09 Apr 2026

I Tried Vibe Coding

For so long i’ve been seeing all over the internet that people are building things crazily by vibe coding them or giving the handle completely in the hands of LLMs.

I’m not an anti-AI guy, i use AI myself to do some of the repetative/boring work. For interesting work i would brainstrom on my own till i feel like i’m not smart enough for this, and just give it upto AI. So i wanted to try vibecoding few projects where i completely give the handle to AI and just proompt my way through. For this i am too broke to afford claude code or openai codex, so i chose opencode with out of the box “Big Pickel” model.

observe

Observe is a native GO command line tool that shows your native CPU/Memory usage.

Skills.md and agents.md

skills.mdagents.md
LLMs do not produce the same result all the time,so skills.md will define a way to get consistent output all the timeLLMs do not have the same personality all the time, agent.md tells the LLM how to think and behave
specific Instructions to follow when a repetitive task is askeda personality/way of thinking to follow
place in different folders such as:SKILLS/timesfm/SKILL.md: How to use timesfmSKILLS/ARIMA/SKILL.md: How to implement ARIMA algoagents.md/calude.md at the root of the project dir.

prompt

A native go application which shows you your cpu and memory usage in realtime.

I like how it asked me questions to clarify.

Question
opencode clarifying questions before actually starting building

Final result: Basically one proompt

final-product
final executable generated by code generated

I am scared for my job and life now!!


Now what??

  • AI is sure going to take my job away!!
  • If writing code has become easy, then valuable skill would be to network and know how to talk to people(which i believe has always been important pre AI era also)
  • I did not learn anything from GO’s perspective, IDEK what it did, i should look into the code and read it and try to recreate this exact thing
  • Maybe i am noob and haven’t done some real world product, and mostly this task i asked has already been done many times and is present in the training dataset of this LLM