Halfway through the Maesa engagement, two designers on the same brand put their work side by side on the big screen. Same product, same launch, same week. One set looked like it came out of the brand team. The other looked like a very competent stranger had been handed the product and told to make it pretty.
The room assumed it was the model. One of them was on a slower, more expensive model, the other on a faster, cheaper one, and people had already started arguing about which one the team should standardize on. I asked both designers to paste what they'd actually typed. The difference wasn't the model. One of them had opened with the brand's voice rules, the three reference campaigns the brand director keeps pointing to, and the list of things that brand never does. The other had opened with the product name and a mood.
That's the whole post, really. But it's worth unpacking, because the model question is about to get louder in every creative team I work with, and it's the wrong question to spend a meeting on.
The model argument is now a taste argument
Every's Context Window team wrote this week about a new round of frontier models splitting their own staff. Some people there prefer the ambitious, slower model. Others prefer the faster, cheaper one that gets to the point. Their framing is the one I'd hand to any creative leader: models are now capable enough that "speed, price, and taste decide which one you use."
Read that as a creative director and it should sound familiar. That's how you pick a photographer. Nobody on your team thinks there's one correct photographer for every job. You pick based on the brief, the budget, the timeline, and whose eye fits the work.
So when a team spends three weeks trying to crown a house model, what they're really doing is trying to make a taste decision once, for everyone, forever. It won't hold. The copywriter will like one thing, the motion designer will like another, and in six weeks a new release will reshuffle the whole ranking anyway. I've watched that argument eat entire quarters.
What actually makes the work consistent
At Maesa the stakes on consistency were real. They run 12+ brands, each with its own voice, its own customer, its own shelf. If output quality depended on which designer happened to be driving which model that day, the whole program would have fallen apart by the second launch.
It didn't, and the reason was boring. We built the brief before we built anything else.
Not a prompt template. A brief. For each brand, one document the whole team works from, whichever model they open:
- Voice rules written as examples, not adjectives. "Warm and confident" means nothing to a model. Five approved headlines and five rejected ones mean a lot.
- References with reasons. Three to five past campaigns the brand director loved, each with a sentence on why. The why is the part people skip, and it's the part that carries.
- The never list. Colors, claims, words, compositions that the brand has killed before. This one catches more bad output than anything else in the document.
- What done looks like. The criteria the approver actually uses, written down, so the person generating and the person approving are grading the same test.
Once that existed, the model argument mostly went away on its own. People kept their preferences. The fast-model designer stayed on the fast model for volume work. The slow-model designer kept it for hero frames. Both sets started looking like Maesa, because both were starting from Maesa.
That program got launches down to 3 months instead of 9, at roughly $280K saved per launch. I'd love to tell you it was clever prompting. It was mostly the brief.
The work happens before the prompt
The same Every piece makes a point about writing that I think is even more important for creative teams: the quality of what comes out depends on the preparation you do before you ever type the request. Most teams I meet treat the prompt as the creative act. It isn't. The prompt is the handoff. The creative act is deciding what you want, what you don't want, and how you'll know when you've got it.
Creative people already know this. It's what a good brief has always been for. You would never send a photographer to set with "product shot, make it nice." But I watch senior designers do exactly that to a model every day, then blame the tool when it hands back the median of the internet.
How I'd run this on Monday
If your team is in the middle of the model debate right now, here's what I'd do instead of picking a winner.
Call a truce on the model. Let people use what they prefer for their own work, within whatever your security and licensing rules allow. You can revisit it later with real data from your own jobs.
Pick one brand or one client and write the brief together. Two hours, the people who actually approve work in the room. Examples over adjectives. The never list. What done looks like.
Run the same job across two or three models with that brief. Put the results side by side. In my experience the gap between models shrinks a lot once the input is solid, and the gap that remains is genuinely a taste call. That's fine. Taste calls are what your team is paid for.
Make the brief the thing you maintain. Models change every quarter. Your brand's voice, references, and never list change slowly. Put your effort into the asset that lasts.
This is also why I don't think AI works as a two-person initiative. If only your two power users know to front-load the brief, you get two people making on-brand work and everybody else making competent stranger work. The brief has to be how the whole team starts, which is exactly the kind of habit the half-day Audit Workshop surfaces and the 8-week Operating Model installs.
The next model release is probably a few weeks out. Your team will argue about it. Let them. Just make sure that whichever one they pick, it's reading the same brief.