Your agent writes text. RichView renders the story.

01Why RichView exists

The analysis is fine. The page is the problem.

Ask a coding agent for an analysis and the numbers usually hold up. The hand-off doesn’t: a wall of Markdown, a chart in library defaults, or a one-off HTML page that looks different every time you ask.

Design shouldn’t be regenerated on every run.

Typography, chart conventions and layout are decisions you make once and reuse. A model re-improvising them token by token is slow, inconsistent and hard to review.

02The idea

Borrowed from JSON renderers.

Tools like json-render settled on a contract that works for generated interfaces: the model writes a small, constrained spec, and a renderer turns it into components it already knows. RichView applies that contract to data stories.

Markdown, charts and media in one plain-text file.

A .rmd file is ordinary Markdown with tagged YAML blocks. Every block is checked against a schema; one that fails comes back as a diagnostic instead of silently disappearing.

• chart
• table
• big
• callout
• image
• embed
• audio
• video

03One file in, one story out

Four rows of YAML. Four bars.

Each data row becomes a column. RichView renders the chart and its interactive controls from your .rmd — no chart-library setup for the agent to handle.

The RichView mark has always been a bar chart. Here its four bars are real: Anthropic’s share of company mentions in Hacker News comments, by year.

Charts that answer back.

The renderer checks every chart as it draws it — empty data, missing encodings, titles long enough to clip, non-numeric values — and returns the verdict as JSON the agent can act on before anyone opens the page.

Loading…

04Themes

Same file, any style.

Content and style live in separate files. The renderer applies a .rstyle.yaml stylesheet to the same .rmd: type, colour, spacing, and the chart’s ink and palette change together. The words and the numbers don’t move.

Three house styles ship with the beta.

sleek is dark and geometric, scientific reads like a working paper, and brutalist sets condensed headlines against heavy rules. Switch between them in the playground. These are real renders of the same demo file:

See every theme on a real report.

The flagship report runs through each house style and the design-system explorations, with the same sources, numbers and annotations in every one.

05Human + agent

You and your agent edit the same file.

The agent drafts; you correct. Change one word — type: bar to type: line — and the chart redraws. The .rmd is the source of truth: diffable in git, readable by both of you, no proprietary editor in between.

Then the agent does the tedious part.

Ask for monthly detail and the agent swaps the data source: forty-five months from the flagship report, two series, one encoding line. In July 2025, Hacker News comments naming Anthropic or Claude outnumbered those naming OpenAI or ChatGPT for the first time.

06The flagship example

The background was the data all along.

Every dot behind this page is part of a month of Hacker News comments that mention OpenAI/ChatGPT (ink) or Anthropic/Claude (clay): what one tech community talks about, not public opinion. It is one of 643 dated sources behind the flagship report.

OpenAI versus Anthropic · a RichView report

Anthropic is winning the people who build with AI. OpenAI still has everyone else.

By October 2026 Anthropic leads OpenAI with developers and the businesses that buy AI, OpenAI still leads consumers by an order of magnitude, and on capability the two are level.

Depending on which population is measured, Anthropic’s share of the two-company total runs from 2.4% (US App Store ratings) to 71% (professional developers’ coding agent at work). Each share comes from a different population and measure; none of them is “the public”. Sources D-ITUNES, S065.

Researched from public data with Apify and Exa, rendered with RichView.