CHAPTERS
- 0:16 – 1:52
Talk setup: Rapid Claude Code releases and today’s focus areas
Ralph introduces himself and explains that Claude Code has shipped many updates recently, so the talk curates a subset. He frames the session around two themes—developer experience and autonomy—and promises pointers for staying current.
- •Speaker intro and context from Anthropic partner/customer work
- •Claude Code has had a dense changelog over recent months
- •Two pillars for the talk: developer experience and autonomy
- •Will end with where to find more updates and details
- 1:52 – 2:53
Remote Control: Run a coding session on your computer, continue from phone/browser
Remote Control lets you start Claude Code on your machine and interact with the same session from the Claude mobile app or any browser. It’s pitched as a way to keep work moving during errands or breaks, with notifications when Claude needs input.
- •Start a session locally and control it remotely from mobile/web
- •Useful for long-running tasks that occasionally need prompts
- •Notifications alert you when Claude needs input
- •Encouragement to replace “doom scrolling” with quick check-ins
- 2:53 – 4:54
Remote Control demo + pro tip: Naming sessions and enabling by default
Ralph demonstrates enabling Remote Control via a command, renaming a session automatically based on activity, and continuing the same conversation across devices. He notes you can configure settings so every session is remote-accessible by default.
- •Enable Remote Control with a simple command
- •Use session renaming for clarity across devices
- •Seamless back-and-forth prompts from desktop and remote UI
- •Settings JSON can default Remote Control to ‘always on’
- 4:54 – 6:25
New terminal full-screen (TUI) mode: Flicker-free Claude Code rendering
Claude Code now offers a full-screen terminal UI that eliminates flickering by virtualizing terminal scrollback. This improves performance in long sessions and enables new interactive terminal elements like clickable expansions.
- •Prior terminal repaint behavior caused flickering on large scrollback
- •Full-screen mode virtualizes scrollback and renders only visible parts
- •Memory usage stays flat even in long sessions
- •Enables clickable UI elements inside the terminal (expandable text/buttons)
- 6:25 – 8:27
TUI demo: Click-to-expand outputs and jump-to-latest navigation
Ralph demos the `/tui` command with a full-screen parameter, showing expandable content blocks and a “jump to end/new message” control. He notes the full-screen UI can also be enabled by default for all new sessions.
- •Enter full-screen via `/tui` with the full-screen option
- •Clickable expansion reveals full generated content (e.g., file contents)
- •UI indicator for new messages and quick jump to the bottom
- •Option to make full-screen TUI the default for new sessions
- 8:27 – 9:28
Claude Code Desktop refresh: Why it’s worth revisiting
The desktop app experience has been overhauled, especially for users who previously preferred the CLI. Ralph positions desktop as ideal for certain workflows rather than replacing terminal use entirely.
- •Desktop app received a major UI/UX revamp
- •Aimed at specific tasks where desktop beats CLI
- •Encouragement for CLI-first users to try it again
- •Preview of richer navigation and review tooling
- 9:28 – 10:59
Desktop demo: Organize sessions by project, inspect plans, and comment inline
Ralph tours the desktop UI: session lists, grouping by repository/project, and viewing “plan mode” outputs. Users can highlight portions of a plan and leave comments that get sent back to Claude for adjustment.
- •Left sidebar lists sessions with grouping options
- •Project-based grouping ties sessions to repos
- •Plan viewer makes step-by-step plans easy to revisit
- •Inline comments on selected plan text feed back into Claude changes
- 10:59 – 12:30
Desktop demo continued: Diffs, file browsing, and GitHub integration
The desktop app supports reviewing diffs and files directly, adding comments, and asking Claude to explain specific lines. Ralph highlights built-in GitHub integration for review and iteration workflows.
- •View PR diffs within the desktop app
- •Add comments directly on code/diff context
- •Ask Claude to explain a specific line from the diff
- •Browse project files and leverage GitHub integration from the UI
- 12:30 – 13:31
Autonomy theme: Reduce unnecessary interruptions in long-running sessions
Ralph transitions to features that let Claude proceed without stopping for trivial permissions or prompts. The goal is to support more asynchronous workflows where Claude only asks for input when truly needed.
- •Common frustration: Claude gets stuck on prompts/authorization
- •Models are better at judging when human input is necessary
- •Focus on letting Claude continue safely on its own
- •Connects to asynchronous coding workflows
- 13:31 – 15:32
Auto mode: Permission decisions with safety checks (destructive actions + prompt injection)
Auto mode reduces authorization interruptions by letting Claude proceed when actions are deemed safe. A classifier evaluates whether an action is destructive or resembles prompt injection; Claude continues, finds alternatives, or escalates to the user accordingly.
- •Auto mode minimizes “can I read/access?” approval prompts
- •Classifier checks: (1) destructiveness risk (2) prompt injection signals
- •If safe, Claude proceeds automatically; if risky, it seeks alternatives
- •Escalates to user only when necessary—Claude must ‘earn’ interruptions
- 15:32 – 17:34
Worktrees: Parallel Claude sessions without conflicts (now easier with native support)
Worktrees let multiple Claude sessions work in parallel on the same repo without stepping on each other’s files. Claude Code now supports creating and managing worktrees more natively, reducing manual directory management.
- •Worktrees create segregated repo copies for isolated work
- •Enables multiple Claudes to run in parallel safely
- •New native support simplifies creation (`claude --worktree`)
- •Desktop and other UIs expose a “start in new worktree” option
- 17:34 – 20:05
Auto-memory: Claude writes durable project notes into memory.md (local, auditable, lightweight)
Auto-memory addresses the “blank slate” feeling across sessions by letting Claude save important project-specific learnings automatically. It writes concise notes and pointers into `memory.md`, loaded alongside `Claude.md`, with progressive discovery to avoid context bloat.
- •Claude captures recurring preferences: style, architecture, debugging insights
- •Writes to `memory.md` selectively—only what’s worth remembering
- •Difference: `Claude.md` = onboarding/manual; `memory.md` = ongoing notes
- •Progressive discovery via an index reduces context bloat
- •Memory stays local (not pushed to GitHub/cloud); inspect via `/memory`
- 20:05 – 22:08
Automated PR review: Multi-agent, multi-phase ‘Ultra Review’ like Anthropic teams use
Claude can perform structured code reviews for pull requests using multiple agents focused on different risk areas, then validate findings against the code. It can run automatically via the GitHub app or manually via `/ultra-review` in Claude Code.
- •Multi-agent reviewers check bugs, vulnerabilities, logic errors, etc.
- •Second phase validates findings to reduce false positives
- •Runs automatically on new PRs with GitHub integration
- •Manual trigger available via `/ultra-review`
- 22:08 – 23:08
Routines (research preview): Scheduled/event-driven Claude Code sessions in the cloud
Routines let you trigger Claude Code sessions automatically on schedules, webhooks, or API calls—without manually starting them and without your computer needing to be on. Ralph positions this as a foundation for asynchronous automation across workflows, including GitHub-driven tasks.
- •Create automated workflows triggered by timers, webhooks, or API calls
- •Runs in a remote environment (cloud) even when your machine is off
- •Supports the same capabilities as interactive Claude Code sessions
- •Works natively with GitHub events (issues/PRs) and can call external systems
- 23:08 – 27:40
Routines demo: Repo monitoring, issue triage, and PR reviews with personality prompts
Ralph demos routines in the desktop app: scheduled checks for commits/issues, an issue-open triage routine, and a pull-request review routine that comments back on GitHub. He shows how webhooks trigger these routines automatically and how logs accumulate over time.
- •Desktop ‘Routines’ UI shows configured routines and run history
- •Example schedule: twice-daily checks for new commits/issues
- •Create an ‘issue open’ routine that generates a triage/report response
- •PR routine posts review comments directly to the PR via GitHub webhook
- •Routine logs persist, enabling later review of outcomes
- 27:40 – 30:12
Agents View (public preview): Monitor and manage many parallel sessions in one place
Agents View provides a single dashboard to track multiple Claude Code sessions, their states, and whether they need input. Ralph demos launching it and interacting with sessions in the background, including sending prompts without fully entering a session.
- •Single view of all sessions and statuses (running, waiting, completed)
- •Sessions continue in the background—less terminal babysitting
- •Quick actions: enter a session or ‘peek/send’ prompts from the list
- •Designed for heavy parallel-session users
- 30:12 – 32:03
Wrap-up: Feedback, enterprise-focused improvements, and where to follow updates
Ralph credits user feedback for rapid iteration and asks the audience to keep reporting bugs and requesting features. He briefly notes enterprise improvements (Windows support, cloud deployment wizards, native binaries) and points to official channels for staying up to date.
- •Call to action: keep using Claude Code and sharing feedback
- •Enterprise/team features: better Windows support, cloud deployment wizards (GCP/AWS)
- •Native binaries for governance/controlled install pipelines
- •Where to track updates: social accounts, docs ‘What’s new,’ changelog, newsletter
