Enabling Claude Code to work more autonomously

Source: Anthropic Date: 2025-09-29 URL: https://www.anthropic.com/news/enabling-claude-code-to-work-more-autonomously

Summary

Anthropic launched a set of Claude Code enhancements for autonomous development: native VS Code extension (beta), refreshed terminal interface with searchable history, and a checkpoint system for reverting code states. Default model switched to Sonnet 4.5. Introduced subagents, hooks, and background tasks for parallel workflows. The Claude Agent SDK (renamed from Claude Code SDK) now supports custom agents for specialized use cases like financial compliance and cybersecurity.

Implications

• Claude Code / agentic engineering thread. VS Code native extension is the distribution move that matters most here — meeting developers in their existing environment rather than requiring terminal context switching. Native IDE presence is how Copilot won; Anthropic is targeting the same real estate.
• Checkpoints for code revert. A checkpoint system that lets users revert Claude Code changes is a safety feature for autonomous operations — it’s the “undo” button for agentic coding, which addresses the primary fear developers have about letting AI make filesystem changes.
• Subagents + hooks. Parallel subagent workflows are the architectural feature that enables complex multi-step engineering tasks. This is the foundation for the “agent teams” that appear in Opus 4.6’s announcement months later.
• Agent SDK rename. Renaming Claude Code SDK to Claude Agent SDK signals Anthropic is generalizing the product beyond coding to all agent use cases — financial compliance and cybersecurity explicitly named as targets.
• Watch: VS Code extension adoption vs. Cursor/Windsurf; whether checkpoints reduce agent-caused incident rates; how the Agent SDK ecosystem grows beyond coding agents.