Hi Everyone,
What a great thread! I’ve been enjoying “vibe coding,” but I also share the concern that it can drift into messy or substandard patterns. Since I’m not a hardcore programmer, I thought I’d see how far I could get using ChatGPT-5.
As a first step, I gave it the original code and simply asked:
“What is substandard about this code? Please optimize it and ensure it follows expert best modern practices. I would like to download the resulting files in a zip file if possible.”
It produced this first pass: V1 ChatGPT5 Improvements Only (CodePen)
Then I shared the observations that [@krilnon] and [@madanva] had made in this thread, and asked ChatGPT to map out:
-
Which issues it had already fixed,
-
Which ones it had missed,
-
Any additional ones it spotted on its own.
It produced a markdown file (which I further edited): V1 Comparative Summary (Markdown)
From there, I asked it to implement all the missing items, and for anything subjective/contradictory, to add HUD checkboxes so you can toggle them on/off.
That gave me a second more collaborative version: V2 sheathe + ChatGPT5 Version [CodePen]
Here is a summary: V2 Implementation Summary [Markdown]
My questions for you all, if interested:
- How did ChatGPT5 do with no help? V1 (CodePen)
- Any thoughts on the collaborative version? V2 [CodePen]
I definitely lean more toward thinking that non-experts can get very good results with the right approach - especially using the best-trained AI models for code reviews and asking targeted questions, combined with rapid learning along the way. I’m curious to hear others’ experience with this approach?
P.S. I’d love to see how Claude, Gemini and Grok would do on this experiment with my same initial prompt.
P.P.S. Hi @kirupa I’ve been a fan since back in the days when you were making games with Flash! Have always loved your content!