📰 Claude Code & Cowork: Autonomous Computer Control Now Live (2026) Anthropic’s Claude Code and
agents anthropic autonomous claude
| Source: Mastodon | Original article
Anthropic has lifted the final barrier to truly autonomous AI assistants: Claude Code and its consumer‑friendly sibling Claude Cowork can now take direct control of a user’s computer. The update, announced on March 24, lets the models move the mouse, type on the keyboard, open files, browse the web and launch development tools without any prior configuration or scripting. The agents decide which actions are needed to fulfil a request, execute them in real time, and report back with results or follow‑up questions.
The breakthrough builds on the desktop‑automation demos we covered earlier this week, when Claude was first shown controlling a Mac via Discord and a custom UI (see our March 24 “Claude Can Control Your Mac” report). Those prototypes required a manual “hand‑over” step; the new release eliminates that friction, turning Claude into a self‑sufficient worker that can, for example, pull data from a spreadsheet, draft a report in a word processor, or debug code in an IDE without a human clicking each button.
Why it matters is twofold. First, it narrows the gap between large‑language‑model assistants and the “general‑purpose agents” that tech giants have been racing to build, potentially reshaping how developers and knowledge workers automate repetitive tasks. Second, the ability to act on a physical desktop raises immediate security and privacy concerns: any compromised prompt could trigger unwanted file modifications, credential theft or ransomware‑like behavior. Anthropic’s documentation stresses sandboxed execution and user‑approved permission scopes, but the shift will likely prompt tighter OS‑level controls and new enterprise policies.
What to watch next are the rollout mechanics and ecosystem response. Anthropic plans a phased release, starting with a beta for enterprise customers, while third‑party tools such as the open‑source Outworked UI are already being adapted to expose the new capabilities. Analysts will be tracking whether competitors like Google DeepMind or Microsoft Copilot accelerate their own autonomous‑agent roadmaps, and how regulators respond to the expanded attack surface introduced by AI‑driven desktop control.
Sources
Back to AIPULSEN