Claude can control your computer now, openclaw and zenmux updated same day

Anthropic has launched a groundbreaking 'computer use' feature for Claude, enabling the AI to directly interact with and control a user's Mac desktop. The model can now perform actions like opening applications, clicking buttons, scrolling web pages, and typing text, moving beyond API calls to simulate human-like computer interaction. This capability allows Claude to execute tasks such as organizing folders or performing multi-step desktop workflows autonomously, without requiring constant clarification. While currently limited to macOS, the feature represents a significant leap toward AI that can complete real work, not just discuss it.

Simultaneously, the agent framework OpenClaw pushed a major update, introducing a structured alternative to Claude's direct control approach. The update includes a new plugin SDK, an official plugin marketplace called ClawHub, and the ability to auto-map skills from various AI models including Claude, Codex, and Cursor. OpenClaw also upgraded its underlying models to M 2.7 and GPT-5.4. This positions OpenClaw as an 'agent OS' focused on orchestration and plugin-based workflows, which users report is better suited for complex, multi-step processes. The dual release highlights two competing visions: Anthropic's human-simulation model for quick tasks and OpenClaw's orchestrated, multi-model agent system for intricate workflows, both aiming to create AI that actively performs digital labor.