Claude Code on Desktop Gets a Built-In Browser Pane
Anthropic highlighted a new in-app browser for Claude Code on desktop. Claude can now pull up docs, designs, or any site and read it, click through, and interact the same way it already does with your local dev servers — no context-switch to Chrome. Launch it with Cmd/Ctrl + Shift + B, and external links shared in a conversation can open in the pane or your default browser.
The Browser pane runs a clean profile with none of your saved logins or history, but it does support website logins — including Google OAuth pop-ups — so Claude can test authenticated apps. Because Claude is now acting on live sites, the first time it tries to act on a given site you’re asked to Allow once, Always allow, or Deny. For identity-bound work, the separate Claude in Chrome extension still shares your real browser session.
API Key Expiration Lands in the Console; Opus 4.7 Fast Mode Set to Sunset
The Claude Developer Platform added API key expiration settings in the Console. You can now set an expiration when creating an API key or Admin API key, and for any key with a lifetime of at least seven days, Anthropic emails the creator before it expires, with admin reporting to track keys across an org — a small but real hygiene win for teams juggling long-lived credentials.
On the deprecation side, fast mode for Claude Opus 4.7 is being retired, with removal on July 24; after that, requests to claude-opus-4-7 with speed: "fast" will error. It follows the recent rate-limit overhaul that raised Sonnet and Haiku limits to match Opus at every tier and consolidated pricing tiers into three: Start, Build, and Scale.
VS Code 1.128 Ships Multi-Chat Claude Agent Sessions
Visual Studio Code 1.128 (released July 8) makes its headline feature multi-chat support for Claude agent sessions. Instead of spreading related threads across separate top-level sessions, a single session can now hold multiple related chats — letting developers compare approaches, branch from an earlier turn, and run work in parallel. Each chat keeps its own history, title, and model selection, and all of them restore with the parent session after a restart.
The release also adds quick chats in the Agents window for questions not tied to a folder; these workspace-less chats open ready for input but skip the Changes and Files side panes. Copilot Vision reached general availability in Chat in the same build — a reminder that Claude now shares the editor with a widening set of agent features.
Claude Code Turns On Auto Mode by Default Across Bedrock, Vertex AI, and Foundry
The latest Claude Code release makes auto mode the default on Amazon Bedrock, Google Vertex AI, and Microsoft Foundry, bringing the hands-off agent workflow to the major cloud deployment targets rather than just the first-party API. The update also moves Bedrock to Claude Opus 4.8 and tightens settings handling across the board.
Under the hood it’s a stability release: improvements to the agent view and Remote Control, plus fixes for freezes, crashes, credential issues, and several worktree and plugin bugs. It’s the kind of cross-cloud parity work that matters for teams standardizing on a single agent workflow no matter which provider hosts the model.
Anthropic Now Worth More Than OpenAI — and Both Have Filed for an IPO
The market backdrop got dramatic. As of early July, reporting has Anthropic valued above OpenAI for the first time in its history, with Claude holding roughly 17% of the US mobile chatbot market. And in a striking bit of timing, both companies confidentially filed for an IPO within a week of each other.
It reframes the shipping cadence of the past two weeks. The steady drumbeat of enterprise governance, cross-cloud tooling, and always-on agents isn’t just product polish — it’s the recurring-revenue, durable-infrastructure story a company tells ahead of public-market scrutiny. Two of the biggest names in AI heading for the exits at once will put real numbers behind the hype.
The Browser Just Became a Feature Inside the Agent
Today’s standout isn’t a model — it’s a browser living inside Claude Code. For years the browser was the thing you tabbed over to; now it’s a pane the agent drives, with its own clean profile, permission prompts, and login support for testing authenticated apps. Pair that with multi-chat sessions in VS Code and auto mode defaulting on across every major cloud, and the shape of the week is clear: the agent is absorbing the developer’s tools, one surface at a time.
That’s the quiet through-line behind the louder headline that Anthropic now out-values OpenAI and both are filing to go public. Investors don’t underwrite demos; they underwrite workflows people can’t give up. A browser you don’t leave, sessions you don’t lose, and agents that run themselves on whatever cloud you already pay for — that’s the stickiness a prospectus is built on. The exciting phase was watching agents get capable; the durable phase is watching them quietly eat the tools around them.