Claude completes the last piece of Agent's puzzle: can now directly manipulate computers and browsers

Anthropic officially opens Claude's computer operating capability and browser tools. Claude can now do clicks, input, keys and screenshots in one job without the need for the model to be called again. The new browser tool is able to read web structure and visual information at the same time, and the web page is operated directly on the basis of the page elements, and it is not easy to fail when the page layout changes occur. In addition, Anthropic has opened skills API and files API to enable developers to encapsulate business processes as reusable skills, to enable Claude to carry out tasks according to a fixed process, and to re-use documents over and over again after uploading them。