i think someone should do a study into AI psychosis and it’s correlation with when the person adopted vibe coding. i think hopping on at a higher model speedruns your psychosis but if you started with the bad models, you develop more tolerance.
example:
my first time dabbling in vibecoding was on replit in early 2024. I built a frontend for an agri-commerce idea with my my friends. a lot of the time i had to jump in and write tailwind css myself because the models were atrocious at UI but somehow they had the rest down.
then i hopped on beta cursor for a course project in late ‘24 to save me after i had heavily procrastinated on some node js microservices. i was still in school so i was getting graded for it. spent the night trying to iron things out with cursor and saw it dig me deeper into trouble, I had to get a few friends together to help out.
then i switched between augment and cursor for my final year project. still had to do a substantial amount of coding by hand. to get things right. but here i was working in areas unfamiliar to me… i was doing integrations with ollama api and hugging face transformers trying to
fit a twi language model into a chatbot and set up RLHF. integrating AST and TTS apis from third parties. i learnt a lot and AI sped up writing the code to get the job done but i had to learn a lot to steer the agents.
i also had one completely vibecoded tool for data gathering. a sort of video clipping desktop app that we needed to split long news broadcast youtube videos into sentences for transcription. i remember it being too heavy for my pc to run so i thought it was bad but it was actually really decent on good hardware and sped up our work.
then for my NSS i had to learn java in a week and submit a
lab for a technical mentor to review and grill me on. so i couldn’t dodge the learning and had to do work hands on week after week for two months.
then in june i gained almost infinte access to the strong but currently outdated models of late Q2 2026. I have done enterprise projects, personal projects, reviewed lab exercises, personal terminal tools and an agent harness for learning. mobius, coming soon
I think i have minimal to optimal psychosis. like on a scale of 0 to 10, 0 being luddite farmer and 10 being Token Jesus Hansen, i’m at a 6.85. meanwhile those who were skeptical until the models got very good are higher on these charts. this is my hypothesis.