Tariq claims HTML beats Markdown for critical AI agent planning
The Shift from Markdown to HTML
For months, the AI coding community has treated Markdown as the gold standard for structuring prompts and receiving agent feedback. However, Tariq from the Claude Code team recently sparked a massive debate by suggesting that Markdown is no longer the optimal format. While Markdown is token-efficient, it often leads to cognitive overload when agents present complex, multi-option architectural plans. HTML, despite its higher token cost, allows for side-by-side comparisons and interactive elements that prevent developers from skimming over critical technical details.
Prerequisites and Tooling
To implement high-fidelity HTML planning, you should be familiar with Claude Code or similar AI agents. You will need an Anthropic API key and a basic understanding of how Large Language Models (LLMs) calculate costs via tokens. Specifically, this technique is most effective when using the Claude 3 Opus model, which handles complex formatting with higher reasoning capabilities.
Implementation via Visual Explainer
One way to achieve this is through the Visual Explainer skill. While not an official Anthropic release, this tool has gained traction for converting text-heavy plans into structured web pages.
// Example prompt for an HTML-based plan
"Analyze these three authentication strategies for my Laravel app.
Provide the answer as a structured HTML page with side-by-side
comparisons, pros/cons, and terminal commands."
The agent uses its internal reasoning to wrap the response in <html> tags. When processed, this opens a browser tab where you can compare Laravel starter kits versus manual implementations without scrolling through endless vertical text blocks.
Syntax and Best Practices
When requesting HTML, use specific tags like <details>, <summary>, and <table> to force the AI to organize data hierarchically. This avoids the "wall of text" common in Markdown. Always explicitly ask for CSS within a <style> tag to ensure the output remains readable in a browser environment.
Weighing the Token Cost
In a test involving a Laravel authentication plan, a standard Markdown response consumed approximately 2% of a 5-hour usage limit, while the HTML version jumped to 5%. This 150% increase in token usage is significant. However, for foundational decisions like database schema or security architecture, the cost is a justified investment against the risk of missing a critical "con" buried in a list of Markdown bullets.
- Anthropic
- 22%· companies
- Claude Code
- 22%· products
- Laravel
- 22%· products
- Claude 3 Opus
- 11%· products
- SpaceX
- 11%· companies
- Tariq
- 11%· people
Planning with AI Agents: Markdown vs... HTML?
WatchAI Coding Daily // 9:46
This channel is not for vibe-coders. It's for professional devs who want to use AI as powerful assistant, while still keeping the control of their codebase. My name is Povilas Korop, and I'm passionate about coding with AI. So I started this THIRD YouTube channel, in addition to my other ones Laravel Daily and Filament Daily. You will see a lot of my experiments with AI: I will try new things and share my discoveries along the way.