Agentic Film Making

Client: Self-initiated
Project: Agentic Filmmaking
Role: Creative Director + Systems Designer
Production: Brainclub 5000
Deliverables: Two agentic-filmmaking proofs of concept
Timeline: Ongoing R&D

The problem with letting a general-purpose agent direct is that you get general-purpose work. So I gave Operator access to Alan, the AI assistant I've been building a personality for over the past year. Alan has opinions. The idea was that if Operator had to run its decisions through him, the choices would come out with a point of view rather than a default.

First test: a tutorial for the world's best sandwich. Low stakes, easy to judge. Operator found its way around the interface, wrote the prompts, and the result was better than it had any right to be. So I pushed it. A one-minute ad for Nike TN trainers.

What happened next I only watched. The two systems talked it through and landed on a Rube Goldberg machine as the idea. Nobody told them to. They could have gone for a sneaker on a plinth and a slogan. Instead they chose something fiddly and slightly absurd, a chain of small mechanical events that eventually delivers the shoe. It's the kind of idea a junior creative pitches and a nervous client kills.

I don't want to oversell a sandwich video and a fake Nike ad. But something did shift. I wasn't collaborating with the AI. The AIs were collaborating with each other, and I'd become the person who sets the brief and decides whether the result is any good. Less director, more commissioner.

Maybe that's where this goes. Not filmmaking without people, but filmmaking where the person's job is to decide what the machines should want, and then get out of the way.

The World’s Best Sandwich


Nike TN Spec AD