it every week. Ron is your AI Talent Manager. He lives in your Slack or Teams and makes sure the growth work actually happens:
Not by itself. The evidence (Doshi & Hauser, 2024) shows AI raises individual output quality while making work across teams significantly more similar. The danger is substitution — handing the model the thinking itself rather than the tasks around the thinking. Teams that keep the insight, the idea and the judgment in human hands keep their edge.
Use the junior assistant model: give AI the pressure‑testing of ideas, iteration at volume (variation thirty‑five in an afternoon instead of a week), and the menial production layer. The senior owns the insight, the idea, and the call on which version is alive.
When an external tool takes over a mental function, the brain reallocates away from it (Risko & Gilbert, 2016). Newer studies find frequent AI use correlates with less critical‑thinking effort (Gerlich, 2025; Lee et al., 2025). It matters because the decline is invisible while outputs keep arriving — and “fine” output is the statistical average, which is exactly what clients can get anywhere.
Four builds: AI goals that are specific and ladder from individual to organization; training in micro‑doses (the spacing effect beats quarterly seminars); mentorship where mentees come prepared; and test briefs — a micro‑project with and without the model — before a junior touches live client work with AI.