AI Codex
Tools & Ecosystemupdate

Claude Docs and Claude Slides: collaborative documents and decks, in beta

In brief

Announced September 16, 2026 alongside the Cowork merge. Ask for a document and you and Claude write it together in a live editor. Ask for a presentation and Claude drafts the slides — editable in place, presentable from Claude, downloadable as PowerPoint or PDF. Beta, paid plans, and Enterprise admins decide when it turns on.

8 min read·Artifacts

Contents

Sign in to save

Claude has been able to produce a document for a long time. What it could not do was let you work in that document. You got a block of markdown or an artifact, you copied it into Google Docs or PowerPoint, and every further change meant going back to the conversation, asking for a revision, and copying again.

On September 16, 2026 Anthropic shipped Claude Docs and Claude Slides in beta, alongside the Cowork and chat merge. Both are editable surfaces that live inside the conversation.

Claude Docs

Ask for a document and you and Claude write it together. You edit directly in the document; Claude edits alongside you. The context of the conversation — the files you attached, the connectors that are wired up, the skills your org has installed — is available to the document as it is being written, which is the part that copy-paste always lost.

The practical difference from the old flow: a revision is an edit, not a regeneration. When you ask for a tighter third paragraph, you get a tighter third paragraph rather than a new draft of the whole thing that you then have to diff against the version you had already fixed.

If your team lives in Google Docs, this does not replace that — see Claude and Google Docs for the connector-based workflow, which is still the right answer when the document has to end up in Drive with your existing sharing and comment history.

Claude Slides

Ask for a presentation and Claude drafts the slides. From there you can:

  • Edit directly in the deck
  • Present straight from Claude, without exporting first
  • Download as PowerPoint or PDF

The PowerPoint export matters more than it sounds. A deck that lands as .pptx goes into your organization's template, gets edited by someone who has never used Claude, and survives the review process that every real deck goes through. A deck that only exists inside an AI tool does not.

Claude Design moves into conversations

Claude Design — prototypes, mockups, one-pagers — was a standalone product. It is now available from inside any conversation as well. If you use Design on its own, that keeps working exactly as before. Nothing was taken away.

The combined move

The example Anthropic gives is one request producing two artifacts: "Write it the way we always do, flag anything that slipped, and put the highlights in five slides." That produces a document and a matching deck from a single conversation, with the same underlying context.

This is the thing that was genuinely hard before. Producing a document and a deck that agree with each other used to mean producing them separately and then reconciling them by hand, which is where the numbers drift apart.

Availability and the admin question

All three — Docs, Slides, and Design-in-conversation — are in beta on paid plans. Enterprise admins choose when to turn them on.

That last clause is the one to act on if you administer a Claude org. Beta features that produce shareable, downloadable business documents are a different governance question from beta features that produce chat replies. Before you flip these on org-wide, be clear on:

  • Where the output goes. A .pptx downloaded to a laptop is outside every retention and DLP control you have configured inside Claude.
  • Who reviews it. A deck drafted in four minutes still needs the same approval as one drafted in four hours, and the speed makes it easier to skip that step.
  • What data the document can reach. The document inherits the conversation's connectors. Audit connector permissions before, not after — the approach is in connector best practices.

If you do not have a written position on AI-generated work products yet, the AI usage policy guide is a starting point that takes an afternoon rather than a quarter.

What to try first

Pick something you already produce on a schedule — a weekly update, a monthly review, a recurring client summary — and run one cycle of it through Docs, then ask for the five-slide version. A recurring deliverable is the right test because you already know what good looks like and how long it normally takes. A novel one-off tells you nothing.

Sources

Weekly brief

For people actually using Claude at work.

Each week: one thing Claude can do in your work that most people haven't figured out yet — plus the failure modes to avoid. No tutorials. No hype.

No spam. Unsubscribe anytime.

What to read next

Picked for where you are now

All articles →