Learning with AI without becoming dependent on it or lazy

The narrow use case for AI in programming boils down to replacing labor you already know how to do. Do I personally enjoy spending 8 hours penning some lovely boilerplate? Yes. Is that an effective use of time? Probably not. So letting the robots do that work is a logical decision, and it works because I have the ability to correct it when it makes a mistake.

I applaud your foresight and think it's great that you're trying to avoid learning a hard lesson about overuse of AI. However, you'll notice that AI in the way that I've described it is seemingly impossible to overuse. Writing things you already planned to write in a faster way is not inherently detrimental. It's like driving an automatic, logically sound irrespective of the myriad individual opinions on the subject.

The problem appears when there is continued insistence on using AI outside of the use case I described. Using an LLM to write code you didn't specifically ask for is symptomatic of an inability to conceive of the forms required for implementing a feature, which is a much more fundamental programming skill than knowing how to type those forms (syntax) and doing the typing (busywork). A skilled programmer isn't completely lost when trying out a new coding language, they have learned intuitions about computing that apply everywhere... except Lisp obviously. 🙃

In terms of building that skill in the first place, you simply need to keep putting information into practice. LLMs are an acceptable source of true information to the extent that they correctly repeat that which is in their training data, but remember they are rather "goal-oriented". They are likely to provide the information you specifically asked for and unlikely to provide seemingly irrelevant information. That's the system working as intended. However when you navigate through a primary source like MDN for example, you're absorbing the domain of things that exist within your subject of study. Knowing what exists empowers you to ask better questions and iteratively refine your knowledge.

To help further explain, consider the skilled programmer from earlier. They know Java very well, but have never used Kotlin. They could try out Kotlin (JVM) and face no immediate friction, writing code that technically works. However, since they know the value of doing their homework, they skim the Kotlin docs for signs of unique syntax and stdlib symbols. They now know that Kotlin has distinct conventions regarding null and that they can use lateinit to better observe those conventions. It's important to build this instinct early, lest you feel compelled to vibe code a 50k LoC JS file because you haven't read the docs and antiquated tutorials neglect to address the advent of modules.

Unsure which question you want answered here as you didn't post an actual question?

Here's a quite similar recent post which might also be relevant: How to stop relying on AI when writing code?

Accepting that AI has made us lazy not to do anything is the FIRST step in progress. Hence start small and celebrate yr progress daily to keep motivated.

Start with simple CRUD and use AI to do auto-complete... The CRUD should target some of yr own need like budget/tasklist to get involved and feel the effort is useful beyond learning. Also use AI to generate yr own backlog of the features u want in yr app by conversing with some chatbots to generate the descriptions and acceptance criteria.