It's been a while since I've written on AI.
After working on clear planning for Vibe coding for several months, I became overwhelmed by too many text files and even felt repelled.
Karpathy mentioned five reasons why he abandoned Markdown (.md) a few days ago.
As I decided to use categories, I want to share my experience.
First of all, I can't abandon .md files. They are efficient when delivered to AIs and agents. So, I make two copies if necessary.
For documents that I want to review in detail, I use HTML, and for those that I just need to check briefly, I use .md. By changing the Claude guidelines with this rule, the burden has been significantly reduced.
Increased efficiency is a bonus.
Making two copies may seem like a waste of tokens, but while I can't say there's no waste, it seems cost-effective compared to the benefits.
"One HTML copy: typically 2~10KB
One MD copy: 1~5KB
Additional output tokens for creating two copies: approximately 1,000~5,000 tokens per response. It doesn't affect subsequent turns (it accumulates in the conversation but minimally). If it's rational to make two copies because you need them, then making them is the right thing to do." (.feat Claude)
I hope this will be helpful if you are also struggling with text that needs to be read.
One-liner review: "Refreshing for the eyes."
