Exercise While Coding: Turn Claude Code Wait Time Into Micro-Workouts

Add up the minutes you spent today waiting on Claude Code. Every "refactor this module", every "fix the failing tests", every plan-mode run — each one is a few minutes where the agent works and you don't. For a heavy user that easily adds up to an hour or more a day.

Most of that hour goes to watching the terminal, checking Slack, or scrolling a phone. This post is about spending some of it on something better: short bursts of movement, sometimes called exercise snacks or micro-workouts, timed to your AI's work instead of to a calendar reminder you'll dismiss.

What is an exercise snack?

An exercise snack is a very short bout of physical activity — typically one to a few minutes — done several times a day instead of (or in addition to) one long workout. Think a set of squats, a minute of marching in place, or a few push-ups against your desk. A movement break is the gentler cousin: a stretch or a walk to break up a long period of sitting.

Neither is a replacement for proper training if you have specific fitness goals. But public health guidance such as the World Health Organization's physical activity guidelines encourages adults to be active regularly and to limit long, uninterrupted stretches of sedentary time — and the guidance is clear that some activity is better than none. For people who sit at a desk all day, frequent small breaks are one of the easiest ways to move more.

Why AI coding breaks are the perfect trigger

The reason most "stand up every 30 minutes" reminders fail is timing. The reminder fires when you're in the middle of a hard problem, so you snooze it. Then you snooze it again.

Waiting on an AI agent flips that. When Claude Code is working:

• You're already interrupted. Your train of thought has been handed to the agent. There's no flow state to protect.
• The break has a natural length. Most agent runs last a few minutes — about the length of one or two sets.
• It repeats all day. Ten, twenty, thirty prompts a day means ten, twenty, thirty chances to move, without scheduling anything.
• There's a clear end signal. The agent finishes or asks for permission — that's your cue to sit back down.

In habit terms, the prompt you send becomes the cue, the exercise is the routine, and getting back to a finished task is the reward. You don't have to invent a trigger; your workflow already has one.

A simple system: one prompt, one set

Keep it boring and repeatable. The rule we use: every time you send a prompt you expect to take more than a minute, do one set.

1. Keep a short list of moves. Five to ten exercises that need no equipment and no changing clothes. Our 15 desk exercises for programmers is a good starting menu.
2. Rotate by body area. Alternate upper body, lower body and mobility so nothing gets overworked and you cover everything by the end of the day.
3. Use a timer or a rep count. 30–45 seconds, or 8–15 reps. Stopping points matter; open-ended exercise is how breaks turn into distractions.
4. Return when the agent needs you. Not when you get curious. Checking the terminal every twenty seconds defeats the purpose.
5. Track streaks, loosely. A count of sets per day is surprisingly motivating. Don't punish yourself for a heads-down day with few prompts.

Making it work without hurting your focus

Keep the intensity low to moderate

You want to come back to the keyboard ready to review code, not gasping. Squats, wall or incline push-ups, marching and mobility work are ideal. Save burpees for the gym.

Don't let notifications cut you off mid-rep

If your AI finishes halfway through a set, finishing the set first is fine — thirty more seconds won't matter to your code. What matters is that you do find out it finished, instead of discovering it five minutes later. That's why pairing movement breaks with Claude Code notifications works so well: you can fully look away from the screen.

Watch for permission prompts

Claude Code often pauses to ask permission before running a command or using a tool. If you miss that, the agent sits blocked while you stretch. A notification that distinguishes "done" from "needs permission" lets you decide whether to finish the set or go approve it now.

Have an off switch

Some days you're debugging production and don't want to see a squat prompt. That's fine. The goal is a default, not a rule you have to obey.

What a micro-workout day looks like

A realistic example for a day with a couple of dozen agent runs:

• Morning: neck tilts, arm circles and cat-cow while Claude reads the codebase and makes a plan.
• Late morning: squats and wall push-ups while it implements the first chunk of a feature.
• After lunch: a high-knee march or step jacks while the tests run — the post-lunch slump is the best time for a little cardio.
• Afternoon: incline push-ups, calf raises and hip circles between review rounds.

None of it feels like a workout, yet by evening you may have done well over a hundred reps and broken up your sitting many times.

Automating it with NotchFit

The system above works with a sticky note and a phone timer. But the friction adds up: picking the next move, timing it, and checking whether Claude is done. So we built NotchFit, a Mac app that runs micro-workouts in your MacBook's notch while Claude Code or Codex works.

• When your AI starts working, the notch expands: an animated exercise with a countdown ring on the left, the agent's live progress on the right.
• Sets only start when you tap Start set, with rest timers between sets — so it never runs ahead of you.
• When Claude or Codex is done or needs permission, you get told. If you're mid-set, that notification waits until the set ends.
• Run /notchfit:plan in Claude Code (or $notchfit-plan in Codex) and your AI writes a weekly plan from your goal, fitness level, equipment, available time and injuries.
• Streaks, a history heatmap and a bit of confetti keep the habit visible. A Hide workout toggle turns it into a pure AI status panel on heads-down days.

The bottom line

AI agents gave developers something new: dozens of short, idle windows every day. You can spend them watching text scroll, or you can spend them moving. Pick a handful of desk-friendly exercises, tie one set to every prompt, and let a notification — not your eyes — tell you when to come back. Stop if anything hurts, keep it easy, and let consistency do the rest.