US artificial intelligence company Anthropic announced that its Claude model now has the capability to manipulate users' computers to perform tasks, including opening applications, browsing the web, filling out spreadsheets, and more. This functionality has been implemented in the macOS versions of Claude Cowork and Claude Code. A demo video shows that Claude can successfully respond to user instructions and complete tasks. This update reflects that AI companies are accelerating the development of 'agent' technology. Previously, after the release of OpenClaw, such capabilities received widespread attention. Nvidia introduced a software stack supporting OpenClaw, and OpenAI also hired its creator to drive related progress. However, Anthropic reminds that the current 'computer operation' function is still in its early stages, and the company has incorporated safety measures. When Claude needs to access new applications, it will request user permission.