What You'll Have By The End Of This Guide

• One line pasted into your CLAUDE.md that stops Claude Code from reprinting an entire file just to change two lines — so the tokens you actually pay for drop hard, and you go far longer before hitting a limit
• The exact copy-paste rule, plus a one-session version you can paste at the top of any chat when you don't want to touch a file
• An honest model setup: a cheaper default model, a one-command switch to Opus only when it's worth it, and cheap models assigned to the small jobs so grunt work runs on Haiku automatically — no magic, just a system

Setup time: ~10 minutes

Skill level: Beginner. If you can open a file and paste a line into it, you can do this today. No new software, no plan upgrade, nothing to install.

First — Why You're Actually Blowing Through Your Limits

Read this part before you skip ahead, because it's the whole reason the fix works.

Claude Code charges you (and counts against your usage limit) by tokens — roughly, the words going in and coming out. Here's the trap: when you ask for a small change, Claude often rewrites the entire file and prints all of it back to you, plus a paragraph explaining what it did. Change two lines in a 400-line file, and it can reprint all 400. Every one of those reprinted lines is output tokens you pay for — for code you already had.

Do that a few dozen times in a session and you've burned a huge slice of your monthly limit on nothing but re-typed code.

The fix is to tell Claude, once, to only show you the lines that changed. That's it.

One honest caveat up front: this does not make Claude dumber, and it is not an unlimited hack. It still reasons just as hard on tough problems — it simply stops printing back code you already have. Usage limits are based on your total token use over a rolling window, so cutting the wasted reprints buys you a lot more runway, but it's a real efficiency gain, not a magic switch.

💼 Want Us To Just Build This For You?

This is the DIY version. If you'd rather have a custom AI setup built, tuned for cost, and wired into your tools for you:

We build done-for-you AI systems for service businesses.

• Free AI audit of exactly where you're losing hours (and burning AI spend)
• A custom 30–120 day build that runs on autopilot
• Works with your existing tools (Claude, your codebase, your stack)