I've been testing AI models for coding, used codex deepseek claude and more for a small project
But, sorry at the end the code was hardly readable by me, too many duplicate codes to many crap code and chained functions
Many useless spaghetti code from what I saw like 8 or 10 functions that could be reduced into a single function
I deleted everything and coded everything by myself with auto complete to speedup so I understand exactly my own code
It toke some time, but right now even the smallest model like gemma does the job perfectly for me as long as I tell him exactly what to do and what to change like filename, function etc
And I do monitor the output to avoid crap code
I'm not good on frontend side especially tailwind so I let the models create a nice design, all models are extremly good at it even the smallest
In my opinion is, for long term vibe coding will not get well over the time, big companies like Claude or Openai focus mostly to make exactly this happend so people without coding knowledges will spend tons of money to create saas businesses
Personally as dev I don't care if there is opus 4.7 or opus 6.0 I understand my code, I use a small model to do what I say so my spending is not even 1$ per month
Ah I forgot, I tried once to ask to refactor my codes the result was just aweful.
Final verdict I love to code by myself it's more fun more challenging and learn new things every day
AI is good, but it should help coders to work faster, not coding full entreprise level apps without monitoring because that will surely go wrong.
What do you think ?