Claude Code's 7 Quiet Updates: From Cross-Session Messaging to Built-In Browser

Eric Tech's 'Don't Miss These 7 Claude Code Updates' gathers seven quietly shipped Claude Code changes into one bench test. The list spans the end of May's temporary quota bump to a new design-canvas integration, each shown with a short live demo. The honest Codex comparison at the close is what lifts the video above a simple feature count.

Sessions can now message each other

The first update lets separate Claude Code sessions message one another directly. Previously a frontend and a backend session stayed in sync only through you; now one session can push a concise summary to the other and changes sync automatically. You list every active main and sub-agent with /list agents, address by @name, and the internal SendMessage tool forwards a compressed summary — not the full conversation history — so the context window stays clean. Naming is flexible: start with claude --name or rename in place with /rename, demonstrated by naming a session 'campaign creator' and passing a Google Sheets ID in a single cross-session note.

Weekly limits trimmed, auto-continue added

The second note is weekly usage limits. May's temporary 50% boost for Pro, Max and Team plans stepped down to 25% on September 14; the five-hour window is untouched but the weekly budget tightens noticeably. What softens the cut is 'auto-continue when limits reset' — toggled in /config in the CLI or in the desktop app — which pauses work at the cap and resumes automatically at reset. The host's example shows a job stuck at 218 auto-waking at 340 without another click.

Built-in browser: Claude opens the web itself

The third change, likely the most talked about, is a built-in browser in the Claude Code desktop app. Once enabled in the Browser tab, the agent can open docs, designs or admin panels, click through, log in, and spin up a local dev server to verify changes with screenshots and DOM inspection. The sensitive part is cookie handling — you choose to reset after tab close, keep per session, or share across tabs, and you gate which sites the agent may visit. You can import Chrome cookies, though the host rightly advises sharing only a narrow set. The live demo navigating to OpenAI's news page to summarize fresh posts reads as a direct answer to Codex's computer use.

iOS simulator and auto mode by default

Fourth is the iOS simulator — public beta for now, Pro/Max/Team only. The desktop app shows code on the left and a live iOS Simulator on the right, so write and test happen in one view when Xcode and the iOS platform are installed. The fifth shift is auto mode becoming the default. Before running a terminal command the agent runs it through a classifier: if allowed it proceeds automatically, if risky it asks you. Anthropic says the classifier adds no extra model cost and aims to balance uninterrupted flow against avoiding harmful actions.

The sixth tweak is output style: /config now offers a 'concise' style that leads with results, skips the preamble and shortens replies. In the demo the same 'what's in this folder' prompt costs 58k tokens in concise versus 66k in default — roughly a 12% saving and visibly less reading load. The seventh and final item is /design, a skill that brings design artboards into the CLI. You prompt with /design, the agent triggers an artifact skill and publishes an editable canvas link; the camera-app example cycles through arcade and other styles, lets you edit text Figma-style, add sticky notes and comments, and push changes back to code in one go.

Codex or Claude Code? The host's scale

What separates the video from a routine tour is the closing stance. The host says the last two months tipped toward Codex, crediting its computer use, more polished desktop experience and thread-per-conversation/worktree management. He still sees Claude Code as terminal-centric and behind on desktop polish. Anthropic's latest moves he reads as borrowing demand Codex already proved — the in-app browser included. The question is clean: single-session speed or multi-session bench — the answer depends on the job.

Together the seven pieces describe a bench maturing more than a model leaping. Messaging sessions close parallel work, the browser closes verification, the iOS pane closes the mobile loop, concise style trims token economics. The weekly quota step-down means that bench will cost more to run; auto-continue only removes wait friction, not the cap. As the host warns, convenience toggles like cookie sharing and auto-allow deserve narrow allow-lists.

A final note: the video briefly pitches the host's own community, interrupting an otherwise product narrative. Most features remain in public beta or staged rollout, so not every account sees them the same week; weekly-limit math and the 'research preview' tag on /design may shift. Still, all seven items trace to Anthropic announcements and independent coverage — so the video is valuable less as a launch and more as a single-screen compilation of scattered releases.