According to @bcherny, custom agents now support running subagents in dedicated Git worktrees by adding “isolation: worktree” in the agent frontmatter, enabling cleaner concurrent experiments and safer code generation workflows (as reported by the original tweet on X). For AI engineering teams, this supports sandboxed agent tasks, parallel feature development, and lower merge conflicts in agent-driven repos, according to the post by Boris Cherny. As reported by the tweet, the configuration targets multi-agent setups where LLM-powered subagents handle discrete branches, improving reproducibility, CI isolation, and rollback hygiene for AI-assisted coding pipelines.

According to @bcherny, developers can enable worktree mode in the Claude Code Desktop app by opening the Code tab and checking worktree mode, which streamlines multi-branch development without using the terminal. As reported by Anthropic’s official docs, Claude Code Desktop supports a Desktop Quickstart with instructions at code.claude.com/docs/en/desktop-quickstart#install, enabling safer branch isolation and faster context switching for AI-assisted coding. According to Anthropic, this setup reduces friction when using Claude for code generation, refactoring, and reviews across parallel branches, improving throughput for teams adopting AI pair programming.