Since Thariq published "The Unreasonable Effectiveness of HTML," it's become a viral sensation across the developer ecosystem.
His core claim is that human cognitive bandwidth has a hard limit. When users are confronted with 100 lines of Markdown, their eyes glaze over, and people stop reading.
This creates a dangerous dynamic in which humans quietly transition from active reviewers to "rubber-stampers."
HTML fixes this by turning a flat text dump into a navigable workspace. You're asking the AI to build a disposable micro-app whose only purpose is to help your brain process dense information before you decide.
Most of the backlash is a misread. Critics point out that HTML costs 2-4x more tokens, kills clean git diffs, and is terrible as LLM input. All true. All irrelevant to what Thariq actually argued.
He's not saying replace Markdown everywhere. Markdown still wins for repo docs, version control, and agent-to-agent loops where no human is reading.
The token cost critique is also a red herring. HTML output is ~0.5% of your total spend in an agentic loop. The real cost drivers are system prompts, tool schemas, and conversation history retransmitted at every turn. You're optimizing the wrong variable.
Machine ingestion and human presentation aren't symmetric. Markdown is for machines. HTML is for humans.
Markdown was built for writing. HTML was built for interacting. When human oversight is the goal, readability is worth paying for.
