• Anthropic showed a robot controlled by Claude.
• The AI model surpassed the company's employees in working with it.
• However, the chatbot still could not successfully complete the experiment's final task.

Anthropic presented the results of the second phase of the Project Fetch experiment, in which the AI model Claude Opus 4.7 learned to work with a commercial quadruped robot on its own. During the tests, the model, without human help, completed a series of tasks for connecting to sensors, programming, and navigating the robot, surpassing the results of Anthropic's employees.

Anthropic first launched Project Fetch in August 2025. Back then, company employees with no robotics experience were split into two teams: one was given access to Claude, the other relied only on its own knowledge and searching for information on the internet.

A guide to Claude: how Anthropic's neural network works with code and text 11.03.2026 Read

The results showed that the team with Claude worked significantly more efficiently. At the same time, the model of that generation itself — Claude Opus 4.1 — could not complete the entire set of tasks on its own and got stuck already at the stage of connecting to the robot.

In less than a year the situation changed dramatically.

Claude surpassed humans

In the new testing, Anthropic connected a laptop with Claude Code directly to the robot. The researcher's role was limited to launching the initial prompt and confirming commands.

Claude Opus 4.7 was able to:

• connect to the robot's cameras and sensors;
• set up access to LiDAR data;
• write control software;
• create a route-tracking system;
• implement object recognition;
• integrate the individual components into a single system.

According to Anthropic, on every task that at least one human team completed during the first phase of the experiment, Claude Opus 4.7 delivered a result at least ten times faster.

If we account for the four tasks that both human teams completed, the model turned out to be more than 37 times faster than the team without Claude and almost 19 times faster than the team that used an AI assistant.

The company noted that the model also produced almost ten times less program code than the humans, while achieving similar or better results.

Where AI still lags behind humans

Despite significant progress, Claude could not successfully complete the experiment's final task — autonomously returning the ball to the starting point using the robot dog.

The model was able to correctly determine the position of the ball and position the robot for a push, but it could not handle precise control of movements in real time.

Anthropic explained that the problem lies in the so-called closed control loop, where the system must continuously evaluate the result of each action and adjust the next commands depending on changes in the environment.

It is precisely in such scenarios that humans still retain an advantage over large language models.

AI that knows how to doubt: testing Claude Opus 4.8 29.05.2026 Read

At the same time, one of the company's experienced roboticists was able to program the autonomous execution of this task. In Anthropic's opinion, current versions of Claude can also achieve such a result given additional tools and time.

Anthropic sees the dawn of an era of physical agentic AI

The company stressed that improving Claude's robotic capabilities was not a separate development goal. The progress was a side effect of the overall scaling and refinement of the models.

Anthropic believes that the industry is gradually going through a familiar development cycle:

• first, the models help humans;
• then humans help the models;
• and in the end, the models begin to perform tasks on their own.

A similar process was previously observed in the field of programming and cybersecurity, where AI agents learned to work autonomously with software tools.

Now a similar transformation is beginning to take place in the physical world.

"We are probably entering an early era of physical agentic AI," — Anthropic noted.

Anthropic continues to expand Claude's capabilities

The results of Project Fetch were another step in the company's strategy for developing autonomous AI systems.

Earlier, Anthropic published a large-scale study of the impact of artificial intelligence on the labor market and concluded that for now the technology mainly increases workers' productivity rather than leading to mass layoffs. More in the article:

Which professions will AI replace? Anthropic conducted its own study 06.03.2026 Read

In addition, after surveying 81,000 Claude users, the company recorded a contradictory attitude toward AI: almost half of the respondents reported expanded capabilities thanks to the technology, although about 20% expressed concerns about job loss.

Also, the head of the Claude Code and Cowork product area, Cat Wu, recently said that the next stage in the development of artificial-intelligence systems will be proactivity — models will be able to anticipate a user's needs even before the user formulates a request.

Source: Incrypted