Plan Mode and Autonomous Workflows

Claude Code can run anywhere on a spectrum from "ask before every keystroke" to "keep working unattended, turn after turn, toward a condition you set once." Neither extreme is automatically correct — the right point on that spectrum depends on how expensive a wrong guess is, how well-scoped the task is, and how easy the result is to verify. This subject covers the mechanisms Claude Code gives you to choose that point deliberately within a single session: plan mode, which separates deciding what to do from doing it, and lets you review and iterate on the plan itself before anything executes; and autonomous loops and goals, which keep an agent working across turns without you driving each one, either on a time interval or toward a verifiable end state.

The goal isn't to memorize keybindings. It's to understand why each mechanism reduces risk or increases throughput, so you can reach for the right one instead of defaulting to whichever one you used last. Once a task outgrows what one conversation can hold — dozens of independent items, or findings that need cross-checking before a human sees them — the right move is a different mechanism entirely, covered in its own subject; this one closes with a short pointer to that material.