Stop wasting your tokens: A practical guide for DevOps engineers using Claude Code
I didn’t care about tokens when I started using Claude Code. Everything felt fast. You type intent. It executes. You move on. Then one day, right in the middle of a task…You hit a limit. A big pause in the middle of your flow state. That’s when it clicks: Wait, what usage limit?
If you use Claude Code or any LLM, this matters more than you might think. This guide is part of my new book, Claude Code Handbook for DevOps Engineer. In this article, I am sharing one of the practical chapters from my book: Saving Your Tokens.
What are tokens? And why should I care!
As you use Claude Code more, you start consuming tokens. Tokens are fundamental to how LLMs work, and they are literally the currency of LLMs. Because every extra word = more tokens. They directly control your LLM cost and your daily usage limits.
Here’s a simple representation of how you are charged by token (credit to Matt Pocock).
How to check my token consumption
Claude Code gives you a built-in way to track token usage in real time. Just run /usage in your Claude Code prompt, and it’ll show details about which model you use the most, and how many tokens you have consumed.
Use arrow keys to navigate, and get more stats such as which model you used the. This should become a habit, especially during long sessions or debugging work. If you don’t track it, you will hit limits unexpectedly.
TIP: The Haiku model uses the fewest tokens and is less capable than Sonnet. Sonnet is usually best for everyday use. However, if you need higher performance, you can use the more advanced Opus model — but keep in mind that Opus consumes tokens significantly faster.
Similarly, you can also check the same from your Claude account usage page. When you run out of tokens, Claude Code will warn you. At that point, you have two options:
- Wait for the reset
- Enable pay-as-you-go with /extra-usage
NOTE: Some users also get a small one-time $20 spent credit for extra usage. But relying on this is not a strategy.
The real problem: unnecessary token burn
Claude Code is powerful. But it is also… generous. It explains more than needed. Adds context you didn’t ask for. That’s great for learning, but terrible for your token usage. Because every extra word consumes tokens, and more tokens = more money + faster you hit your limits.
And most of the time, you don’t even need that much detail. As DevOps engineers, we already optimize everything: CPU, memory, storage. But suddenly, we ignore the one resource we use the most with AI.
🗿Meet caveman — A simple way to reduce token usage
One practical solution is to force Claude to be concise. The shift is small, but important. You stop asking: “Can you explain this?” and start asking: “Give me exactly what I need”. You stop treating it like ChatGPT and start treating it like a tool inside your workflow.
As DevOps engineers, we care about signal, not noise. You can do this using a plugin called caveman. It tells Claude to respond in a short, minimal “caveman” style.
You can install caveman using Claude plugin marketplace:
claude plugin marketplace add JuliusBrussee/caveman
claude plugin install caveman@cavemanIt’ll be automatically enabled, and over the period of time, you will notice: Shorter responses, less fluff, lower token usage.
Go save your tokens!
Token usage is not just about cost. It affects how you work. If you burn tokens inefficiently, you: break your flow, hit limits faster, lose momentum.
Start treating tokens like any other resource in DevOps. You optimize CPU. You optimize memory. Now optimize tokens.
This is one chapter from my book Claude Code Handbook for DevOps Engineer. It is a practical guide to using AI directly in the terminal for DevOps engineers. It is currently free for a limited time while I continue shaping the book. If this helped, check it out and explore the rest.
