Anthropic's Claude can now control a user's computer to complete a variety of tasks, and the company is working on building an AI agent to rival the breakout OpenClaw. Anthropic announced on Monday that users can now send mission instructions to Claude through their mobile phones, and the AI agent will complete the operation.
Anthropic said that after receiving instructions, Claude can open applications on the computer, operate browsers and fill out spreadsheets. A demonstration video released on Monday shows a scenario where a user is late for a meeting and asks Claude to export the presentation to PDF format and add it to the meeting invitation. The video shows Claude successfully completing the task.
Anthropic's latest upgrade highlights the efforts of major AI companies to develop so-called "intelligent agents" - this type of AI can autonomously complete various tasks for users around the clock.
After OpenClaw was released and quickly became popular, AI agent-related functions have attracted much attention this year. OpenClaw can connect to the AI models of OpenAI and Anthropic, and users can send instructions to it through mainstream applications such as WhatsApp and Telegram to complete operations. Similar to Anthropic's new feature, OpenClaw runs locally on the user's device, providing direct access to files.
Nvidia CEO Jensen Huang told CNBC last week that OpenClaw is “definitely the next ChatGPT,” and technology companies are racing to develop similar products. The chip giant also released an enterprise version of OpenClaw last week - NemoClaw.
Last month, OpenAI hired OpenClaw developer Peter Steinberg to "drive the development of the next generation of personal intelligence."
safety measures
Anthropic cautions that Claude's computer control capabilities are "still in its early stages" compared to his ability to write code or interact with text.
"Claude may make mistakes, and although we continue to improve security protection mechanisms, threats are constantly evolving." Anthropic warned.
The company added that when developing its computer control features, it "incorporates security features that minimize risks" and that Claude will always ask users for permission before accessing new apps.
Users can use the Dispatch feature of Claude's collaborative office suite launched last week to continuously talk to Claude through their mobile phones or computers and assign tasks to the agent.