According to Anthropic’s Engineer, HTML is the New Markdown for AI Agents

2024-06-18 · Source: To Data & Beyond · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, medium

Summary

Anthropic engineer Thariq Shihipar argues that HTML is the new Markdown for AI agents, advocating for AI-generated outputs to be interactive interfaces rather than static reports. While Markdown suits short, disposable text like commit messages, HTML transforms complex AI outputs—such as code reviews, research briefs, and planning documents—into dynamic, reusable working surfaces. This shift enables features like SVG diagrams, interactive widgets, in-page navigation, collapsible sections, filtering, sorting, and state tracking, which Markdown lacks. The core idea is that "Markdown gives you something to read; HTML gives you something to use," moving beyond merely producing more text to creating better, actionable artifacts. Examples include turning a flat code review into a dashboard with severity badges and filters, or a task list into a draggable Kanban board that preserves state. The underlying principle is to optimize the artifact, not just the prompt, to reduce the cost of subsequent human action.

Key takeaway

For AI Engineers or Prompt Engineers designing agent outputs, you should prioritize generating interactive HTML interfaces over static Markdown reports for complex tasks. This approach transforms AI-generated content into actionable workspaces, allowing you to filter, sort, and update information directly within the output. By optimizing the artifact for human interaction, you reduce the cost of subsequent actions and streamline workflows. Evaluate whether your output is a disposable endpoint or a persistent workspace to decide if HTML is warranted.

Key insights

HTML transforms static AI outputs into interactive, reusable interfaces, enabling human action.

Principles

• Reports are read and discarded; interfaces let people keep working.
• Optimize the artifact, not just the prompt.
• A good AI response reduces the cost of the next human action.

Method

Use a three-question decision flow to choose between Markdown for static, disposable outputs and HTML for interactive, reusable interfaces. Employ a structured prompt template to generate single-file HTML reports with features like collapsible sections, filters, and interactive elements.

In practice

• Add "prefer single-file HTML for complex outputs" to agent instructions.
• Create a personal HTML report template with specific features.
• Publish and share a generated HTML artifact with a teammate.

Topics

• AI Agents
• HTML Interfaces
• Markdown
• Prompt Engineering
• Artifact Optimization
• Interactive AI Outputs

Best for: AI Engineer, Prompt Engineer, Software Engineer

Related on AIssential

• Why this Claude Code engineer uses HTML files as AI specs | Thariq Shihipar (Anthropic) — HTML is replacing Markdown as the preferred medium for human-AI interaction due to its richer, more engaging visual capabilities.
• Anthropic killed Tool calling — Anthropic's tool calling updates enhance agent efficiency and accuracy for complex tasks by optimizing token use and improving tool interaction.
• Faster Code, Slower Understanding: A Survival Note from a Year of Agent-First Engineering — AI shifts the engineering bottleneck from code production to comprehension and maintenance, demanding new optimization strategies.
• How To Build Your Own Agentic PDF-to-Markdown Converter — Combining document processing tools with AI agents enhances PDF-to-Markdown conversion flexibility and accuracy.
• I legit caught one of my Ai Agents watching a youtube and tiktok videos instead of working — AI agents can exhibit behaviors interpreted as "slacking," prompting questions about their internal processes and monitoring.
• How to Use Agent Skills — Agent skills enable reusable, dynamic AI capabilities, moving beyond static prompts for improved performance and reliability.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by To Data & Beyond.