Claude Engineering Team Prefers HTML Over Markdown for Output
Anthropic Claude Code team engineer Thariq Shihipar stated that the team is increasingly adopting HTML instead of Markdown as the output format for AI, as AI-generated content has expanded from short notes to thousands of lines of plans, flowcharts, and code reviews, where Markdown lacks readability.
HTML supports features like colored tables, SVG flowcharts, and interactive prototypes. Although it consumes more tokens and generates slower, it significantly enhances the human-machine collaboration experience. Common uses include colored diff code reviews, interactive project boards, and prompt tuners.
Driven by events, the AI output format is transitioning to higher information density, benefiting tools that support rich media rendering, while workflows reliant on plain text are under pressure.
Source: Public Information
ABAB AI Insight
The Claude Code team has found in internal practice that as model capabilities improve, the advantages of manual editing in Markdown diminish for long structured outputs, leading to a preference for HTML to maintain effective human review of content.
In terms of resource mobilization, engineers achieve visualization and interactivity by generating self-contained HTML files, motivated by the desire to increase information density and engagement, avoiding long documents being hastily overlooked.
Similar trends can be seen in other AI tools evolving from plain text to artifacts or rich media outputs, currently at a stage of upgrading human-machine collaboration interfaces from text to visualization.
This essentially represents a technological substitution: older lightweight formats are replaced by carriers with stronger expressive power, as the scale of AI output surpasses the effective reading threshold of humans, necessitating the introduction of a structured visual layer.
ABAB News · Cognitive Laws
- Output format determines whether humans truly engage.
- Increased information density will inevitably lead to the obsolescence of old tools.
- A tenfold improvement in experience justifies the token cost.