The majority of creative agencies have basically thrown their team an AI model and called it a strategy.
A prompt from the creative director generates an output that is adjusted and prompted again. It’s good, often great, but the process is artisanal: slow, unscalable, and completely dependent on who’s doing the prompting that morning—a craft masquerading as infrastructure.
But the shift has started a push beyond learning better prompts and towards a pipeline approach.
The Analogy: Chef Logic vs. Restaurant Logic
If you think about the difference between a great chef and a great restaurant, the chef has incredible taste, incredible depth of knowledge, incredible instinct built up over years and years of experience. None of those qualities is scalable. The restaurant, on the other hand, has a process, a standardisation, a menu, a series of positions and jobs that produce high-quality results no matter who walks through the door. Most creative teams working with AI today are working with Chef Logic on a restaurant problem. They’re working with the prompt as the dish. And if every dish requires the same chef’s intuition every time, then you don’t have a kitchen, you don’t have a restaurant, you have a show, you have a performance. And the moment that the chef walks out the door, the entire thing shuts down.
Transitioning to Production Mode
Building an actual AI pipeline means moving beyond performance and entering production mode. This means moving beyond “what do we want to make?” and instead thinking “how do we make it repeatedly, consistently, and at scale?” so we eliminate the over-dependency on a single person’s intuitive prompting.
The Three Elements of a Functional Pipeline
There are three key elements of a functional creative pipeline.
First, a brief layer, which is the structured input that defines intent, constraints, audience, and tone before any model ever sees any kind of request.
Second, a model stack, which is the intentional selection and ordering of tools based not on personal preference, but on task type. For instance, Claude for reasoning and copy scaffolding, MidJourney or Luma for visual ideation, Runway or PICO for motion, etc.
Third, a human-in-the-loop checkpoint with a defined moment at which judgment, taste, and strategic oversight reenter the process. This last part is the one most teams don’t do, and it’s the most important one.
Building the Infrastructure
To build the pipeline, we start with an AI brief; this is different from a creative brief. Think of it as a pre-prompt layer: a set of variables before they’re a prompt, such as audience context, brand voice attributes, format requirements, deviation range, and goals. Think of it as the standardized workflow. Without it, everything starts from scratch. With it, everything starts from the same “kitchen”.
Then there’s the model stack. This is where most teams get distracted by “shiny objects”. New tools come out weekly promising to revolutionise everything. The art of the game here is to avoid the temptation of chasing each new release and instead ask, “What role does this model play in our stack?” and “Does this model perform better than what we currently have in that role?” A model stack is an architecture of decisions, not a collection of shiny new objects.
Finally, there’s the human in the loop, where strategy, taste, and domain expertise truly can’t be replicated. Rather than putting creative energy into execution, the team puts creative energy into curation, direction, and evaluation.
The Future: The System as the Asset
There’s a lot of anxiety in our industry right now about AI and creativity. There’s a fear that everything’s going to look the same because we’re all working with the same models. And that’s true. But what people are forgetting is that the models are going to be a commodity in one to two years.
What’s going to drive differentiation down the line is the pipeline we built around those models, the institutional knowledge of how to brief, how to sequence, and when to override that system.
“The most valuable creative asset of the next decade won’t be a single campaign or an AI model. It’ll be the system that a team develops to produce creative assets that look nothing like anyone else’s.”
And the chef’s essential role in all of this will be to build the restaurant.
About the Author
Hiba Hassan is an AI Adoption Strategist, Luma AI MENA Lead, and Fractional CMO of Emilio Schuberth. A neurodivergent leader, she translates emerging tech and inclusive values into high-impact strategies.
Disclaimer: This article was originally written by Hiba Hassan. It is shared here for educational purposes. All rights belong to the original author.
