Anthropic has crossed a significant threshold. On March 24, 2026, the company launched the most ambitious consumer AI agent to date, giving Claude the ability to directly control a user’s Mac鈥攃licking buttons, opening applications, typing into fields, and navigating software on the user’s behalf.
This isn’t science fiction or vaporware. The features are available immediately for Claude Pro subscribers on macOS, marking a new phase in the competition to build AI that actually does work rather than merely discussing it.
From Conversation to Action
The computer use feature transforms Claude from a sophisticated chatbot into something closer to a remote digital operator. When a user assigns a task, Claude follows a layered priority system: first checking for direct connectors to services like Gmail, Google Drive, Slack, or Google Calendar; then falling back to navigating Chrome via browser extension; and finally interacting directly with the screen by clicking, typing, scrolling, and opening applications.
This hierarchy reveals how Anthropic thinks about reliability versus reach. Direct connectors are fastest and most reliable. Screen-level interaction is most flexible but slowest and most fragile.
Dispatch: Your iPhone as an AI Remote Control
The strategic centerpiece may actually be Dispatch鈥攁 feature that pairs Claude on your phone with Claude on your desktop. Users scan a QR code to link devices, then text instructions from anywhere. Claude executes on the desktop and sends back results.
Anthropic envisions practical applications: having Claude check email every morning, pull weekly metrics into reports, organize cluttered folders, or compile competitive analyses from local files. Scheduled tasks allow setting a cadence once and letting Claude handle the rest.
Early Testing: Promising but Imperfect
Anthropic calls this a research preview for good reason. Early hands-on testing shows the feature works well for information retrieval and summarization but struggles with complex multi-step workflows.
Detailed testing found Claude successfully located specific screenshots, summarized notes from Notion, added URLs to databases, and recalled screenshots from earlier sessions. However, it failed at more complex tasks. The verdict: about a 50/50 shot whether what you try will work.
The Agentic AI Race Heats Up
Anthropic’s timing is deliberate. OpenAI, Google, Nvidia, and countless startups are all chasing the same goal: building AI that operates inside your existing tools rather than beside them.
Claude’s computer use works through screenshots of the user’s desktop. Anthropic trains it to avoid stock trading, inputting sensitive data, or gathering facial images, but acknowledges these guardrails are not absolute.
We’re not there yet. But with Claude now clicking buttons on Macs worldwide, the trajectory is becoming clear. The AI agent revolution hasn’t arrived in full force, but its advance guard is definitely here.