A–04 AI ENTERTAINMENT
AI Entertainment and Video Systems
New formats for story and video, shaped by generative intelligence.
Generative models changed the cost structure of video and story production, but tools are not formats. We are building the systems layer for AI-native entertainment: pipelines where models do the heavy lifting and human editorial judgment decides what ships.
Current state of the AI entertainment work
Format experiments in short drama and generated video are underway and not yet public. As with our interactive work, we announce when something is real, not before.
The bet underneath this work
Generation stopped being the bottleneck. Anyone can produce hours of passable video now, which means volume has no value and the scarce input is judgment about what deserves to exist. We are building toward pipelines where models absorb the production cost and a human decides, at a small number of clearly defined gates, what actually ships.
Format is the second half of the bet. A story told well in a medium nobody watches reaches nobody, and the platforms that carry short video reward structural choices — pacing, opening seconds, episode length — far more than they reward production polish. Those constraints belong in the system that generates the work, not in a review at the end.
The failure mode we design against is fluent emptiness: output that is clean, on-brief, and says nothing. It is the characteristic failure of generative pipelines, and it is invisible to any automated quality check, because nothing about it looks broken. The only reliable filter we have found is a person with taste and the authority to reject.