The AI agent Claude has surpassed humans by a head in controlling a robot dog: speed and efficiency have increased by orders of magnitude.
I have been closely following the development of Anthropic's Project Fetch, and the latest results have truly impressed me. The new version of the model, Claude Opus 4.7, demonstrated the ability to configure and program a four-legged robot 20 times faster than entire teams of human engineers. This is not just another improvement — it is a paradigm shift in the question of who should control physical systems.
For context: in August 2024, Anthropic first conducted a similar experiment. At that time, employees with no experience in robotics tried to program a robot dog, and the AI only acted as an assistant, speeding up the search for solutions. In the new testing phase, the situation changed dramatically. Claude Opus 4.7 worked almost completely autonomously with minimal human oversight. The neural network independently completed a whole range of tasks:
- connected to video sensors and LiDAR;
- wrote a program for manual control;
- created a system for monitoring the robot's path;
- configured an object recognition algorithm.
The numbers speak for themselves. The Opus 4.7 model turned out to be 18 times faster than a team using older AI versions, and 37 times faster than people who worked without chatbot assistance. But the most striking thing is the quality of the code. The volume of code written by the neural network turned out to be 10 times smaller than that of human teams. This indicates a much deeper understanding of the task and the ability to find elegant, efficient solutions.
It is critically important to emphasize that Anthropic did not introduce specialized algorithms for controlling the hardware. This progress in robotics has essentially become a side effect of the general scaling of language models. This confirms my long-standing hypothesis: the further development of AI will depend less and less on niche solutions and more and more on the fundamental capabilities of base models.
However, one should not think that AI has already completely replaced humans. Claude still experiences serious difficulties with precise physical actions that require complex real-time feedback. The model managed to guide the robot to the goal, but failed at the task of gently pushing a ball to a specific point. In this area, humans still maintain their superiority.
At Anthropic, they believe we are entering an era of "physical AI agents." I completely agree with this. In the near future, neural networks will use standard tools and equipment as effectively as they work with software code today. The labor market in engineering and robotics is facing fundamental changes.
My conclusion: This experiment clearly demonstrates that AI is already capable not just of assisting, but of independently solving complex engineering tasks. The key question now is not whether AI can replace humans in controlling robots, but how quickly and in which specific niches this will begin to happen. Investors and developers should closely monitor this trend — it will determine the development vector of the entire industry for years to come.