Why does markdown go unread?
Working with Claude Code, everything piles up as markdown — planning docs, code analyses, research summaries. The content is there, but past 100 lines the screen becomes a wall of symbols and indentation, and you end up skimming the top and moving on.
An Anthropic engineer on the Claude Code team admitted the same thing on the company blog this May: "HTML is the new markdown." He no longer reads markdown past 100 lines, so he started receiving output as HTML. The problem is not a shortage of documents — it is that generation speed has overtaken reading speed.
What changes with a one-line HTML conversion?
One prompt is enough: "Convert my research markdown to HTML. Organize channels as tabs on top, and group the content into topic cards." The same content, rearranged into tabs and cards, becomes visible at a glance without scrolling — comprehension speed changes noticeably compared to reading markdown.
Layouts, charts, and tables come out exactly as specified. Start with a single phrase — "make it HTML" — and progressively refine the structure you want.
What does the Playground plugin add?
The last mile static HTML cannot reach is interaction — dynamic views where you press buttons and drag sliders to explore. A plain conversion request rarely produces this well, so Anthropic packaged it as the Playground plugin. Installation is a single command in Claude Code — /plugin install playground@claude-plugins-official — with built-in templates per problem type.
Visualize a codebase's dependencies and modules appear as a color-coded node map, with per-layer preset buttons to focus on what you need. Comments you attach to nodes convert into feedback prompts — paste one into Claude Code and a targeted analysis of that module begins.
Execution flow works the same way. An eight-stage diagram shows how a question becomes an answer, and pressing a question preset highlights the relevant path. It takes less time than reading a text explanation and yields deeper understanding.
When HTML, and when text?
Receiving everything as HTML costs significantly more tokens. The rule is simple: finished artifacts best understood as tables, diagrams, or cards go to HTML; intermediate artifacts — drafts, memos, agent-to-agent data — stay in text to save tokens.
Why it pays off anyway: AI already produces documents faster than humans can absorb them. Closing that gap is what productivity means right now — not the ability to delegate to AI, but the ability to read and verify AI output quickly.