Introduction
Not long ago we got agentic coding agents, which have been immensely useful for churning out code and reducing the boilerplate you have to write. We have come a long way since copy-pasta StackOverflow development. However, these tools have a lot more to offer. You don’t have to use them for coding only.
In this short article I’ll talk about how I’m using Claude Code and Codex to learn Kubernetes (K8s).
I’ve been aware of K8s for a while but never had to use it directly until I joined a team at work that uses it. Note that what I have learned so far is a work in progress, as I’ve yet to finish the lessons.
How I’m using AI to learn
Books are great, and I have learned a lot from them, but they can get boring pretty quickly. Some books are dense with paragraphs and offer you very little opportunity to get practical. Sure, you can copy and paste code from them, but you’ll be left fixing bugs, as some of these may rely on outdated libraries or frameworks.
I’m not anti-books. I do encourage you to use them as a reference manual and as a way to understand the why behind the things we do in tech — that’s one thing they’re good at. They excel at this aspect. Oftentimes I have to nudge my agent a little to get it to explain things for me.
So how am I using AI to learn?
The first thing I do is choose a topic I want to understand. These are usually tech-related and involve some code or running commands. It might be mastering new Java features, solidifying my TypeScript knowledge, and so on. Recently I embarked on learning K8s so that I’m effective at work using it. As you can already see, it’s important to first have a goal in mind.
Once you have the goal, your AI agent (Claude or Codex) can develop your learning plan with lessons and dense explainers in Markdown. Normally I have it create a lessons directory, covering various aspects.
For K8s I have 20 or so lessons. The aim is to develop a microservice architecture app and deploy it on K8s using MiniKube. For the lesson plans I used Claude Code, as it’s perfect at this, and Codex is great for implementation.
Each lesson basically introduces a K8s concept and then leaves out tasks for me, such as using Spring Initializr to generate an app and applying YAML files using kubectl.
A master CLAUDE.md and AGENT.md are created to guide how the lessons work. The agent will basically tick off a lesson when you tell it to. That way you can keep track of your progress, and the agent can also avoid interacting with lessons that are already completed.
You must tell it to create this file along with the lessons. All you need to tell it is to develop the Markdown lessons with great detail, covering various K8s commands and concepts. So like a book, you read the lessons, and each lesson has steps for you to complete at the end. It will normally drop code snippets for you, such as commands.
The killer feature in using AI for learning is that if you get stuck, the agent can get you unstuck. I stopped implementing the microservices myself, for example — I was only generating the starter with Spring Initializr, which gave me the barebones project structure. I would then instruct it to implement the microservice but leave things like setting up infrastructure and various other K8s commands for me, as the whole point is to learn operating K8s, not programming. This was so good, because I can see something akin to a real-world application and all the various commands I need to run to get it working on K8s.
If you get stuck understanding something, you can always instruct the agent to explain the concept or show you what you need to do. You are free to choose how much you want it to hold your hand.
I like to call this method of learning interactive-live-learning. It’s like a book that is alive: you get stuck somewhere, and you ask the book to point you in the right direction; the code is outdated — it fixes it in real time. No need to go post on some forum and wait for answers for hours. It’s hands-on and helps you solidify what you’re learning.
It’s not perfect, but it’s a start
Sometimes the explainers are very terse, perhaps because they don’t want to waste tokens. So I have to supplement that by querying Gemini for more in-depth explanations. The lessons generally point you in the right direction, but you might have to do some work.
I ran into occasions where it promised to do something in a later lesson but never did, so I had to nudge it.
Summary
I’ve been using coding agents to learn K8s and various other topics, and the results have been impressive so far. The tools do something books could never do for me: interactive-live-learning.
If you’ve been putting off learning something new, give it a shot. Pick a topic, set a goal, and let the agent draft the lessons for you. Read them like a book, do the hands-on bits yourself, and lean on the agent when you get stuck. You’ll be surprised how much ground you can cover when the book talks back.