
What happened
Anthropic beta-launched Claude Slides and Claude Docs, letting users build slides and documents directly in Claude chat, with Claude Cowork folded into the main chat.
Why it matters
The output space looks deliberately human-defined so that an AI can generate safely and render deterministically, which should make mass-producing internal decks more predictable and consistent.
What to watch
The test was run only in a developer environment, so real chat response speed and Japanese font rendering in claude.ai remain unverified; watch whether MCP exports such as Google Drive actually work.
WHO IT HITSTeams that generate large volumes of internal, template-style decks and documents are the clearest beneficiaries, since a closed output space favors stability and accessibility compliance over expressive freedom. Whether the feature feels the same in the real claude.ai chat UI is still unverified.
Summaries like this, in your inbox every morning.
Anthropic's announcement sits inside a broader consolidation: Claude Cowork, previously a separate workspace, was merged into the normal Claude chat under the framing of "One Claude," so users no longer choose between chatting and working. The editorial team went past the announcement and wrote files directly from the developer-facing environment that powers the feature, generating a four-slide deck from rough meeting notes. Each slide turned out to be a single HTML file, and the whole deck a single JSON index.
The internal design is unusual in where it puts the constraint. Allowed CSS properties are a fixed list, and the values each may take are limited to px ranges, enumerated keywords, or hex/rgb/hsl colors; margin, float, grid-template-areas, :hover, and keyframe animations are excluded. Custom tags such as <x-shape>, <x-icon> and <x-connector> replace freehand SVG drawing, and SVG itself is capped at 52KB with scripts banned and local references only. Contrast is specified numerically — 4.5:1 for body text against the background, 3:1 for large text — and typography is limited to 4–5 reusable sizes and at most four typefaces per deck, usually two.
The trade-off is legible: expressive range is narrowed so that every output fits the canvas and follows the same visual rules, which is coherent with producing internal, standardized materials in volume. The main open question is whether that design survives contact with the real product — the team could not open the chat UI, so response speed and Japanese rendering quality are unconfirmed, and the ChatGPT Canvas contrast is explicitly framed as inference from public behavior rather than verified internal access.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
After OpenAI staff raised concerns on Slack, president Greg Brockman abandoned the second half of his $50 mill…

GovAI research fellows Alan Chan and Sam Manning said at a Sept

A practitioner listed five Japanese-language books he keeps re-opening, from '機械学習 100+ページ エッセンス' by Andriy Bu…

The article lays out the split in Claude Code — CLAUDE.md is the file the user writes with instructions and ru…

A solo founder audited his Claude Code Auto-memory store of 41 memory files

The author calculates one AI prompt costs about 0.01 yen in electricity — ChatGPT at 0.34 Wh and Gemini at 0.2…
