[Claude Completes the Final Piece of the Agent Puzzle: Now Capable of Directly Controlling Computers and Browsers] Beating AI Newsflash: Anthropic has officially enabled Claude's computer operation capabilities and browser tools. Previously, every mouse click, input, or action performed by Claude required the model to be invoked again; now it can complete clicks, inputs, keystrokes, and screenshots continuously within a single task. The updated browser tool no longer relies solely on screenshots to locate buttons but simultaneously reads webpage structures and visual information, allowing it to directly interact with page elements. When webpage layouts change, it is less likely to fail due to button position shifts. Anthropic has also launched the Skills API and Files API. Developers can encapsulate enterprise workflows into reusable Skills, enabling Claude to execute tasks according to fixed processes; uploaded files can also be reused across requests. [Original Link]