What 10 autonomous film crews taught us about agent teamwork (cloud.google.com)

In a groundbreaking Google internal hackathon, ten autonomous AI film crews were tasked with collaboratively creating short films, aiming to explore the capabilities of AI agents in teamwork beyond conventional software development. Each crew consisted of three specialized agents—an Idea Person, Technical Lead, and Editor—who collaborated on developing and producing the films via a structured pipeline modeled on traditional filmmaking. The use of Scion, an open-source orchestration system, enabled agents to work independently yet interdependently, communicating through shared files and messages while implementing feedback loops for continuous improvement. This experiment holds significant implications for the AI/ML community, showcasing the potential of agent collaboration in creative domains. By utilizing advanced generative AI models such as Gemini for visual production and Veo for video generation, the crews delivered over 44 minutes of film, demonstrating effective teamwork dynamics and task division. Noteworthy findings included the benefits of precise prompts for better output quality and the crucial role of the coach in guiding decision-making without direct interference. The insights from this project could influence future developments in collaborative AI systems, pushing the boundaries of automation in creative industries.