πŸŽ™οΈ How I AI: HTML is the new Markdown: How Anthropic engineers are building with Claude Code

Β· Source: Lenny's Newsletter Β· Field: Technology & Digital β€” Artificial Intelligence & Machine Learning, Software Development & Engineering, Emerging Technologies & Innovation Β· Depth: Advanced, extended

Summary

Anthropic engineer Thariq Shihipar advocates for HTML as a superior communication format for AI agents, replacing Markdown for project planning and specifications. Recorded live at Anthropic's Code with Claude event, Shihipar demonstrates how HTML artifacts enable richer expression through interactive elements, visual mockups, and scrollable sections, improving engagement with lengthy AI-generated plans. He highlights that engineers are evolving into "compute allocators," focusing on defining project boundaries and staying synchronized with agents, rather than solely writing code. The discussion covers building custom, disposable UIs for editing specific plan sections, creating living design systems in HTML, and the importance of simple, trusting prompts for Claude. Shihipar notes that only about 1% of generated tokens become production code, with the majority used for internal tools and rich interfaces.

Key takeaway

For AI Architects and AI Product Managers designing and overseeing AI-driven development, adopting HTML for project specifications and design systems is crucial. This shift from Markdown to HTML allows for more interactive and visually engaging plans, improving your ability to stay in sync with AI agents and refine outputs. Focus on defining clear project boundaries and leveraging AI to create custom interfaces for complex tasks, rather than getting bogged down in manual code writing or editing.

Key insights

HTML offers richer, more interactive communication with AI agents than Markdown, enhancing engagement and product quality.

Principles

Method

Use HTML for AI-generated project plans, interactive specs, and living design systems. Create disposable micro-UIs for specific editing tasks, and employ simple, open-ended prompts to guide the AI.

In practice

Topics

Best for: AI Architect, AI Product Manager, AI Engineer, Machine Learning Engineer, Director of AI/ML

Related on AIssential

Open in AIssential β†’

Editorial summary, takeaway, and curation by AIssential. Original article published by Lenny's Newsletter.