In the AI era, how do you actually learn and understand new concepts?

I've been in IT for a long time, and lately I've been going deep into DevOps/Kubernetes. One thing keeps bugging me about how we learn this stuff.

When I want to understand a new concept, docs and most articles feel written for people who already get it. Even with experience, I often only get it after reading the same thing explained a few different ways. Videos are the opposite problem: they run at the author's pace, not mine, so I keep pausing and rewinding, and often spend a lot of time to learn very little.

So both feel like extremes — articles too static, videos too dynamic and out of my control. I keep wondering if the missing piece is something in between: text at your own pace, with small interactive bits where you can poke at the idea until it clicks.

AI pushes me this way too. Now that it writes a lot of the code for me, I care less about memorizing details and more about quickly getting a correct mental model — knowing what I want to do, and letting AI handle the how.

So I'm curious:

• Do you also find docs too dense and videos too slow to pay off, or is it just me?
• When a new concept finally clicks for you, what usually does it?
• Have you seen any sites or articles that explain a concept with interactive animations or little playgrounds you can poke at — and did that actually help you, or is it a gimmick?

Comments

[Aldo]

Getting deep into Kubernetes and similar distributed systems often feels like trying to drink from a firehose. I've found that simply reading through official docs or tutorials, while necessary, frequently leads to a surface-level understanding. My own breakthrough with complex tooling usually comes from trying to build the absolute simplest possible working example first. For Kubernetes, that meant getting a single container deployed, then a basic service, then adding a persistent volume, and so on, incrementally. Each step reveals a new set of problems and concepts that become much clearer when they're concrete rather than abstract.

In this process, AI tools have become less about generating full solutions and more about accelerating my debugging and knowledge acquisition. When I'm wrestling with a Pending pod or a CrashLoopBackOff, I'll often feed the logs and my manifest into a large language model. It's not about getting a magic fix, but about rapidly exploring potential causes, understanding obscure error codes, or getting a quick explanation of a specific field in a Helm chart or Kustomize overlay without having to context-switch away from my IDE. It acts as an extremely fast, context-aware reference guide, letting me spend more time on the actual problem-solving and less on information retrieval.

However, the fundamental work of building a robust mental model still requires active engagement. After I've used AI to debug a specific issue, I'll often spend time drawing out the network flow, sketching how different control plane components interact, or even trying to explain the concept to a colleague (or myself, out loud). That synthesis, connecting the dots between components like the kube-scheduler, kubelet, and etcd, is where true understanding forms. AI can help us navigate the complexity, but it doesn't build the intuition or the architectural foresight that comes from wrestling with these systems yourself.