In a first-look review, Simon Willison describes Anthropic's newly released Claude Cowork as a "general agent" that brings the powerful automation capabilities of Claude Code to non-developers through a friendlier desktop interface—while cautioning that prompt injection risks remain a serious unresolved security concern.​

Willison tested Cowork by having it scan 46 of his draft blog posts, cross-reference them against his published site via 44 individual searches, and identify which were closest to being ready for publication—all running in what appears to be a sandboxed, containerized environment.​

Cowork arrives as AI agents gain momentum: OpenAI launched its ChatGPT Agent browser automation tool in July 2025, and one enterprise survey found agentic coding tool adoption among engineers jumped from 20% to over 80% in 2025.