For two years running, Coca-Cola spent the holidays teaching the rest of us an expensive lesson about AI and branding. Almost nobody took notes.

In 2024, Coke remade its 1995 "Holidays Are Coming" ad with generative AI. The internet recoiled at the uncanny faces and the trucks that would not hold their shape. In 2025 the company came back and did it again, this time leaning on animals to dodge the uncanny humans. Viewers caught the shape-shifting trucks anyway, and the warmth still was not there. Coke's own head of generative AI defended both as human-guided storytelling. The audience kept saying the same thing: this feels cold.

Here is what makes the Coke case useful instead of just another AI pile-on. The ad-effectiveness scores were strong. Testing firms rated the spot high on attention and recall, the metrics that usually predict an ad will work. By the numbers a CMO trusts, it performed. By the only number that matters for a ninety-year-old brand built on warmth, it leaked equity. Both things are true at once, and the gap between them is the whole problem with how most companies are thinking about AI in branding.

AI is good at generating the asset. Holding the brand is a different job, and it is bad at that one. Generating an asset is production: make a thing that looks like the reference. Holding a brand is judgment: know what this brand is allowed to feel like, and refuse the version that scores well and means nothing. A model cannot do the second job, because the second job is taste, and taste is what the brand is selling.

This is the cross-section I work in. I vibe-code, I build products with AI in the loop, and I direct brands. The view from that intersection is not the one you get from either camp alone. The "AI replaces creatives" crowd and the "AI is slop" crowd are wrong in the same way. Both think the question is whether the machine can make the asset. It can. The real question is who decides whether the asset should ship, and that decision got more valuable the moment generation became free.

Coke is not the only example. Meta's ad platform quietly swapped a brand's top-performing human ad for an AI-generated granny the marketer never asked for. Vogue ran an AI model and readers threatened to cancel. The generation is never the failure. It is fine, often impressive. What fails is that nobody with brand judgment stood between the output and the audience, or the person who did got overruled by a cost-efficiency argument.

So the actual AI x branding skill in 2026 is editorial judgment at machine scale, not prompting. Prompting is the easy part, and it gets easier every month. The skill is the ability to look at fifty on-brand-ish options and kill forty-nine, fast, for reasons you can defend. That is creative direction. AI did not replace it. AI made it the bottleneck, because everything upstream of it just got automated and everything downstream depends on it.

The brands that win the next few years will run AI hard on production and guard the judgment layer like the asset it is. The ones that lose will do what Coke did: let the efficiency of generation make the call that only taste should make, then act surprised when the brand feels a little less like itself every quarter.

The machine can make the ad. Whether it should is still your call. Stop outsourcing that one.