Anthropic's AI agent programmed a robot dog 20 times faster than humans: a breakthrough in physical AI
Anthropic has unveiled the second phase of the Project Fetch experiment, and the results are impressive. The Claude Opus 4.7 model demonstrated the ability to autonomously configure and program a four-legged robot, completing tasks 20 times faster than teams of human engineers working with the previous version of the AI.
As a reminder, in the first phase of the project (August 2024), the artificial intelligence only assisted humans, helping them find solutions faster. Now, we are witnessing a qualitatively different level of autonomy. The Opus 4.7 model, under minimal researcher supervision, independently completed the full cycle of work:
- connected to the sensor system, including video and lidar;
- wrote code for manual robot control;
- developed a trajectory monitoring system;
- configured object recognition algorithms.
The key performance indicator is speed. Opus 4.7 turned out to be 18 times faster than a team using older versions of AI assistants, and 37 times faster than engineers working without a chatbot. At the same time, the neural network generated code that is 10 times more compact than human-written code — fewer lines, higher efficiency.
It is important to emphasize: Anthropic did not introduce specialized algorithms for robot control. According to the developers, the progress in robotics became a side effect of the general scaling of language models. This is direct evidence that fundamental improvements in AI systems can open up new, unforeseen horizons.
However, there were limitations. Claude still struggles with precise physical manipulations in real time. The model managed to guide the robot to the target but failed to gently push a ball. This point clearly demonstrates that even the most advanced neural networks still lag behind humans in complex sensorimotor coordination.
My analysis: We are on the threshold of the era of "physical AI agents," as rightly noted at Anthropic. The ability to work with real equipment through standard interfaces is not just hype. It is a transition from a digital assistant to a full-fledged autonomous operator. If the current pace of progress continues, in the next 2-3 years we will see AIs capable not only of writing code but also of managing entire production lines, warehouses, and possibly even cars.
Incidentally, on June 13, Anthropic suspended access to the Fable 5 and Mythos 5 models due to a directive from the U.S. government on export controls. This serves as a reminder that the AI technology race is taking place not only in laboratories but also on the political field.