CHAPTERS
Talk framing: developer experience + autonomy + where to keep up
Ralph (Anthropic technical staff) sets expectations: many recent Claude Code releases, but the talk will focus on a curated subset. He organizes updates into two themes—developer experience improvements and autonomy features—then promises pointers to ongoing release info.
Remote Control: keep a coding session running while you’re away
Remote Control lets you start a Claude Code session on your computer and continue interacting from the Claude mobile app or any browser. It’s positioned as a way to dispatch work and only check in when Claude needs input, with notifications to pull you back in.
Remote Control demo: activate, rename sessions, and continue prompts anywhere
Ralph demonstrates enabling Remote Control with a command, issuing prompts, and taking over the session from another device. He shows renaming the session so it’s easier to identify on mobile and continuing the back-and-forth seamlessly.
Flicker-free terminal: full-screen TUI with virtualized scrollback
Claude Code introduces a new full-screen mode that virtualizes terminal scrollback to prevent flicker and keep memory usage flat, even in long sessions. This also enables richer terminal interactions like clickable elements and easier navigation.
Full-screen TUI demo: clickable expansions and “jump to end” navigation
Using the /tui fullscreen command, Ralph shows Claude Code reloading into a full-screen interface. He demonstrates expanding generated content via clickable elements and using a “go to end/new message” control to jump to the latest output.
Claude Code Desktop refresh: sessions list, project grouping, and workflow UI
The desktop app has been revamped, and Ralph encourages CLI-first users to revisit it for certain tasks. The new UI emphasizes session management, especially grouping sessions by repository/project for people running parallel work.
Desktop app demo: plan view, inline comments, diffs, and file browsing
Ralph walks through desktop features: viewing the plan behind a session, commenting on parts of the plan to request changes, and reviewing code via diffs. He also shows asking Claude to explain specific lines and browsing project files from within the app.
Why autonomy features matter: fewer blockers in long-running sessions
Ralph motivates autonomy improvements with a common frustration: Claude getting stuck on permissions or minor decisions mid-run. The next features aim to let Claude proceed safely without interrupting the developer unless necessary.
Auto Mode: smarter permissions with destructive-action and injection checks
Auto Mode reduces authorization interruptions by using a classifier to decide whether an action is safe. It evaluates whether an action is destructive and whether the situation resembles prompt injection; if safe, Claude proceeds, otherwise it seeks alternatives or asks you.
Worktrees: parallel Claude sessions without stepping on each other
Worktrees let multiple Claude sessions work on the same repo safely by isolating each session into its own working copy directory. Claude Code now offers simpler, native worktree support so users don’t have to manage the complexity manually.
Auto-memory: Claude writes durable project notes without bloating context
Auto-memory addresses the ‘blank slate’ feeling in new sessions by having Claude capture durable insights—like style choices, architecture decisions, and debugging learnings—into a memory.md system. It’s designed for progressive loading so context stays lean, and stays local to your machine.
Automated PR code review: multi-agent, multi-phase ‘Ultra Review’
Anthropic’s internal practice becomes a product capability: automated PR review that spins up multiple agents to inspect different aspects (bugs, vulnerabilities, logic errors). A second phase cross-checks findings against the actual code to improve precision and speed.
Routines (research preview): scheduled and event-driven Claude Code sessions in the cloud
Routines let you run Claude Code sessions automatically—on schedules, webhooks, or API triggers—without manually starting them. They run in a remote environment (cloud), support the same capabilities as a normal session, and integrate well with GitHub events like issues and PRs.
Routines demo: issue triage + PR review personas triggered by GitHub webhooks
Ralph demonstrates creating routines from the desktop app: one scheduled to check repo changes, another triggered when an issue is opened, and a PR-review routine that comments directly on pull requests. He shows how a newly created GitHub issue triggers the automated ‘medieval knight’ triage response.
Agents View (public preview): one dashboard for many parallel sessions
Agents View provides a centralized interface to manage multiple Claude Code sessions running in the background. It shows status (working, complete, waiting for input) and enables quick interaction—either by entering a session or sending prompts without switching context fully.
Enterprise & staying current: deployment improvements and update channels
Ralph closes by crediting user feedback and highlighting enterprise-focused improvements like better Windows support and cloud deployment wizards for AWS/GCP, plus native binaries for governance pipelines. He points viewers to official channels to keep up with frequent releases.
