Anthropic Ships Claude 3.5 Sonnet with Public Computer Use API
Anthropic released an API endpoint that lets Claude 3.5 Sonnet move the mouse, type on the keyboard, and drive the browser. Users send screenshots and task instructions through the API; the model returns coordinate clicks and keystrokes until the desktop task finishes. No custom code or developer hiring is required for routine file, spreadsheet, or web workflows.
This demonstrates agentic control loops where an LLM observes the screen state and issues low level actions. Practitioners can replace brittle scripts with natural language instructions that adapt to changing interfaces. The workflow shifts from writing automation code to describing desired outcomes and letting the model iterate.
Anthropic documented the feature at anthropic.com/news/3-5-models-and-computer-use. Early testers report completing invoice data entry and form submissions in under two minutes without writing Python or hiring contractors.
Step 1: Visit console.anthropic.com, create a project, and enable the computer use beta. Step 2: Send a POST request with a base64 screenshot and the prompt 'Click the export button and save the CSV'. Step 3: Parse the returned action list and execute the mouse and keyboard events locally; the file appears in your downloads folder.