Anthropic has released a new capability for Claude: the ability to control your computer and perform tasks autonomously — scrolling, clicking, opening files, interacting with apps — as if it were sitting at your keyboard. It is a significant step in the evolution of agentic AI, and it carries implications that go well beyond productivity.
What Claude computer use actually does
The feature, currently available as a research preview for Claude Pro and Claude Max subscribers on macOS, allows Claude to take direct control of your machine to complete tasks you assign to it. Claude can use your web browser, open and edit files, interact with development tools, manage calendar entries via Google Calendar, send messages via Slack, and more.
The mechanics are straightforward: if a dedicated connector exists for an app, Claude uses it. If not, Claude operates the interface directly — moving the cursor, clicking, scrolling, and typing — in the same way a human operator would. According to CNET's coverage of the launch, Claude always asks for permission before performing any action, and can be stopped at any point.
Anthropic has also announced integration with Dispatch, a companion feature that allows users to assign tasks to Claude remotely from a mobile device — checking email, launching a Claude Code session, or generating a morning briefing — even when away from the computer.
The security considerations
Giving an AI model direct access to your operating environment is not a trivial decision. Anthropic has been explicit about the current limitations: this is a research preview, the feature is new and may contain errors, and the company recommends against using it with applications that handle sensitive data. Several such apps are disabled by default.
The risks that experts have flagged for agentic AI more broadly — prompt injection attacks, unintended actions, data exposure — apply here. Anthropic says it has implemented automatic scanning for these vulnerabilities, but the advice remains measured: treat this as a powerful early tool, not a production-ready autonomous agent.
Why this matters for AI visibility
The strategic significance of this release extends well beyond its immediate utility as a productivity tool. What Anthropic is demonstrating — and what OpenClaw, NemoClaw, and similar frameworks are collectively accelerating — is a fundamental shift in how AI interacts with the digital environment.
Until recently, AI models were primarily retrieval and generation engines: they answered questions, summarised content, and produced text. The agentic layer changes this. AI systems are now beginning to act — to navigate interfaces, retrieve information from live sources, and make decisions autonomously on behalf of users.
This has a direct consequence for brand visibility. When a user asks Claude to find a supplier, compare service providers, or research options in a given category, Claude will not return a list of links for the user to evaluate. It will navigate, read, synthesise, and recommend. The brand that gets cited in that recommendation — or excluded from it — is determined not by search position, but by how clearly and consistently AI models understand what that brand offers and why it matters.
In an agentic AI environment, brand visibility is not about appearing in results. It is about being understood, trusted, and selected by the AI acting on behalf of the user.
The citability question becomes more urgent
The Citability Score framework was developed precisely for this transition. As AI moves from answering questions to performing tasks, the brands that have invested in machine-readable content, consistent entity definitions, and structured authority signals will have a measurable advantage — not just in AI-generated responses, but in AI-driven actions.
If Claude is managing a user's calendar and the user asks it to book a venue for an event, Claude will draw on what it knows about available options in the relevant category. If your brand is well-cited, clearly described, and structurally consistent across authoritative sources, it enters that decision process. If it is not, it is invisible to the agent — regardless of how strong its SEO profile may be.
The AI Visibility Index measures exactly this kind of presence: not just citation frequency, but citation quality, context, and position in AI-generated responses. As agentic capabilities expand, this measurement layer becomes more — not less — relevant.
A research preview, not a finished product
It is worth being precise about what this release is and what it is not. Anthropic has been careful to position computer use as an early-stage capability. Complex tasks may not work reliably on the first attempt. The integration with Dispatch is promising but also new. The broader ecosystem of "agentic" frameworks — OpenClaw, NemoClaw, and others — is still consolidating.
What the release signals, however, is unambiguous: the major AI labs are actively building toward a world where AI agents act on behalf of users across digital environments. That transition is not speculative. It is in production, available today to millions of subscribers, and expanding.
For brand and marketing teams, the question is no longer whether to take AI visibility seriously. It is how quickly they can build the infrastructure that makes their brand legible to the AI systems that are increasingly acting as intermediaries between intent and action.
Source: CNET — Anthropic's Claude Can Now Control Your Computer. For Anthropic's official documentation on computer use, see anthropic.com.