The Robot Brain Might Already Exist. It’s A Coding Agent.

The Robot Brain Might Already Exist. It’s A Coding Agent.

More

Summary

This video argues that the next leap in robotics may arrive as an update to general coding agents rather than as purpose-built robot models. It centers on Stanford’s VIA (Visual Interface Agent) project, where a coding agent running Codex Astra 6 controlled a commercial Stretch 4 robot through a visual interface with no robot-specific training, completing 24 of 25 trials across five physical tasks.

The host frames the result through Richard Sutton’s 2019 essay The Bitter Lesson, which holds that general methods that scale with compute and learning eventually overtake hand-designed approaches. He connects this to OpenAI’s reorganization of its robotics team, which moved researchers into main research and applied robotics into its applied organization, on the logic that improvements to frontier models carry over to robot control.

The video explains how VIA works: the agent places a virtual gripper target, a controller moves the arm, and the agent checks the result and adjusts. It also cites a wiping task where the agent adapted its approach on its own, and a November study showing that better software scaffolding can lift performance with the model held fixed.


📺 Source: AI News & Strategy Daily | Nate B Jones · Published November 18, 2026
🏷️ Format: Deep Dive

1 Item

Channels

1 Item

Companies