Absolutely not.
I have tried the past weekend, ChatGPT, Gemini, Claude and Junie(Claude/Agentic Agent), and even Ollama. I tried Python, Typescript/Node and Java
It tends to use outdated packages and dependencies, which in some cases are vulnerable.
Here is my recent experience.
- I have tried having it do a simple frontend, and backend. What would have taken me an hour, took an entire weekend, and in the end, I had to feed it specific reasons as to why it wasn't working, and eventually just finished it off myself. Test cases it bends to pass them, without understanding why it's failing
- Yesterday, I had a Kubernettes issues after doing an upgrade, knowing the reason, I thought let me try the AIs again to see if it would catch it. After around 5 hours ensuring me it's solution this time is foolproof, I again, gave up, and just did an apply with ignore flag.
These AIs do work well when you treat it as a faster google aggregator, and treat the solutions it spits out as broken.
Vibe coding works well for the average joe that wants to write a simple script for their own use, but it will never get you to be an actual paid developer. And the current Graduates/Juniors relying on AI, will never attain the experience to be a confident feature/project lead.
