Claude Code Spinner Verbs Explained: What “Pondering” and “Brewing” Mean, How to Change Them, and Why You Should Stop Watching the Spinner

If you have used Claude Code for more than an afternoon you have seen the words. Pondering. Noodling. Brewing. Spelunking. A random verb next to a spinner, a timer counting up, a token count, and underneath it a one-line tip about some command you did not know existed. People search for "claude code spinner verbs" and "claude code thinking words" for two reasons: they want to know whether the word means anything, and they want to change it. This post answers both from the current documentation, and then makes the case that the spinner is the one UI element you should stop staring at.

What the spinner verb actually means

Nothing. That is the short answer, and it is the honest one. The verb is picked at random each time a turn starts and shown while Claude Code waits for the model's response. "Pondering" is not a different state from "Brewing"; neither tells you whether Claude is reading files, writing code or reasoning about your question. The built-in list is a bit of personality, in the same family as the random loading messages in a video game, and the documentation describes the setting that controls it in exactly those terms: the verbs shown while a turn runs.

What does carry information is everything around the verb. The elapsed time tells you how long the turn has been going. The token count tells you roughly how much has been read and written. The tip line under the spinner is a rotating hint about a feature. And when the spinner is replaced by a tool call line, a permission prompt or a finished response, that change of state is the thing worth noticing, which is where the rest of this post ends up.

Changing the verbs: spinnerVerbs

The verbs are a settings key, so they can be replaced in any settings file: your user settings at ~/.claude/settings.json, a project's shared .claude/settings.json, or the local .claude/settings.local.json. The value is an array of strings, and the documented example replaces the built-in verbs with your own:

{
  "spinnerVerbs": ["Pondering", "Contemplating", "Analyzing"]
}

Claude Code picks one at random each time a turn starts. That means a single-entry array gives you a fixed word, which is what most people who search for this actually want: a plain "Working" instead of a thesaurus. A longer list keeps the variety but makes it yours. The one thing the array cannot do is make the word meaningful: there is no mapping from verb to activity, so do not write verbs that look like states ("Reading files", "Running tests") or you will mislead yourself.

Because it is an ordinary setting it follows the normal precedence rules: managed settings beat the command line, which beats project local, which beats shared project, which beats user. If a teammate committed a spinnerVerbs list to the repo's shared settings and you would rather not see "Vibing" forty times a day, put your own array in .claude/settings.local.json and it wins for you alone.

The tips under the spinner: spinnerTipsEnabled and spinnerTipsOverride

The line under the spinner is a separate feature with two settings. To turn the tips off entirely:

{
  "spinnerTipsEnabled": false
}

To add your own tips to the rotation, or replace the built-in ones, use spinnerTipsOverride, which is an object with a mode and a tips array. In append mode your tips join the built-in list and one is picked at random from all of them; in replace mode only yours are shown.

{
  "spinnerTipsOverride": {
    "mode": "append",
    "tips": [
      "Run npm run check before you ask for a review.",
      "Migrations live in db/migrations; never edit a shipped one."
    ]
  }
}

That second example is the genuinely useful trick hiding in this feature. A team can put its own onboarding reminders into the shared project settings, and every engineer who runs Claude Code in that repo sees them in the dead time while a turn runs. It is a low-pressure place for the rules that otherwise live in a wiki nobody opens: branch naming, the command that must pass before a PR, where the fixtures are. Keep each tip to one line; the spinner shows one at a time and the whole point is that it is read in a glance.

A caution from the settings documentation's own worked example: a value you set can be overridden from above. If you turned tips off in your user settings and they came back, a project settings file or a --settings flag on the command line is turning them on again, and the fix is to set your preference at a higher level (project local beats shared project) rather than to keep editing the user file.

Thinking words versus thinking

The search for "thinking words" sometimes means something else: the extended thinking output, the model's reasoning that some versions of Claude Code display as it works. That is not the spinner. Extended thinking is toggled with Option+T on macOS or Alt+T on Windows and Linux, and on the current model generation (Opus 5.5, Sonnet 5.5 and the Fable models) the toggle has no effect because those models always use extended thinking. The detailed view of what Claude is doing, tool by tool with timestamps, is the transcript viewer, toggled with Ctrl+O. If you want to know what Claude is actually doing while the spinner spins, that is the answer; the verb never will be.

Why you should stop watching the spinner at all

Here is the thing about a randomly chosen verb and a counting timer: they are designed to be looked at, and looking at them is the most expensive thing you can do with a coding agent. A turn that takes four minutes is four minutes of your attention if you watch it and about five seconds if you do not. The spinner gives you no signal about when it will end, so there is nothing to wait for; the only event that matters is the change of state at the end, when Claude finishes, or stops with a permission prompt, or fails.

That event is exactly what Claude Code's notification hooks and the preferredNotifChannel setting exist for, and what AI Done Now turns into a phone or desktop notification you cannot miss. Set that up once, following our Claude Code notifications guide, and the spinner becomes decoration: you kick off the task, switch to something else, and come back when the alert fires. The verb can say whatever it likes.

If you run Claude Code inside tmux or over SSH, where terminal bells and desktop alerts do not reach you, the tmux and SSH notifications post covers the routes that do. And if you want the waiting time to be worth something physically rather than just mentally, NotchFit turns the same "Claude is done" event into the end of a guided micro-workout in your MacBook's notch.

Quick reference

• spinnerVerbs: array of strings; replaces the built-in verbs; one is picked at random per turn. One entry means a fixed word.
• spinnerTipsEnabled: false hides the tip line.
• spinnerTipsOverride: object with mode (append or replace) and tips (array of strings). Good for team reminders in shared project settings.
• The verb has no meaning. The elapsed time, token count and the end of the turn do; let a notification tell you about the last one.

Settings names are taken from the official Claude Code settings reference as of October 2026. They do change; if a key here is rejected, check the reference for the current name.