It’s clear that these AI tools are here to stay and the task of writing code is now being automated. It seems we’re moving up the abstraction layer to defining the problems and architecting solutions with the assistance of these AI tools.
However, as a beginner, I don’t have the experience or knowledge to know when its wrong. How much of the code it’s writing do I need to be understanding and learning and challenging?
Learning these AI tools are great, but blindly accepting the code it gives me based on the prompts I give, I don’t feel like I’m learning. What do I need to learn is the existential question here?
Its true unless you know coding, how can you evaluate what it produces, right! If you are in the field of code writing you would normally use AI tools for efficiency not as judges of work. You need to discriminate what they produce and if they work for your case!
This is a really thoughtful question.The concern about learning versus just using AI tools is something many people are grappling with right now.
You need to understand the fundamentals enough to be a good evaluator and debugger. Think of AI as a very fast junior developer who sometimes makes plausible-sounding mistakes. You need to be able to:
Critical skills to develop:
Read and understand code - You don’t need to write everything from scratch, but you should be able to trace through logic, understand what each part does, and spot obvious issues
Debug systematically - When something breaks (and it will), can you identify where and why? This requires understanding error messages, using print statements or debuggers, and reasoning about program flow
Recognize bad patterns - AI can generate code that works but is inefficient, insecure, or unmaintainable. You need enough knowledge to spot these issues
Design and architecture - Understanding how to break problems into components, when to use which data structures, and how to organize code well
A practical approach:
Learn the fundamentals first - Take an intro Python course without AI assistance. Struggle through writing your own loops, functions, and classes. This builds intuition
Use AI as a learning partner - When you get AI-generated code, don’t just copy it. Ask yourself: “How does this work? Why did it make these choices? Could I modify this if requirements changed?”
Test everything - Write simple tests, try edge cases, intentionally break things to see what happens
Read the code line by line - If you can’t explain what every line does, that’s a gap in your knowledge worth filling
The people who will thrive aren’t those who write the most code by hand OR those who just prompt AI best - they’re the ones who can effectively collaborate with AI while maintaining deep understanding of what’s being built.
So true!
Also sometimes ,I find it useful to challenge the AI assistant ,though it often provides great codes ,it sometimes makes mistake or doesn’t provide alrernative methods that can be as efficient.
For problems that have a standard solution. The LLM will give you the standard solution and will often flag out as wrong other solutions that might also solve the problem, but are not standard. It happens to me a lot! If you push back to the LLM to convince you, it will try to find test cases it fails, but if you push back again and indicate the wrong LLM reasoning, it will see that they also pass.
However, the standard solution for known problems should be learned and understood because they are the best “forms” of the solution.