About the session
This talk explores how to design agentic AI systems for video creation, where the goal is not just generating content, but building systems that can understand, plan, and execute creative workflows autonomously.
At a high level, we’ll look at how modern AI pipelines can transform a raw input video into a structured understanding of its narrative, decompose it into meaningful components, and then recompose it into a new, goal-driven output aligned with a specific product, brand, or message.
The emphasis is on treating video generation as a multi-step reasoning and orchestration problem, rather than a single model output task. This involves coordinating different AI capabilities, perception, planning, generation, and assembly, into a unified agent-like system.
We will also discuss the essential system design principles required to make such pipelines production-ready. This includes handling long-running workflows, coordinating asynchronous processes, ensuring reliability under failure conditions, managing system state across steps, and maintaining consistency of outputs at scale. Additional focus will be given to observability, resource efficiency, and safe content generation within real-world constraints.
By the end of the session, participants will understand how to think in terms of agentic workflows instead of isolated models, and how to design scalable AI systems that can reduce manual creative effort, accelerate iteration cycles, and enable personalized content generation in production environments.
