Anthropic's AI agent taught a robot dog to follow commands 20 times faster than humans
Anthropic has presented the results of the second phase of the Project Fetch experiment, and they are impressive. The Claude Opus 4.7 model fully autonomously handled the setup and control of a four-legged robot, completing tasks 20 times faster than a team of human engineers working with the previous version of the AI.
As a reminder, in the first phase, which took place in August 2024, the neural network only assisted humans, speeding up the search for solutions. Now the situation has changed dramatically: Claude Opus 4.7 worked almost without human involvement, under minimal researcher supervision. The algorithm independently connected to video sensors and LiDAR, wrote a program for manual control, created a trajectory monitoring system, and configured an object recognition algorithm.
Numbers That Speak for Themselves
The performance of the new model is astonishing: it turned out to be 18 times faster than a team using older AI versions and 37 times faster than people working without chatbot assistance. Moreover, the code written by the neural network turned out to be 10 times more compact than human-written code. This is direct proof that AI is capable not only of accelerating processes but also of optimizing them in terms of efficiency.
Particularly important, according to the developers, is that this progress was a side effect of the general scaling of language models. Anthropic did not introduce specialized algorithms for robot control—the model simply learned to work with hardware by expanding its cognitive abilities.
Limitations and Prospects
However, there were caveats. Claude still experiences serious difficulties with precise physical actions. The model managed to guide the robot to its target but failed at the task of gently pushing a ball. This requires complex real-time feedback, where humans still hold an advantage.
Nevertheless, Anthropic is confident that we are entering the era of "physical AI agents." I share this optimism: current results show that neural networks can already work with standard tools and equipment as effectively as with software code. This paves the way for fully autonomous robotic systems capable of performing complex tasks without human intervention.
My comment: The Anthropic experiment is not just a demonstration of one model's capabilities but a signal for the entire industry. If AI can adapt to the physical world so quickly without specialized training, then we are on the brink of a revolution in robotics. In the coming years, we will see neural networks begin to manage not only virtual but also real assets, which will fundamentally change the automation market.