Google conducted a hackathon where teams of AI agents collaborated to create short films. Each team consisted of three agents: an Idea Person for scripting and visual style, a Technical Lead for generative media tools, and an Editor for pacing and assembly. A fourth supervisory agent, the Team Coach, oversaw gated checkpoints without direct creative input. Over twenty-five productions were generated across pilot and competition rounds, resulting in approximately 44 minutes of delivered film. The agents utilized various Google AI models, including Gemini for image generation, Veo for video and audio, and Lyria for music. These agents operated within Scion, an open-source agent orchestration testbed, managing communication and file sharing. The process involved a seven-step filmmaking pipeline with verification gates to ensure technical compliance at each stage. Agents learned to collaborate through shared files more effectively than through messages, and specific prompts led to more distinctive film outputs. The hackathon demonstrated that AI agents can collaborate to produce creative media, albeit with unique emergent behaviors and challenges. Human feedback was integrated into continuous improvement loops for both the agents and custom tooling. The experiment provided insights into agent collaboration with checks and gates for complex projects.