Claude Gains Mac Control, Pushing AI Agents Into Live Enterprise Workflows
The Development
Anthropic on March 24 released a research preview enabling Claude to directly control a user's Mac—opening applications, clicking buttons, typing into fields, and navigating software autonomously. The feature ships inside both Claude Cowork and Claude Code, and integrates with Dispatch, a mobile task-assignment tool launched the prior week. Claude operates via a priority hierarchy: native connectors first (Gmail, Slack, Google Drive), then Chrome browser automation, then direct screen interaction via screenshot analysis as a last resort. Access requires a Claude Pro subscription at $17/month or a Max plan at $100–$200/month. Early testing by MacStories found roughly a 50 percent task success rate. The launch arrives as Anthropic competes directly with OpenAI for enterprise accounts, and as open-source frameworks OpenClaw and Nvidia's NemoClaw have already normalized autonomous computer control. Enterprise audit logs currently do not capture Cowork activity—a material compliance gap.
Our Take
The 50 percent success rate is a distraction. Anthropic is not shipping a finished product—it's shipping category ownership. OpenClaw proved users will tolerate rough edges for autonomous computer control; Anthropic is now absorbing that demand into a managed, subscription-based product with guardrails and enterprise pricing. The real strategic move is the Dispatch pipeline: mobile-to-desktop task assignment with scheduled execution transforms Claude from a chat interface into background infrastructure. That's a fundamentally different procurement conversation with enterprise buyers. The compliance gap—audit logs that don't capture agent activity—is the single largest enterprise barrier right now, and Anthropic hasn't solved it. Marketing and ops teams evaluating agentic tools for Q2 pilots need to treat that gap as a blocking issue, not a footnote.
The Exploit
Action Item
If your organization is evaluating Claude Cowork for any regulated workflow, formally document the audit log gap with your compliance team before piloting—Anthropic's enterprise controls currently produce no centralized record of agent actions, which creates liability exposure that won't be resolved informally.