Agent Warfare Heats Up: Claude Can Operate Computers on Behalf of Users, AI Agents Moving Toward Practical Implementation

American artificial intelligence company Anthropic announced that its Claude model can now manipulate users’ computers to complete tasks, indicating that the company is also working on developing an AI agent that can compete with OpenClaw (commonly known as “Lobster”).

According to Anthropic, after receiving user commands, Claude can open applications on the computer, browse the web, and fill out spreadsheets. This feature is already available in macOS versions of Claude Cowork and Claude Code.

Anthropic released a demonstration video showing a scenario: a user is about to be late for a meeting and sends a command via phone for Claude to export a presentation as a PDF and attach it to the meeting invite. The video shows Claude successfully completing this task.

This latest update from Anthropic highlights the company’s accelerating efforts to develop “agents.” With the release of OpenClaw, these types of AI agents have become a focus of attention and quickly gained popularity.

Last week, NVIDIA CEO Jensen Huang announced the launch of the NemoClaw software stack at the annual GTC conference to support OpenClaw. In an interview, Huang stated that OpenClaw “is definitely the next ChatGPT.”

Last month, OpenAI hired Peter Steinberger, the creator of OpenClaw, to help develop the “next generation of personal AI agents.”

However, Anthropic cautions that compared to Claude’s capabilities in programming or text interaction, “computer operation” is still in its early stages. “Claude may make mistakes, and although we are continuously improving safety measures, threats are also evolving.”

Anthropic added that it has incorporated “safety protections to minimize risks” into this functionality, and Claude always requests user permission before accessing new applications.

(Source: Cailian Press)