Anthropic has taken a significant step towards fully autonomous AI agents. The company has updated Claude with the ability to take direct control over the user's computer — and perform tasks independently, even while the user is away from the screen.
Claude Does the Work Itself
The new capabilities are built into the Claude Code and Cowork tools, allowing the AI model to operate on the machine without the user needing to follow every step. According to Anthropic's own announcement, Claude can, among other things, open and process files, navigate web browsers and apps, and activate developer tools — all without prior setup.
The user is prompted to grant permission before Claude acts autonomously, which is intended to ensure that humans still have the final say. Nevertheless, the functionality represents a clear shift: from AI as an assistant to AI as an agent.
The feature is currently available as a research preview, reports The Verge.
From AI as an assistant to AI as an agent — Anthropic crosses an important boundary with Claude Code and Cowork.
A Growing Trend: AI Agents Take Over Work Tasks
Anthropic's launch is not an isolated case. It is part of a rapid development where autonomous AI agents are increasingly taking over tasks that previously required human intervention. According to background research on AI and the future of work, such systems — which can interpret user intent, interact with software via APIs, and perform actions across entire workflows — are fundamentally reshaping the world of work.
Who Will Be Hit Hardest?
It is particularly within technology, finance, law, and consulting that the consequences could be greatest in the short term. Anthropic's own CEO, Dario Amodei, has previously stated that nearly half of all entry-level positions in these sectors could be replaced or eliminated by AI — a claim that should naturally be read in light of the fact that he himself leads one of the leading AI companies and has commercial interests in the field.
Research reports point in the same direction, albeit with varying estimates. A report from Goldman Sachs estimates that AI automation could affect up to 300 million full-time jobs globally, while McKinsey Global Institute predicts that AI technology could add around $13 trillion to the global economy by 2030.
Trust and Control: An Unresolved Equation
The question of who truly has control is central. Anthropic stipulates that the user must grant explicit permission, but as AI agents become faster and more capable, the line between «assistant who asks» and «agent who acts» will become increasingly blurred.
It is also worth noting that the features are currently only available in a research preview. This means that Anthropic is not presenting this as a finished product, and that behavior, limitations, and security mechanisms may still change.
For businesses and individual users considering adopting such tools, questions of data security, privacy, and liability will be crucial — especially as AI agents gain access to sensitive files and systems.