The production problem is already solved. That's not the interesting part.
Let's get the obvious thing out of the way. Making stuff is cheap now. A mid-size brand can generate 400 ad variants before lunch, localize them into 30 languages, and swap the product shot for a regional SKU without a photographer. Two years ago that was a pitch deck. Today it's a Tuesday.
What nobody in those planning meetings wants to say out loud is that this mostly benefits the people who were already good. If your positioning was muddy, you now have 400 muddy ads instead of four. Volume amplifies whatever you fed it.
I keep thinking about the Coca-Cola holiday ad from late 2024, the AI-generated one that redid the "Holidays Are Coming" trucks. The backlash wasn't really about the visuals, which were fine if a bit uncanny. It was that people felt the brand had shortcut something they cared about. A brand with 130 years of goodwill spent a little of it to save a production budget. That trade will be offered to every marketer, every quarter, for the next five years. Some will take it too often.
What I think actually changes
1. The brief becomes the product
If generation is free, the thing that's expensive is knowing what to generate. The brief, the strategy doc, the tone guide, the list of things the brand would never say: these stop being PDFs nobody reads and become the literal input to the machine that makes everything.
I'd bet that by 2029 most serious brands have something like a living brand model. Not a style guide. A structured, machine-readable definition of voice, visual rules, claims that are allowed, claims that are banned, the three customers we care about and the two we don't. Every piece of output gets checked against it before a human sees it.
This is not glamorous work. It's the marketing equivalent of writing good tests. The teams that do it will ship faster and embarrass themselves less. The teams that skip it will keep getting surprised by what their own tools produce.
2. Personalization finally means something, and it'll be creepy for a while
We've been promised one-to-one marketing since the late nineties. What we got was "Hi {FIRST_NAME}." The difference now is that the message itself can change, not just the salutation.
By 2028 I expect the normal case to be: same campaign idea, thousands of executions, each one shaped by what the model infers about you. Not just which product, but which argument. The person who cares about durability gets a different ad than the person who cares about how it looks in a photo.
I have mixed feelings about this. It works. It also has a way of feeling like the brand is reading your diary. The brands that win here won't be the ones with the most data. They'll be the ones that learn where the line is between "that's useful" and "how did you know that." Regulation in the EU is going to force some of that restraint whether brands like it or not, and honestly, I think that's healthy.
3. Your customers' AI is the new gatekeeper
This is the one I think most people are underestimating.
More and more purchase decisions now start with someone asking an assistant. "What's a good running shoe for flat feet under $150." The assistant answers. Maybe it lists three brands. Maybe it lists one. The customer never sees your homepage or the headline you spent three weeks A/B testing.
So who is your brand talking to? Increasingly, a model. And models don't respond to vibes. They respond to what's written about you, what's structured on your site, whether your product specs are clear, whether reviewers say the same thing you say.
I don't think traditional SEO dies, but it gets a sibling that matters more. Call it whatever you want. The practical question for a brand in 2027 is: when a model summarizes us, is the summary accurate and flattering, or is it a shrug? Most brands have no idea what the answer is today.
4. Agencies split in two
I've watched a few agency friends go through this and it's not fun. The middle of the agency business, the part that turned a strategy into 40 deliverables, is where the cost has collapsed. That was a lot of headcount.
What's left splits. On one end, small teams selling judgment. The idea a model wouldn't have come up with, because it's trained on what everyone already did. On the other end, technical shops that build and maintain the brand model, the generation pipelines, the review tooling. The people who used to sit in the middle, translating one into the other, are the ones I worry about.
Big holding companies will say they're doing both. Some are. Most are cutting the middle and hoping nobody notices the seams.
5. Human-made becomes a label, like organic
Around 2023 this sounded like wishful thinking from copywriters. I no longer think so.
When everything can be generated, provenance starts to carry a premium. Not for everything. Nobody cares if the banner ad for a discount mattress was written by a person. But in categories where the brand is the point (fashion is the obvious one, spirits is another), "a person made this" will be a claim brands make explicitly, and sometimes charge for.
The wrinkle is that it'll be hard to prove, and some brands will lie about it. I expect at least one decent-sized scandal along those lines before 2030. A brand that built its identity on craft gets caught with a generation log.
Where I'm unsure
I don't know whether the flood of generated content makes people care more about brands or less. There's a version of the future where everything looks so similar that the few distinctive voices stand out more than ever. There's another where people just tune the whole category out and buy on price. I go back and forth on this weekly.
I also don't know how much of the "AI marketing" spend right now is real. A lot of budgets got reallocated in 2025 on the basis of demos. Some of those demos are becoming products. Some are quietly getting shelved. The 2027 numbers will be more honest than the 2026 ones.
If you're planning a budget right now
Spend less on making things and more on deciding what to make. Write the brand model. Get someone technical to help you make it machine-readable, then be boring and disciplined about enforcing it.
Find out what the major assistants say when someone asks about your category. Ask them yourself. If you're not mentioned, that's your actual competitive problem, and no amount of generated video fixes it.
Keep some things human on purpose, and be honest about which ones. The audience can tell more often than you think, and they're getting better at it.
And when someone puts up the "AI will transform marketing" slide, ask them what changes on Tuesday. If they can't answer, the slide is decoration.