Anthropic has demonstrated: AI Claude controls a robot dog 20 times faster than engineers
A new phase of Anthropic's Project Fetch experiment has demonstrated an impressive leap in AI autonomy. The Claude Opus 4.7 model handled the setup and control of a four-legged robot 20 times faster than the best teams of human engineers working with the previous version of the neural network a year ago. This is not just an acceleration—it is a qualitative shift: the AI took over virtually the entire programming process.
Unlike the first phase in August 2024, when employees with no robotics experience only used AI as an assistant, the new testing phase showed that the model can operate with minimal oversight. Under the supervision of just one researcher, Claude Opus 4.7 autonomously performed a series of complex tasks:
- connected to video sensors and LiDAR;
- wrote a program for manual robot control;
- created a trajectory monitoring system;
- configured an object recognition algorithm.
The numbers speak for themselves: Opus 4.7 was 18 times faster than teams using older AI versions and 37 times faster than humans working without chatbot assistance. Moreover, the code generated by the neural network was 10 times more compact than that of human teams. This indicates that the AI is not merely repeating patterns but finding more efficient solutions.
It is important to emphasize that this progress was achieved without introducing specialized algorithms for hardware control. As the developers note, this is a side effect of the general scaling of language models. In other words, the ability to control physical objects grows on its own as the underlying AI architectures improve.
However, there were limitations. Despite success in navigation, Claude still struggles with tasks requiring fine physical coordination. The model guided the robot to the target but could not gently push a ball to the right spot. This confirms that in the realm of complex real-time feedback, humans still hold an advantage.
At Anthropic, they are convinced: we are on the threshold of an era of "physical AI agents." In the near future, neural networks will be able to work with standard hardware as naturally as they currently do with software code. In my view, this claim is not without merit: the progress shown over the past year is impressive, and if the trend continues, the boundary between the digital and physical worlds will become even more blurred.
Analytical commentary: The automation and robotics market may be in for a tectonic shift. If AI learns to effectively manage physical agents without specialized training, it will radically change logistics, manufacturing, and even household robotics. However, it is too early to talk about a complete replacement of humans—tasks requiring high tactile sensitivity remain an area where AI falls short.