I recently talked about how my students were getting frustrated using ChatGPT 4.0 as a coding assistant. After some helpful tips here, we switched to CoPilot since it’s powered by GitHub. I also showed them how to integrate GitHub with Visual Studio. One student is making progress and really trying to learn C#, but the others are struggling (one even dropped out). Out of the remaining five, one new student is showing some promise.
In a recent session, two students complained about the code they got from CoPilot. I explained how they could get better results by writing clearer prompts. I also told them not to believe all the AI hype they’ve seen online, especially from people like Nvidia’s Jensen Huang.
Does anyone know a good YouTube channel I can recommend to them about using AI for coding? Also, can I share your advice with them? From what I’ve seen, you still need solid programming knowledge to use AI tools effectively. I’ve been sending them code and walking through it together, plus giving them assignments. They still think they can just jump in without understanding the basics. What do you all think?
The hype is real, but it only works well if you already know how to code. You need to spot issues and fix them yourself. AI tools can summarize things and help with new languages, but they’re not magic.
I’m not a hardcore coder, but I used to work with Perl a lot before retiring. From my experience, the better you understand the language, the better your prompts will be. When I tested ChatGPT, I was able to tweak the scripts it gave me and make them work for what I needed. If you don’t understand the basics, though, you can’t expect good results. You’ve got to do the hard work to learn coding first, then use AI to assist you—not the other way around.
Take it step by step. Use AI to generate small pieces of code, then test and refine them as you go. Big, sweeping changes usually lead to messy results.
You’re underestimating AI. I got Claude to write a full RF/EM simulator for me—a tool that usually costs tens of thousands of dollars. It had all the custom features I needed, and it only cost me a fraction of that.
I built an entire AI-assisted CRM in a single day. It has task automation, personalized email scheduling, team calendars, and more. The code is rough—no documentation or proper error handling—but it works. I’m now taking my time to refine it. Even as a junior-level coder, I managed to create something I never thought I could.
Copilot shines when you’re actively coding and using its autocomplete feature. It can predict entire functions or blocks of code. But using it to generate full programs? Not so much—it often misses the mark.
The real solution is for students to learn coding the old-fashioned way—by solving problems and building a strong foundation. Relying on AI too much will just lead to bad habits and poor-quality code.
Have them create a checklist for what the AI needs to generate good code. Once they realize they have to provide proper context, they’ll start getting better results. It’s all about learning how to communicate with the tool effectively.
