Claude Can Now Control Your Computer

Claude steps off the chat window and onto your desktop

Anthropic has launched a significant new capability for its Claude AI: the ability to take control of a user's computer and autonomously complete tasks. Users can now send a prompt from a smartphone, and Claude will open applications, navigate browsers, fill spreadsheets, and carry out multi-step workflows on their desktop. It is a meaningful step towards genuinely useful AI agents, and it places Anthropic squarely in competition with a fast-moving field of autonomous AI tools.

What Claude Can Actually Do Now

The new feature, announced on Monday, allows Claude to act on a user's computer in response to natural-language instructions sent from any device, including a phone. In a demonstration video, Anthropic showed a user running late for a meeting who asked Claude to export a pitch deck as a PDF and attach it to a calendar invite. Claude completed the task end to end without further input.

The capability extends to opening desktop applications, navigating a web browser, and populating spreadsheets. Claude runs locally on the user's device, which means it has access to local files and applications, rather than operating through a cloud-based intermediary. This design mirrors the approach taken by OpenClaw, the viral third-party agent that sparked the current industry-wide sprint towards agentic^✦ AI.

"Claude can make mistakes, and while we continue to improve our safeguards, threats are constantly evolving." - Anthropic

The company was notably candid about the product's current limitations. Anthropic acknowledged that computer use is "still early compared to Claude's ability to code or interact with text." As a safeguard, Claude will always request permission before accessing a new application, and the company says it has built the feature with controls designed to minimise risk. Whether those safeguards prove sufficient in real-world use at scale^✦ remains to be seen.

The OpenClaw Effect

No discussion of AI agents in early 2026 is complete without addressing OpenClaw, the app that changed the conversation. OpenClaw is a third-party tool that connects to AI models from both OpenAI and Anthropic, and can be messaged through WhatsApp or Telegram to carry out tasks on a user's device. Its consumer-friendly interface and viral growth put agentic AI on the mainstream radar in a way that API^✦ announcements never quite managed.

"OpenClaw is definitely the next ChatGPT." - Jensen Huang, CEO, Nvidia

The response from the industry has been swift and competitive. OpenAI hired Peter Steinberger, OpenClaw's creator, explicitly to lead the development of its next generation of personal agents. Nvidia announced NemoClaw, an enterprise-grade equivalent designed for business use. And now Anthropic is bringing its own native answer to market. The race is no longer about language model benchmarks. It is about who can most reliably act on your behalf in the real world.

For readers interested in the broader competitive dynamics at Anthropic, the company is also navigating significant corporate milestones, including a potential IPO process covered in our report on Anthropic's rumoured October IPO at a $380 billion valuation. The launch of agentic features like computer use will directly influence that valuation conversation.

Dispatch and the Cowork Ecosystem

Alongside the computer use announcement, Anthropic has integrated the feature into its emerging productivity platform. Dispatch is a newly released feature inside Claude Cowork that allows users to maintain a continuous conversation with Claude across phone and desktop, assigning the agent tasks that it then carries out asynchronously. Think of it as a persistent AI assistant that does not disappear when you close a tab.

This architecture matters. Rather than a one-shot query-and-response model, Dispatch enables an ongoing working relationship between user and agent. You can assign a task, move on, and check back when it is done. This mirrors how people actually work, and it is a meaningful design shift from the chatbot paradigm that has dominated AI consumer products since 2022.

• Computer use: Claude can open apps, browse the web, and handle files on your desktop.
• Dispatch: Assign tasks from phone or desktop and maintain a continuous agent workflow.
• Permission gates: Claude requests access before entering any new application.
• Local processing: The feature runs on-device, giving Claude access to local files without cloud routing.

For those looking to get the most from this kind of agentic workflow, understanding how to frame instructions clearly becomes critical. Our guide to the 8-part Claude prompt framework that works is worth reading before you start delegating your entire task list to an AI agent.

The Asia-Pacific Picture

The agentic AI race has particular resonance across Asia-Pacific, where mobile-first behaviour and high smartphone penetration make the remote-prompt, desktop-execution model especially compelling. In markets like Japan, South Korea, and Southeast Asia, users are often managing work across devices throughout the day. An agent that can receive a WhatsApp or Telegram instruction from a commuter and complete a desktop task before they arrive at the office is not a hypothetical. It is a genuine productivity unlock.

The viral spread of OpenClaw is itself a testament to this dynamic. The app gained significant traction precisely because it integrated with messaging platforms already embedded in daily life across Asia. In countries such as Indonesia, Vietnam, and Thailand, WhatsApp and Telegram are primary communication tools, not secondary ones. Anthropic's move to build native computer use capabilities into Claude is partly a response to that momentum.

The broader AI agent market in Asia is also expanding rapidly. According to the Boao Forum for Asia, the region's AI sector is on track for explosive growth, and the race to dominate the agentic layer is intensifying. Our coverage of the Boao Forum's declaration of Asia as the new AI epicentre with a $400 billion market by 2030 provides useful context for how regional governments and businesses are positioning themselves.

Singapore, which has invested heavily in AI infrastructure and regulation, is likely to be a key early-adopter market for enterprise agentic tools. The city-state's upcoming SuperAI 2026 conference is expected to feature significant discussion around autonomous agents as the next frontier for enterprise deployment across the region.

Safety, Autonomy, and the Trust Problem

The enthusiasm around computer-use agents carries a shadow: trust. Allowing an AI model to operate applications, access files, and take actions on your behalf requires a level of confidence in the system's reliability and integrity that current AI cannot fully justify. Anthropic's own warning that "threats are constantly evolving" is a rare piece of corporate candour, but it raises questions about what those threats actually look like in practice.

There are plausible failure modes. An agent that misinterprets an instruction and deletes files, sends a draft email prematurely, or fills in a form with incorrect data could cause real harm. Prompt injection attacks, where malicious content in a webpage or document tricks the agent into taking unintended actions, are a known concern in the research community. Anthropic has not detailed how its safeguards address this specific threat vector.

The permission-gate model, requiring approval before accessing new apps, is a sensible start. But for power users who may grant broad permissions up front, the safety net essentially disappears. This is the core tension in agentic AI: the more autonomous the agent, the more useful it becomes, but also the more damage it can do if something goes wrong. Our editorial on why you should stop letting AI do your thinking for you explores the broader cognitive and professional implications of over-reliance on AI agents.

Frequently Asked Questions

How does Claude's computer use feature actually work?

Users send a text prompt to Claude from their smartphone or another device. Claude then operates directly on the user's computer, opening applications, navigating browsers, managing files, and completing multi-step tasks without requiring further input. It runs locally on the device rather than through a remote server, which gives it access to local files and apps.

What is OpenClaw and why does it matter for understanding Claude's new capability?

OpenClaw is a third-party AI agent application that connects to models from both OpenAI and Anthropic. It can be operated via WhatsApp or Telegram and runs on the user's device to complete tasks. Its viral growth in early 2026 accelerated the entire industry's push towards agentic AI, prompting responses from Anthropic, OpenAI, and Nvidia in quick succession.

Is it safe to let Claude use my computer?

Anthropic has built in permission gates that require Claude to request access before using new applications. However, the company itself has cautioned that computer use is an early-stage capability and that risks remain. Users should be aware of the potential for errors, misinterpreted instructions, and evolving security threats before granting broad permissions.