The race to build truly useful AI agents just entered a new dimension. On Monday, Anthropic announced a groundbreaking update to its Claude AI assistant: the ability to directly control a user’s Mac computer. This isn’t a gimmick or a demo—it’s a research preview available now to paying subscribers that transforms Claude from a conversational assistant into what amounts to a remote digital operator.
What Claude Can Now Do on Your Mac
The new computer use capabilities allow Claude to perform a wide range of desktop actions:
- Clicking buttons and interacting with UI elements
- Opening and closing applications
- Typing text into any field
- Navigating software the way a human would
- Taking screenshots to understand what’s on screen
- Managing files across the filesystem
- Composing and sending emails through connected clients
- Scheduling tasks and setting reminders
The feature is available immediately for Claude Pro subscribers (starting at $17 per month) and Max subscribers ($100 or $200 per month), though currently only on macOS. Windows users have been left out of this initial research preview.
How It Works: The Three-Tier Approach
Anthropic has implemented a clever priority system for how Claude decides whether to use direct connectors, browser navigation, or screen-level interaction.
Tier 1: Direct Connectors
First, Claude checks for integrations with services like Gmail, Google Drive, Slack, and Google Calendar. These connectors provide the fastest and most reliable path to completing tasks. Pulling messages through a Slack connection takes seconds, whereas navigating Slack through screen-level interaction would be much slower and more error-prone.
Tier 2: Browser Navigation
If no direct connector is available, Claude falls back to navigating through Chrome using Anthropic’s dedicated extension. This is more flexible than connectors but slower and more prone to errors.
Tier 3: Screen-Level Interaction
Only as a last resort does Claude interact directly with the user’s screen—clicking, typing, scrolling, and opening applications the way a human operator would. This mode is the most flexible since it can theoretically work with any application, but it’s also the slowest and most fragile.
Dispatch: Your iPhone as a Remote Control
The real strategic innovation might not be computer use itself, but how Anthropic is pairing it with Dispatch—a feature that lets users assign Claude tasks from their mobile phone.
A user pairs their iPhone with their Mac by scanning a QR code, and from that point forward, they can text Claude instructions from anywhere. Claude executes those instructions on the desktop—which must remain awake and running the Claude app—and sends back results.
Imagine sending a text from your commute: “Compile my weekly metrics into a report and email it to the team.” By the time you arrive at the office, the task is done. This is the vision Anthropic is selling—not just an assistant at your desk, but an autonomous agent working for you even when you’re not there.
Early tester Gagan Saluja captured the significance: “Combine this with /schedule that just dropped and you’ve basically got a background worker that can interact with any app on a cron job. That’s not an AI assistant anymore, that’s infrastructure.”
The Competition Heats Up
Anthropic’s timing is far from accidental. The company is shipping these capabilities into a market that has been rapidly reshaped by the rise of OpenClaw, the open-source framework that enables AI models to autonomously control computers. OpenClaw exploded earlier this year, proving that users wanted AI agents capable of taking real actions on their computers.
Nvidia entered the fray with NemoClaw, its own framework for simplifying OpenClaw deployment. Smaller startups like Coasty are also pushing into the space, marketing “full browser, desktop, and terminal automation with a native experience.”
Meanwhile, Reuters has reported that OpenAI is actively courting private equity firms in what it describes as an “enterprise turf war with Anthropic”—a battle in which the ability to ship working agents is becoming the decisive weapon.
The Reality Check: 50% Success Rate
Despite the hype, early hands-on testing reveals significant limitations. John Voorhees of MacStories published a detailed evaluation showing mixed results. While Claude successfully located specific files, summarized notes in Notion, and added URLs to databases, it failed to open the Shortcuts app, send screenshots via iMessage, list unfinished Todoist tasks, or fetch URLs from Safari.
Voorhees’ verdict was measured: “It’s not good enough to rely on when you’re away from your desk” but “a step in the right direction.” Other users have reported that the new features are consuming usage quotas at alarming rates, with one Max subscriber complaining that Dispatch was eating 10% of their monthly allowance in a single prompt.
Security Concerns: Letting AI Control Your Desktop
Perhaps the most significant concern is security. Computer use runs outside the virtual machine that Cowork normally uses for file operations, meaning Claude is interacting with the user’s actual desktop and applications—not an isolated sandbox.
Anthropic has built several layers of defense: Claude requests permission before accessing each application, sensitive apps like investment platforms are blocked by default, and users can maintain a blocklist of applications Claude is never allowed to touch. The system also scans for signs of prompt injection during computer use sessions.
But the company is remarkably forthright about the limits of these protections. The help center documentation explicitly warns users not to use computer use to manage financial accounts, handle legal documents, process medical information, or interact with apps containing other people’s personal information.
For enterprise customers, there’s an additional complication: Cowork conversation history is stored locally on the user’s device, not on Anthropic’s servers. Critically, enterprise features like audit logs, compliance APIs, and data exports do not currently capture Cowork activity. This means organizations subject to regulatory oversight have no centralized record of what Claude did on a user’s machine.
As one user pointedly asked on social media: “When the agent IS the user (same mouse, keyboard, screen), traditional forensic markers won’t distinguish human vs AI actions. How are we thinking about audit trails here?”
The Bigger Picture: Agents as Infrastructure
What Anthropic is really selling is a vision of AI as infrastructure—background systems that handle repetitive work without constant human oversight. The testimonials Anthropic has gathered suggest this pitch is landing with some organizations.
Larisa Cavallaro, an AI Automation Engineer, described connecting Cowork to her company’s tech stack and asking it to identify engineering bottlenecks. Claude returned “an interactive dashboard, team-by-team efficiency analyses, and a prioritized roadmap.” CTO Joel Hron offered a more philosophical framing: “The human role becomes validation, refinement, and decision-making. Not repetitive rework.”
Looking Forward
We’re witnessing a pivotal moment in AI development. The ability for AI agents to actually perform work—rather than just suggest or describe how work could be done—is rapidly becoming the central battleground in the AI industry.
Anthropic’s computer use feature represents the most ambitious consumer-facing implementation of this vision to date. Whether its limitations—inaccuracy, security concerns, and regulatory gaps—will be overcome in time to fulfill the promise remains to be seen. But one thing is clear: the era of AI as a passive tool is definitively ending.
The question now isn’t whether AI will control our computers. It’s whether we’re ready for what that means.