How Puma used AI agents to create an ad from scratch

An experimental spot for Puma, produced with virtually no human touch, shows how AI agents are helping marketers automate a growing portion of the creative development process. Puma worked with agency Monks to train a team of AI agents to create an ad from scratch, from designing preliminary concepts to writing scripts to generating high-quality shots and, ultimately, finalizing a commercial. AI agents are autonomous bots that execute specific functions and can be programmed to work together to complete tasks that are far more complex. Puma and Monks engineered hundreds of agents to create a one-minute ad in five weeks. The spot, which Monks released yesterday, is primarily an experiment. But the effort potentially offers a much different view of how AI could affect creative marketing: Rather than an individual prompting a large language model (LLM) to produce ads, a group of agents will do the prompting—and perhaps the rest of the creative duties—for them, said Henry Cowling, chief innovation officer at Monks. “We’re moving into a world with fewer agencies and more agents,” Cowling said. The effort comes as brands are entrusting AI with more responsibility in the development of their ads. Coca-Cola ran AI-generated TV spots during the holidays at the end of last year. Vodafaone also recently produced a fully AI-powered ad. The rise of video tools, which creative agencies are testing for their own experiments, has precipitated this deeper trust. Also read: Secret Level is at the forefront of AI-generated advertising Where Puma and Monks’ spot differs from many previous efforts is in its heavy reliance on AI agents, which many in the ad industry view as a force that could accelerate automation in marketing. Adobe this week rolled out a suite of agentic AI tools for brands, following similar launches from OpenAI and Meta. Non-agentic AI leverages generative capabilities, but typically requires a human to prompt it, improve it and move it along. Agents, on the other hand, can be programmed to automate ideation, iteration and transitioning content between phases in a process, Cowling said. Also read: Watch Puma’s biggest ad effort to date For example, in Puma’s ad, numerous agents compiled potential concepts based on a brief, while others received those concepts and curated the best ones. The differences in this material depended on the differences in agents’ training, Cowling said. For example, some bots were programmed to conceptualize or curate specifically for Gen Z audiences, while others were programmed to do so for older ages. After curation, the winning concepts were handed over to still other agents that developed them into scripts before sending them into the next stage of the ad creation process—a brigade that culminated in a final, one-minute spot. Human oversight was still involved at the stages in which material was curated, Cowling said. The ad features numerous vignettes of realistic athletes competing in their sports, as well as a final shot of a Puma shoe. This last detail is notable given the trouble that marketers have had when using AI to depict products and logos. Still, the spot is not perfect. At least one of the characters is shown with an unusual number of fingers—a mistake that is commonly made by generative AI bots. And previous attempts by brands to go all in on AI-generated ads have almost always incited backlash, especially from members of the production industry who fear that the technology is being used to replace them. “I do think what we’ve done here will be controversial,” Cowling admitted. For Monks, the primary reason for the experiment was to develop a pipeline of AI agents, trained specifically for Puma’s marketing, that can be used in the future for producing more content for the brand, Cowling said. Monks is currently building similar pipelines for clients including Google, General Motors and BMW. Monks is also seeking to promote an expansion of its partnership with Nvidia, which it also announced yesterday during the Nvidia GTC conference. The agency launched an “Agentic AI Advisory Group” that will use Nvidia technology to create agent pipelines for brands. It also formed the “Monks Foundry,” which is essentially a team of 150 engineers that will power the advisory group. Roughly 60% of these engineers are current employees; Monks will hire the remaining 40% over the next six to 12 months, Cowling said. Using Nvidia’s AI hubs to create the spot, Monks leveraged the chip giant’s Cosmo video model, as well as one of Runway’s models. The effort also used Flux for image generation and Meta’s Llama model to help power the agents. ~ ~ ~CLARIFICATION: A previous version of this story indicated that Puma would test the ad in public, as stated by Monks. Monks has since clarified that the ad will not be tested in public.

How Puma used AI agents to create an ad from scratch

Featured Stories