Claude Code Field Guide
17 / 35

Extended Thinking

Claude's ability to reason through a hard problem step by step before acting — on by default, no magic words required.

Updated since the video
Currency check: Accurate, but the “magic words” era it describes (typing “think hard”) is explicitly obsolete, and effort levels are now the primary dial.

Some tasks are trivial (“rename this file”); others genuinely need Claude to plan and reason through multiple steps before touching anything. Extended thinking gives Claude a dedicated reasoning budget in tokens for that planning. The video correctly notes you used to have to invoke this with words like “think,” “think hard,” or “ultrathink” — that's gone. It's on by default with a built-in cap.

What's changed since the video: the primary way to control thinking depth now is the /effort command (see Models) rather than magic phrases or toggling thinking on/off directly. You can still view the reasoning itself with Ctrl+O (it's collapsed by default), and it's worth knowing thinking tokens are billed as output tokens whether you're looking at them or not — relevant if you're watching cost.

Planning a multi-step migration

Before touching a database schema migration that affects a dozen files, extended thinking lets Claude Code reason through the dependency order and edge cases up front, rather than discovering a broken foreign key mid-edit.

Turning thinking down for a simple bulk edit

For a purely mechanical task (updating a copyright year across every file), a user lowers effort rather than leaving thinking at its default depth, since no real reasoning is needed and the token cost isn't worth it.