Enabling Claude Code to work more autonomously (www.anthropic.com)

Anthropic rolled out a major productivity upgrade for Claude Code, introducing a native VS Code extension (beta), a refreshed terminal interface (v2), checkpoints for safe autonomous edits, and an expanded Claude Agent SDK — all running on Sonnet 4.5. The VS Code extension surfaces Claude’s edits in real time with a sidebar and inline diffs; the terminal update adds clearer status, searchable prompt history (Ctrl+R), and visual improvements. Sonnet 4.5 is now the default model (switchable with /model), the extension is in the VS Code Marketplace, and terminal/SDK updates are available now. Technically the launch is focused on enabling longer, parallel, and more trustworthy agentic workflows. A checkpoint system automatically snapshots Claude’s edits so you can rewind (Esc twice or /rewind) to restore code, conversation, or both — note checkpoints track Claude edits, not user edits or shell commands (use version control alongside them). The Claude Agent SDK adds subagents, hooks, and background tasks so teams can split work (e.g., backend subagent while frontend is built), auto-run tests/linting at hooks, and keep dev servers alive without blocking progress. Early adopters are using the SDK for compliance, cybersecurity, and debugging agents, signaling broader utility for building robust, autonomous developer assistants.