Agentic creative is not a slide

Every agency pitch deck has an AI slide now. Most of them are the same slide: a logo soup of tools, an arrow, and a promise that creative will get “10x faster.” Almost none of them have shipped a single asset through that pipeline into a live campaign.

The difference between a slide and a system is whether the pipeline survives contact with a real brief, a real brand reviewer, and a real launch deadline.

What a real pipeline has to handle

A creative pipeline that actually runs in production has to answer some unglamorous questions:

• Who owns the output? Every asset that ships needs a named human who approved it — not a model, not a vague “the system.”
• Where does brand voice actually live? Not in a one-time prompt someone wrote in January. It has to be a maintained reference the agents pull from on every generation.
• What happens when a model gets swapped? If your entire creative program breaks because a vendor deprecated an API, you didn’t build a system — you built a dependency.
• How does the loop close? Performance signals from the live campaign need to route back into the next brief automatically, or you’re just doing slow manual creative with extra steps.

The shape we run

Our internal pipeline moves through six stages — brief, research, concept, direct, produce, ship — and we’re explicit about which stages are agentic and which stay human. Strategists write the brief. Research and concept agents do the first-pass synthesis and ideation. Creative directors edit the shortlist. Production agents render every size and format. Media leads launch and the results feed the next brief.

The humans sit at the two ends and the middle checkpoint — framing the problem and making the judgment calls. The agents handle the parts that are genuinely mechanical: synthesis, generation at scale, and resizing.

That’s the whole trick. Not replacing judgment — removing the bottleneck between judgment and volume.