I believe people should understand the products and technologies they work with in order to use AI effectively and build good solutions. For a new project, I'm trying to learn several technologies and problem-solving approaches at once. Compared with a few years ago, learning feels harder because AI can produce solutions so quickly that I often feel overwhelmed trying to understand them. I end up circling back to verify whether the solution is correct, which can leave me even more confused. How are you balancing AI-assisted productivity with actually learning the underlying concepts?
5 Answers
Slow down and separate the goal of learning from the goal of shipping. Break the project into small, isolated pieces and stay with each one until you can explain it clearly. Build the simplest version first, repeat the exercise, and avoid adding complexity until it’s necessary. Ask yourself why each part works, what happens at different layers of the system, and whether the result matches your expectations. AI is useful for feedback, but it shouldn’t choose the entire path for you. Keep notes, use version control or snapshots, and be willing to read authoritative documentation more than once.
I try to write the solution myself first and use AI for hints, corrections, examples, and review. When learning something like configuration files, I’ll study a simple example, reverse-engineer it, then write my own and ask for feedback. In a lab, breaking the setup and fixing it is often more educational than getting a perfect answer immediately.
Reading and asking focused questions still works. Use AI to break a difficult topic into smaller explanations, clarify specific gaps, or quiz you, but don’t outsource the entire learning process. The important distinction is whether you’re using it to understand a concept or merely to avoid doing the thinking yourself. Learn new technology when it serves a real project or goal, so the knowledge has somewhere practical to stick.
Use AI as a tutor and study-plan generator rather than an end-to-end solution machine. Give it your current knowledge level and ask for a tailored sequence of lessons, labs, and evaluations. Have it explain every unfamiliar command or configuration, then ask questions whenever something is unclear. At the end, take a quiz, tear down the lab, and rebuild it from memory. That repetition exposes what you actually understand.
The rebuild-from-memory step is a great addition. I’ve mostly been building once and moving on, which probably explains why the knowledge doesn’t stick.
A good approach is to make AI explain the reasoning before showing the implementation. For example, when learning a protocol, build a small broker or client step by step in a language you already know. After each change, ask which part of the protocol was added and why. The same method works for things like container runtimes: implement a tiny piece, test it, break it deliberately, and investigate the failure instead of asking for the whole finished system.
This is exactly the kind of concrete process I was looking for. I’m going to try it myself first and then use the same approach with my younger teammates so we focus on understanding rather than just getting a result.