Mike Wilkerson writes here

tldr;

1. I never use AI to compose or even review blog posts.
2. I often use AI to help me write better code myself (some refactoring, writing boilerplate, investigate/review my code), and rarely to fully-vibecode something.
3. I will explicitly put a disclaimer on anything that I do vibecode.
4. I am responsible for every line of code I publish, even if I allowed AI to write 100% of it.

On AI and writing

I am generally very optimistic about the potential of AI, but not for content creation. I can't stand the thought of using AI to generate written words that I pass off as my own. I write here because I enjoy the process of writing, and because it helps clarify and distill my thoughts. Or because I want to talk about something I find cool. Or because I have something to share that I hope will be useful to others.

My blog posts are in no way written or even shaped/reviewed by AI. Any grammatical errors or structural misuse of the English language is part of my own unique communication fingerprint, and I don't want that to be stamped out or homogenized by an LLM.

On AI for coding

I love to write code. Even if AI completely replaces software engineers at some point in the future, I will still write code because I enjoy it. I do not currently believe that AI is capable of fully replacing humans in most aspects of life.

That being said, I am most certainly not anti-AI. I leverage LLM's daily in some form or another for assisting with code. It may be worth a full blog post at some point on my current approach/workflow, but to-date this mostly looks like using Claude Code in a tmux split to investigate my code, figure out something I'm stuck on, write boilerplate, or generate small examples for me to play around with and then build into my codebase myself.

I have explored vibecoding and have even had AI completely write a codebase from scratch. However, I have not yet figured out a fully-agentic workflow that actually solves the big, meaningful problems I'm working on (for example, pushing the state of the art of bias detection and remediation in AI/ML models at SolasAI).

But so far, for small and inconsequential things, vibecoding is a delightful way to quickly go from idea to tangible proof of concept in no time at all. I like that vibecoding helps me touch and feel things I have a idea for, to see if I think they are worth pursuing in a more robust fashion. I'll probably do a full post about this in the near future, but the most notable example of this for me is an LSP multiplexer for Emacs that I had Claude write in Go. I use it every single day and it works great.