Back from three days at TechCrunch Disrupt: how the panels answered my three questions on robots, networks and timing, and the one nobody answered.

Last week I attended TechCrunch Disrupt in San Francisco, three days of sessions on Physical AI, AI infrastructure, and agents. I have written about Physical AI since May, and I went to test that thinking against what founders and investors were saying on stage.

Before I left, I wrote that I would ask every robotics founder three questions. I kept them in mind through every session.

The first was what a robot does when the network drops. The Physical AI panel gave the clearest answer I heard all week. A founder building a foundation model for sensor data said his customers run air-gapped, and a 300ms cloud round trip is too slow for them. His fix is to train a large model, then compress it into a small agent that lives on the device. The Slow Brain is moving onto the machine.

The second was which part of the stack a founder would rather not own. The tokens panel answered with the opposite instinct: own your intelligence and your harness, so no single provider sets your costs or your roadmap.

The third was who guarantees the timing, and to what standard. That is the one nobody answered. The session titled "Physical AI Isn't a Model Problem. It's an Execution Problem" spent its time on deployment, data access, and compression. The first two days of the agenda have no session on timing at all.

The same panel added a reality check. There will be no ChatGPT moment for robots. ChatGPT reached 100 million users in two months, the iPhone took about five years, and a panelist put motors at 60 to 70 percent of a humanoid's cost.

So the pattern from Automate in June and from Intel's session at MassRobotics held up. The Slow Brain gets better and cheaper to run every month. The layer between model and motor that guarantees timing still has no owner. Disrupt did not change that.

If you build robots, chips, or the software in between, I would like your answer to the third question. Who owns the timing in your stack? Thoughts?

This perspective was originally published on LinkedIn. Join the discussion, add your thoughts, and follow for regular updates.

View on LinkedIn