Agentic AI
Anthropic Ships Computer Use API for Autonomous Desktop Agents
Developers can now let Claude move the cursor, click and type across operating systems, opening a path to automating desktop workflows that resist traditional scripting.
Technology
Multimodal AI
Vision and language together for screen and document understanding.
Related news
Agentic AI
Google DeepMind Previews Project Mariner for Browser Agents
A prototype agent running inside Chrome can complete multi-page research tasks, extract structured data and fill long forms while showing its reasoning in a side panel.
Research
Research: Multimodal Agents Approach Human-Level Performance on Web Tasks
A benchmark of realistic web errands shows frontier multimodal agents completing tasks at roughly 70% of human speed and accuracy, up from under 20% a year ago.