The Trap

In the past 2 years I’ve used AI constantly to accelerate my learning. But the way I’ve done it has differed significantly depending on the topic. I’ve now realized that one of the ways I was using this tool was slowing down my learning and atrophying my coding muscles.

Picture this, you want to learn coding so you start thinking about what kind of project you could work on. You brainstorm and possibly use chatgpt to get some ideas and orientation as to where to start (so far so good). Then you begin coding. You forgot some stuff in your language so you ask chatgpt to break some stuff down for you but you feel like things are too slow, things are painful. So you ask chatgpt to generate some code thinking it’s boilerplate anyways and you understand the underlying concept so “it should be fine”.

This right here is the trap. Not the brainstorming or the breaking down of concept, the mass generation of code you could’ve never written yourself.

What learning should feel like

Learning should be painful, true learning comes as a result of friction and when you remove it, you get empty satisfaction. The problem here is that we’ve convinced ourselves that output constitutes learning. So when a tool like generative AI comes along with the promise of fast outputs, our monkey brain goes “fast learning yay me do”. But unfortunately it’s been proven time and time again that learning comes from process and that the more frictionless the process, the weaker the aquired understanding is.

How to frame code generation

There is a way to frame code generation in a way that is not harmful. Simply treat sessions where you generate code, use AI tools etc. as AI tool learning. Do not fool yourself into believing that you are learning to code or math or anything else, otherwise you will find yourself vastly disapointed time and time again when you’re unable to convert those skills or explain your own projects. If you frame these sessions as such, you will make time for proper learning sessions in which you actually learn things.

All that being said, AI can absolutely be used to learn and to accelerate learning by a lot but it needs to be done the right way and you need to be aware of what kind of skills you’re developping as well as what kind of friction you’re running into. If you’re struggling to understand a coding concept and digging into it and asking questions about it, you’re learning about coding, if you’re struggling in making an AI code editor understand your prompts and getting it to pull up MCP docs, you’re learning AI tool use. If you’re asking AI to give you the answer to your math homework, you’re learning prompting but if you’re digging into a concept, asking all the necessary questions to rederive it yourself, understand where the ideas come from and how they fit into a grander scheme, you’re learning mathematics.

Essentially this all comes down to “you have to do what you want to learn”. So if you want to learn mathematics or coding do math and code. If you want to learn AI tool use, use AI, but just be very careful not to delude yourself into thinking you’re doing one whilst doing the other.