Anthropic’s Claude Takes Control of Your Computer: Could It Dethrone OpenClaw?

On March 23, Anthropic announced a significant new capability for Claude, allowing it to control a user’s computer to execute tasks, marking a major milestone for autonomous AI agents that interact with applications and manage software like a human would. This update comes in the context of the rising trend of AI agents, especially fueled by the growing interest in the open-source agentic framework “OpenClaw.”

The enhanced functionality of Claude now enables it to carry out tasks like locating and transferring files directly from a user’s hard drive. Users can initiate these processes through prompts in the Claude app on their smartphones, which commence with Claude searching for appropriate tools via integrations with applications such as Google Calendar or Slack.

If Claude fails to identify the right application or tool, it can perform the task manually, mimicking human-like actions such as scrolling, clicking, typing, or moving the cursor to open files, browse the web, and launch developer tools, effectively acting as if it were using a keyboard and mouse.

While users must approve access for any applications Claude intends to utilize (in the absence of available connectors), it is not necessary for every action it undertakes. In a demonstration video, Claude articulates its planned actions upfront and confirms completion once the task is accomplished. This computer-use update has been rolled out as a research preview for users with MacOS devices who subscribe to Claude Pro and Claude Max plans.

Anthropic has introduced a new feature, Dispatch, allowing users to assign tasks to Claude via their phones. “Assign a task from your phone, shift your focus to other tasks, and return to complete work on your computer,” stated Anthropic in a post on X. Users can delegate a variety of routine tasks to Claude, from checking emails to exporting files, and may also initiate live Claude Cowork or Claude Code sessions to generate code or execute tests. Due to being in research preview, some more complex tasks may encounter difficulties initially.

Anthropic’s latest update for Claude is particularly notable as LLM-powered AI agents are gaining traction, in part due to the rise of OpenClaw. This open-source framework enables developers to orchestrate AI agents in structured processes.

Anthropic’s Acquisition Strategy

Nearly two years ago, Anthropic first hinted at the concept of computer-use by AI systems. In 2024, the company unveiled an upgraded Claude 3.5 Sonnet model capable of understanding and engaging with any desktop application by imitating keystrokes, button clicks, and mouse movements.

The ‘Computer Use’ API was launched in open beta, but it reportedly faced issues, often hampered by errors and excessive token usage due to its dependence on LLM reasoning. The model needed to take screenshots of the user’s computer, interpret various UI elements, and decide the course of action. To optimize computer interactions at enterprise scale, Anthropic announced the acquisition of Vercept in February 2026. This AI startup specializes in tools designed for complex agentic tasks, exemplified by its cloud-based Vy product, which remotely operates Apple MacBooks.

The specifics of the acquisition financials remain undisclosed. However, Vercept’s founders and some employees have joined Anthropic as part of the deal. “It’s been less than four weeks since we joined, and with the team here behind us and joining forces, we just shipped our first product launch,” Kiana Ehsani, a co-founder, mentioned in a post on X. “Four weeks down, many more to come. We’re just getting warmed up,” she added.

While Anthropic’s LLMs rank among the most advanced models available, internal reasoning alone proved to be insufficient for enhancing the dynamism of AI agents. They require comprehensive understanding through grounded interactions in dynamic environments. Vercept’s focus on bridging this gap contributed to its appeal as an acquisition target for Anthropic.

This acquisition followed Anthropic’s earlier purchase of Bun, a startup that developed tools for executing and managing AI agents in production settings. The integration of these layers is crucial for Claude’s transition from merely a conversational model to a robust execution platform.

Is Claude the End of OpenClaw?

The announcement from Anthropic can be interpreted as a strategic response to the increasing prominence of OpenClaw. This open-source platform has generated a rich ecosystem of autonomous AI agents or ‘claws,’ capable of spawning their own sub-agents to autonomously carry out specialized tasks with access to local files and data.

Initial implementations of OpenClaw, often set up on economical Mac Minis, primarily focused on running Claude Code through WhatsApp on users’ phones. Thanks to its open-source format, projects by OpenClaw developers have evolved to facilitate AI agents performing functions like booking flights, managing smart home devices, and overseeing social media campaigns across various platforms and devices in a structured approach.

Even though setting up OpenClaw may present a more makeshift and experimental feel, Anthropic’s newly introduced computer-use capability offers a potentially more reliable, integrated solution that demands relatively minimal technical knowledge. This update, anticipated to support Windows devices shortly, could alleviate the requirement for self-hosting on Mac Minis, which have reportedly seen high demand in China during the OpenClaw surge.

As OpenClaw continues to capture the attention of developers, companies, and regulators, concerns over security are growing. The requirement for OpenClaw to access a user’s comprehensive data and systems poses risks of agents going rogue and mistakenly altering or deleting critical files. These security concerns have spurred the emergence of safer forks and alternatives. Recently, Nvidia unveiled NemoClaw, a software toolkit designed for easily configuring and installing OpenClaw with several enhanced security measures.

However, it is crucial to note that Claude’s computer-use feature introduces its own set of security considerations. Anthropic has stated that safeguards are in place against threats such as prompt injection attacks, but has also cautioned users that this feature remains new. Therefore, Claude should not be granted access to applications managing sensitive data.

Karan Mahadik specializes in technology’s intersection with public policy for The Indian Express, based in Delhi-NCR.