Physical AI looks closer than ever.

But the hardest part in robotics is not getting a machine to do one impressive task on camera. It is building systems that can improve from real-world experience, handle edge cases, and scale across different robots and environments.

In this episode of Automated, Brian Heater speaks with Sergey Levine of Physical Intelligence about why robotics has reached an inflection point, and why progress now requires more than great models in a lab.

Sergey explains why the next phase of robotics will depend on something much less flashy than a viral demo: collecting the right real-world data, learning from it efficiently, and building systems that improve through deployment.

The conversation explores what makes a robot experience useful in the first place. Sergey describes a concept borrowed from child psychology called the “zone of proximal development,” where the best learning happens when a system is challenged just beyond what it can already do. For robots, that means creating environments where they can succeed, fail, adapt, and improve.

Brian and Sergey also discuss how the bottleneck in robotics is changing. Basic motor skills are improving fast. The harder problem now is judgment. A robot may be able to clean dishes, but if it drops a clean plate on the floor, it still has to understand that the plate needs to be washed again. That kind of common sense remains one of the biggest unsolved challenges in physical AI.

They also dig into one of the biggest debates in robotics right now: data. Sergey argues that real-world data collection is not the impossible obstacle many researchers once assumed. In fact, he believes the long-term path to better robots is more practical than people think. Deploy systems, collect experience, improve the model, and repeat.

The conversation also covers why Physical Intelligence is focused on a general intelligence layer rather than a single-narrow product, why robots should not just be treated as metal versions of people, and what surprised Sergey most about controlling very different robot platforms with the same model.

Finally, Sergey reflects on why Physical Intelligence is structured more like a lab than a traditional startup, why experimentation matters so much in modern AI, and how we may one day look back on this era as the moment AI moved beyond internet data and into the physical world.

Connect with Sergey Levine

https://www.linkedin.com/in/sergey-levine-5a31a24

Learn more about Physical Intelligence

https://www.physicalintelligence.company/

We’d love to hear from you.

Have thoughts or guest suggestions?

Reach us at podcast@automate.org.

You can find the transcript and more episodes of Automated at automated.fm

Unlock full access to Automated and explore everything automation.

Subscribe today and leave a review on YouTube, Apple Podcasts, and Spotify.

Subscribe to the Automated Newsletter:

https://www.automate.org/automation/automated-newsletter

You can also find us on:

LinkedIn https://www.linkedin.com/showcase/automated-podcast-by-a3/

Instagram https://www.instagram.com/automatedpod/

Hosted on Acast. See acast.com/privacy for more information.

Sergey Levine on Why Real-World Data Will Define Physical AI

Automated with Brian Heater

Sergey Levine on Why Real-World Data Will Define Physical AI