Anthropic launches production browser tool and Chrome side panel for Claude

Browser Use reaches general availability with a structured, non-visual model

Anthropic's Browser Use tool is now generally available, giving Claude a structured view of a web page in addition to what is visually rendered. One explainer notes the tool "doesn't actually run a browser," instead exposing elements that Claude can target through references rather than raw screenshots. A second write-up frames the same release as part of a broader package that brings Computer Use, browser access, custom Skills, and larger file storage to enterprise AI agents, which gives developers a single surface for browser automation, on-screen action, and file handling.

For practitioners, the practical shift is that page interaction moves from pixel-level perception to structured element references, which changes how prompts, error handling, and test fixtures need to be written.

Claude moves into Chrome via a Cowork-branded side panel

The Claude in Chrome side panel has been folded into Claude Cowork, turning browser sessions into full Cowork sessions that share state with the Claude apps. According to Anthropic's own post, "Your conversations, skills, and connectors carry between the browser and the Claude apps," which removes the line between a browser tab and the agent surface developers were already running locally. A social post from a developer describes the same launch as "Claude inside Chrome," listing automation tasks such as web research and data collection.

For working developers, this means Skills and connectors defined in one environment no longer need to be re-created when switching context to a browser tab.

What this changes in an enterprise agent build

The combined release groups Browser Use, Computer Use, Skills, and expanded file storage as one production-ready agent stack rather than separate beta surfaces. Channel coverage frames the bundle as "production-ready Computer use, browser access, custom Skills, and larger file storage for enterprise AI agents," which matters because each capability previously had its own access story. Consolidating them under a generally available label reduces the version-skew problem teams hit when one capability was GA and another was preview, and it gives platform owners a single set of controls to audit.

How to think about the new browser model

Because Browser Use works through structured element references rather than rendered pixels, prompts that used to assume visual reasoning will need to be rewritten. The New Stack write-up makes this explicit: the tool "gives Claude a structured view of a web page in addition to what is visually rendered," so developers can opt into the cheaper and more deterministic path while still falling back to vision when a page lacks semantic markup. Teams that built selectors or DOM snapshots for older Claude releases will need to verify their assumptions against the structured-reference model before rolling out broadly.

Follow-up for working developers

A reasonable next step is to audit any in-house agent that already scripts a browser, confirm which Anthropic capability it relies on, and test it against the GA Browser Use release to surface breakages from the structured-reference change. Teams that standardise on Skills and connectors should re-validate that state carries correctly between the Chrome side panel and the Claude apps, since the Cowork integration is the new shared surface. Where a workflow depends on page interaction, building small local probes — such as a MIME Type Lookup or a URL Extractor — can help characterise what the structured view actually exposes before rewriting production prompts.

For organisations still deciding policy, pending EU AI Act deadlines around watermarking make the broader governance picture worth tracking as agent surfaces expand.