The Audio Freeze

Google isn't scared of the audio race. Google owns YouTube, which is to say Google owns the single largest repository of human speech, music, ambient noise, lecture, argument, and crying-on-camera content that has ever existed. The idea that they're scrambling to catch up on audio is like worrying that Exxon is going to run out of pipes.

Everyone else, though — gone. Silent. The audio release landed and the rest of the model companies did what companies do when they don't have an answer yet: they stopped talking. Runway is the one exception worth watching, not because of their audio work but because their world simulation stuff has this emergent quality to it that feels like it's doing something nobody planned. That's the interesting tell. Everything else right now is deliberate, manufactured, benchmarked. The Runway stuff looks like it surprised the people who built it.

The more interesting read came out of the OpenAI court filings — which, as a source of strategic intelligence, have been doing more work than any of their blog posts. What they're actually building toward, buried in the legal documents where nobody expects to find product roadmaps, is T-shaped agents. Broad enough to cover everything, deep enough in specific domains to actually be useful — the generalist-with-specialization structure that every management consultant has been drawing on whiteboards since 1990, now apparently the architecture of the thing that replaces knowledge work.

Nobody announced this. It showed up in a lawsuit.

That's where we are. The real roadmap is in the discovery documents.