First published on Travis's LinkedIn, where he shares his take on agentic advertising, ad tech, and the business of building in this space. A marketer turned entrepreneur, Travis is co-founder and Executive Director of Adzymic, with over two decades across adtech and digital media in Asia Pacific — including a stint leading SAP's regional digital media practice. Follow him on LinkedIn for more.
There are a lot of “AI creative platforms” now.
Some generate images. Some generate copy. Some can turn a product feed into hundreds of variations. Increasingly, you can type a prompt and get something that looks remarkably close to a finished ad.
The technology has moved very fast. But after spending time building around this space, I think we may be using the term AI creative platform too loosely.
Generating an asset is only one part of making advertising.
Anyone who has worked with a large advertiser knows what happens after the first creative comes back. The logo is slightly wrong. The font isn't approved. The product needs to move 20 pixels to the left. Legal wants another line added. The agency wants a 300x250, 300x600 and 970x250. Then somebody asks for a mobile version, a video version and maybe a DOOH execution.
And by the way, the promotion changes next Tuesday.
This is the less exciting side of advertising that most AI demos don't show.
The model isn't really the product
Image generation is getting very good. So is copy. Video will get there.
My guess is that, fairly soon, access to a good generative model will be the easiest part of building an AI creative product. Everyone will have one, either directly or through an API.
The harder part is everything around the model.
Take something as basic as brand guidelines. A brand might have very specific rules around how its logo is placed, what colours can sit behind it, which fonts can be used, how much whitespace is required, what tone its copy should take and which product images are approved.
You can dump some of this into a prompt, but that doesn't really solve the problem.
If a brand manager still needs to correct the AI ten times before approving an output, you haven't removed that much work. You've just moved the work around.
That is why I think brand understanding will become an important part of these platforms. Not just uploading a logo and selecting three colours, but retaining enough knowledge about the brand that the first output is already reasonably close to something usable.
A good ad also needs an idea
There is another part that gets overlooked when we talk about generation.
Someone still has to decide what to make.
Suppose the brief says: An airline brand is running a Double 11 promotion across multiple destinations. What should the ad actually do?
It could be a simple promotional banner. It could open with the sale message before moving into a destination gallery. Prices could be dynamically populated. The destinations could change based on where the person is travelling from. Maybe there is a countdown as the promotion gets closer to ending.
A designer or creative team makes dozens of small decisions like these without necessarily describing them as “reasoning”.
For AI creative to become genuinely useful, I think it has to start participating in this part of the process too. Give it the brief, brand and campaign objective, and it should be able to suggest how the creative could work before generating all the assets.
We have been experimenting with this ourselves, and it is much harder than generating a nice-looking visual. There are many possible answers, and the most visually impressive answer isn't always the best advertising idea.
Then you hit production reality
This is where I think many AI creative demos stop too early.
They show you a beautiful 1:1 visual. Advertising rarely ends there.
For one campaign, you may have standard display banners, high-impact units, video, social, CTV and DOOH. Even within display, a desktop skin behaves very differently from a 300x250 banner. You can't just resize everything and expect it to work.
Rich media makes this even more obvious. A carousel has behaviour. A gallery has states. A countdown needs a data source. A live-score creative needs rules for what happens before, during and after the match.
So templates aren't going away just because generative AI has arrived.
In fact, I suspect they become more useful.
AI can generate the idea, content and variations, while a design system gives those outputs some structure. You get flexibility without having to trust a model to reinvent the technical implementation of an ad every single time.
That balance between generation and structure is something the industry is still working out.
The part I'm most interested in is what happens after generation
Creative automation becomes much more powerful once it connects to data.
A retailer shouldn't have to manually produce a banner for every product. A travel advertiser shouldn't need separate creative production every time it wants to promote another destination. A sports sponsor shouldn't need somebody sitting around updating a score.
We have been doing this through DCO for years, using things like product feeds, location, weather, time and live sports data.
AI changes what can happen on the creative side.
Historically, you might build a template and define all the possible combinations before the campaign begins. The system then assembles those combinations according to a set of rules.
With generative AI, some of those creative decisions could eventually happen much later in the process. The system might decide what message, layout or imagery makes sense for a particular context and generate what it needs.
We're not fully there yet. There are plenty of questions around brand control, consistency, latency and whether marketers even want AI making that much of the decision.
But this is far more interesting to me than generating another hundred banner variations.
And someone still has to run the ad
There is a very practical test I use when looking at AI creative products.
What happens after I press generate?
If the answer is that somebody downloads a JPEG, sends it to the agency, waits for approval, asks a designer to make the remaining sizes, uploads everything into an ad server and checks that the tracking works, then we have improved one step of a much longer process.
That's still valuable. But there is a lot more left to solve.
The direction I find more interesting is where the brief, brand knowledge, creative generation, formats, dynamic data, QA and activation start coming together.
Eventually performance should feed back into the process as well. Not simply telling you that Creative A had a higher CTR than Creative B, but helping the system make a better creative decision the next time.
The image model gets most of the attention because that's the part you can demo in 30 seconds. But the companies that eventually build meaningful AI creative platforms may win by solving all the boring things around it: understanding brands, producing real ad formats, handling revisions, connecting data, checking outputs and getting the creative live.
Generating the picture may turn out to have been the easy part.
If that resonates, take a look at the AgenX Creative Agent — Adzymic's creative agent built for exactly this gap. One brief in, deployable rich media out: every size, every language, every channel, brand-consistent by design and ready to serve programmatically or agentically. [Explore the Creative Agent →]