Plan mode makes the agent research and propose an approach before it edits anything. You approve or redirect, then it executes. It’s the single highest-leverage habit for multi-step work.

Why it matters

A wrong approach discovered after the agent has edited ten files is expensive to unwind. Caught at the plan stage, it costs one message. Planning also produces better work: the agent has thought through the whole task before touching code, so the edits are more coherent.

When to use it

Anything touching more than one or two files.
Anything you’re unsure how to scope.
Anything where a wrong approach is costly to reverse (migrations, refactors, schema changes).

When to skip it

Tiny, obvious edits where planning is slower than just doing the change. Plan mode is a tool, not a tax — reserve it for work where the plan is worth reviewing.

How to improve

If most of your prompts run in auto mode, try starting your next substantial task in plan mode. Review the plan, tighten it, then let it run uninterrupted — which also cuts interruptions.