AI engineer Claude Opus 4.7 handled robot dog control 20 times faster than humans — a new stage of Project Fetch
We are witnessing a landmark milestone in the development of artificial intelligence: Anthropic's Claude Opus 4.7 language model has demonstrated the ability to independently program and control a four-legged robot, completing tasks 20 times faster than teams of human engineers. This is the second phase of the Project Fetch experiment, and the results are impressive.
In August 2024, when the first phase of the project launched, the AI only acted as an assistant, helping employees without robotics experience find solutions faster. Now, the Opus 4.7 model worked almost autonomously, under minimal researcher supervision. The neural network independently completed the full cycle of tasks:
- connected to video sensors and lidar;
- wrote a program for manual control;
- created a robot path monitoring system;
- configured an object recognition algorithm.
Comparative analysis shows a colossal gap in performance. Opus 4.7 turned out to be 18 times faster than the team using previous AI versions, and 37 times faster than people working without chatbot assistance. Moreover, the neural network generated code that was 10 times smaller in volume than that of human teams. This indicates not only speed but also a qualitatively different level of efficiency.
The key conclusion reached by the experiment's authors: progress in robotics has become a side effect of the general scaling of language models. Anthropic did not introduce specialized algorithms for controlling hardware — the model simply learned to understand context and apply its knowledge to physical objects.
However, not everything is so smooth. Claude still struggles with precise physical actions. The model managed to guide the robot to the target but failed at the task of gently pushing a ball to the right spot. This requires complex real-time feedback, where humans still maintain an advantage.
Anthropic states that the industry is entering an era of "physical AI agents." In the near future, neural networks will be able to use standard tools and equipment as effectively as they currently work with software code.
Expert opinion. This experiment is a clear signal that the boundary between the digital and physical worlds is blurring. The fact that a model without specialized training can control a robot faster and more efficiently than a human changes the rules of the game in automation. However, the failure with the ball reminds us: coarse physics and real-time nuances remain the last bastion that AI has not yet stormed. Investors and developers should prepare for the next generation of neural networks to not only write code but also control machine tools, drones, and possibly cars.