Claude Code and Cowork can now use your computer
apple claude
| Source: Mastodon | Original article
Anthropic has lifted a key restriction on its AI assistants: Claude Code and the newer Claude Cowork can now act directly on a user’s computer. In a brief announcement posted to the company’s help centre, the firm said the tools run locally, letting the model point, click and edit files the way a human would. Users grant access to specific folders, and all code execution happens inside an isolated sandbox, but the model can now open applications, drag‑and‑drop data, and commit changes without the user typing a single line.
The move builds on the capabilities we covered earlier this month, when we explored Claude Code’s role in pull‑request reviews and productivity hacks. Those stories showed the model’s strength in understanding and generating code, but the workflow still required the developer to copy‑paste snippets or run commands manually. By giving Claude a “virtual hand” on the desktop, Anthropic turns a conversational code assistant into a true co‑pilot that can, for example, refactor a repository, update configuration files, or generate a playlist in Spotify without leaving the chat window.
The significance is twofold. For developers, the integration promises to shave minutes—or even hours—off repetitive tasks, making AI‑driven automation feel more immediate and less abstract. For the broader AI market, it narrows the gap between large‑language‑model assistants and the tightly integrated agents offered by Microsoft and Google, raising the stakes for safety and privacy. Anthropic’s sandboxed execution and explicit file‑sharing consent aim to mitigate the risk of unintended changes, but the ability to control a user’s UI also opens new vectors for abuse if mis‑configured.
What to watch next: Anthropic has not disclosed a full rollout schedule, but early adopters can enable the feature through the Claude Help Center today. Expect tighter OS support (macOS, Windows, Linux) in the coming weeks, pricing details for enterprise‑grade usage, and a wave of third‑party plugins that expose more apps to the model. Competitors are likely to accelerate their own desktop‑agent roadmaps, and regulators may soon scrutinise how much control users cede to AI. The coming months will reveal whether Claude’s newfound hands‑on ability translates into measurable productivity gains or sparks fresh security debates.
Sources
Back to AIPULSEN