One Claude: The Merge, Claude Docs & Claude Slides
On September 16, 2026, Anthropic collapsed three products into one. Chat, Cowork (the agentic workspace) and Claude Design stopped being separate places you had to choose between. There is now one conversation box, and Claude decides what the task needs.
The official framing is short: "Ask Claude for what you need, and it can decide which tool to use." The practical consequence is longer, and it is what this page is about — because a merge like this moves settings, retires habits, and ships two new document tools whose limits are not obvious until you hit them.
- Understand what actually merged on September 16, 2026 — and why 'no mode to choose' changes how you open a task
- Find every setting that moved: global instructions, web search, research, and file storage
- Pick correctly between Claude Docs, Claude Slides, Claude Design, a plain artifact, and a downloaded .docx/.pptx
- Use Claude Docs the way it's designed: tabs, inline comments, @Claude edits, and real-time co-authoring
- Know the eight limits that will bite you — no version history, no comment-only role, charts that don't refresh, and the compliance configurations where Docs is simply unavailable
- Predict the usage-limit cost of agentic work now that everything runs in one conversation
What merged, in one picture
Nothing you had is thrown away. Per Anthropic: "Existing Cowork tasks, projects, connectors, skills, artifacts, and files carry over, and those Cowork tasks open as they did before, so you can continue them."
What changes is the entry point. Previously you decided the shape of the work before you described it — chat for a question, Cowork for a project. Now you describe the work and Claude picks the machinery, with the same context, skills and connectors available from any conversation.
Where everything moved
This is the part that generates the most "where did my button go" confusion, and it is the part no third-party write-up covers. Four things relocated:
| What you used | Where it is now |
|---|---|
| Global instructions | Settings → General |
| Web search toggle | Gone — it's automatic. Claude decides when to search |
| Research | Type /deep-research, or use the + button |
| Your files | Settings → Storage folder |
- The web-search toggle is not hidden, it is removed. If your workflow depended on forcing search off for a private or offline-reasoning task, that lever no longer exists in the UI — say it in the prompt instead ('answer from the conversation only; do not search the web').
The three templates: which one do you actually want?
The merge shipped Claude Docs and Claude Slides as new tools, and moved Claude Design inside conversations. All three are templates — a starting shape for an artifact, not a separate app.
- Rich-text documents saved to your Claude account, with headings, tables and other formatting, and more than one tab (like sections of a notebook). You and Claude and your colleagues all write in the same document, at the same time. Exports to Word (.docx), PDF, Markdown (.md) and Google Docs. Reach for it when the document itself is the deliverable and it will keep changing.
- Decks built from your notes, a report, or the work already in the conversation. You can edit any slide directly, present without leaving Claude, and export to PowerPoint and PDF. Reach for it when you need a deck out of material you already produced.
- Visuals, mockups, prototypes, one-pagers and landing pages, built against your design system. Exports as a .zip, PDF, PowerPoint, or standalone HTML. Reach for it when the output is visual or needs to match a brand.
- Still there, unchanged, on every plan including Free. A runnable mini-app, a chart, a diagram, a script. Reach for it when the thing needs to execute rather than be read. See Artifacts.
- A one-shot file Claude builds and you download. No live editing surface, no sharing, no comments. Reach for it when you need a file to hand off and never touch inside Claude again. See Generating Real Files.
The distinction that matters most is the last two: a Claude Doc is a living document with a URL, collaborators and comments; a generated .docx is a download. If you asked for "a Word doc" pre-merge and got a file, asking the same way now may get you a Doc you then export — which is usually what you wanted, but it is a different object with different sharing rules.
Three ways to create one
- Plain language is enough — 'turn the plan we just worked through into a product spec I can share with the team.' Claude asks a few questions before it starts, then drafts the doc on screen.
- Select Output in the message box and choose a template. Use this when Claude keeps picking a different shape than you want.
- Pick a template from the gallery, then brief Claude. Useful when the document, not the conversation, is the thing you came for.
Brief a Claude Doc so the first draft is close
Create a Claude Doc: a one-page product spec for the export feature we just designed. Structure it in four tabs: 1. Summary — problem, proposed solution, who it's for 2. Requirements — numbered, testable, each with a rationale 3. Open questions — things I still need to decide, with your recommendation and the tradeoff for each 4. Rejected alternatives — what we considered and why not Rules: - Pull the decisions from this conversation. Do not invent requirements I did not state. - Where I was vague, leave a comment asking me the question instead of guessing. - Add a table comparing the three export formats we discussed. Ask me anything you need before you start drafting.
Claude Docs in depth
Editing, comments, and @Claude
Three editing paths coexist, which is the point of the tool:
- You type. Click into the doc and type. Changes save automatically.
- You ask. Request a change conversationally; Claude edits and explains what it did and why.
- You comment. Select text and leave a comment for your collaborators — and mention
@Claudein a comment to have Claude make that edit in place. Claude replies in the thread, makes the change, and explains it.
Claude also leaves its own comments while drafting, to explain a choice or ask you a question. This is the behavior worth designing your prompt around: telling Claude to ask in a comment rather than guess (as in the prompt card above) converts silent hallucination risk into a visible question you can answer.
You can also ask Claude to add a chart, diagram, graph or timeline to a doc.
Collaboration and sharing
People with edit access work on the same doc at the same time as you and Claude, and everyone's edits appear in real time.
| Viewer | Editor | |
|---|---|---|
| Read | ✅ | ✅ |
| Edit | ❌ | ✅ |
| Comment | ❌ | ✅ |
| Export | ❌ | ✅ |
Sharing rules differ sharply by plan:
- Pro / Max — share with anyone who has the link.
- Team / Enterprise — invite specific people or a group, or share with everyone in your organization. Docs cannot be shared outside your organization, by link or by email invitation.
- Either way — opening a shared doc requires a Claude account, and only the owner can rename or delete a doc.
Where you can use it
Docs work on web, desktop, Claude Code, and mobile — but the iOS and Android apps are view-only: no editing and no template selection from a phone. Plan around that if your review loop is mobile.
The limits that will bite you
Every one of these is documented, and every one surprises somebody in week one.
- There is no way to see or restore an earlier state of a doc. Anyone with edit access can overwrite anything, and your only recovery is an export you took earlier. For a document that matters, export to Markdown at each milestone until history ships.
- The roles are viewer and editor, nothing between. A reviewer who should comment but not edit does not exist as a permission — grant edit and rely on trust, or grant view and collect feedback elsewhere.
- A chart Claude added is a snapshot. Change the underlying numbers in the doc and the chart keeps showing the old ones until you ask Claude to regenerate it. Treat every visual as stale until re-asked.
- No link sharing, no email invitation outside the organization. If you need to send a doc to a client, export it (Word, PDF, Markdown, Google Docs) and send the export.
- iOS and Android can read a doc. They cannot edit one or pick a template.
- Claude Docs is not available for organizations configured with customer-managed encryption keys, zero data retention, or HIPAA. This is the single most important fact for regulated teams, and it is not a rollout delay — it is an exclusion.
- Activity inside a doc is not recorded in the Compliance API. If your audit posture depends on that export, doc activity is currently a blind spot.
- If your team used the standalone site, migrate design systems from the Artifacts page before then — chats, comments and public project links from the standalone version are deleted when it closes. Full checklist on Dashboards & Motion.
- Docs, Slides and Design are now on by default everywhere, including Enterprise as of October 15, 2026. The two beta templates are not: Dashboards and Motion are off on Enterprise until an owner turns each one on in organization settings, under Artifacts.
Also not supported yet in the merged experience
The unified experience does not yet support GitHub integration ("Add from GitHub"), conversation branching, search across older Cowork tasks, Dispatch (existing users only), or incognito chats for file and code creation. If your workflow leans on branching or on searching your Cowork history, that capability is temporarily gone — not moved.
What this costs you
There is no separate price. The relevant sentence is about usage limits: "Everything you do with Claude counts toward your plan's usage limits. Longer agentic tasks that search the web, run code, or create files generally use more than a quick question." Claude Docs counts the same way.
The merge makes this sharper than it sounds. Previously, opening Cowork was a deliberate act — you knew you were starting expensive work. Now a casually-phrased request in the same box can trigger a long agentic run that searches, executes code and writes files. The cost signal moved from the door to the prompt. If you are on a plan where limits matter, say what you want explicitly:
Keep a request cheap when you only want an answer
Answer from this conversation only. Do not search the web, do not run code, and do not create any files or artifacts. Two paragraphs.
And the inverse, when you do want the full machinery:
Ask for the expensive version on purpose
Take this as a full task, not a chat answer. Research the three vendors below, check their current pricing pages, build a comparison spreadsheet, and turn the recommendation into a Claude Doc I can share with the team. Cite every figure with a link and the date you read it. Vendors: [A], [B], [C]
Migrating your own habits
- They moved to Settings → General. Open them once and re-read them: instructions written for chat-only Claude ('you cannot browse the web', 'ask before doing anything long') can now actively fight the merged behavior.
- It's gone. Put search intent in the prompt — either 'check current sources' or 'do not search'.
- Type /deep-research or use the + button. It is a command now, not a mode.
- Scheduled tasks, connectors and skills carried over, but prompts that said 'this is a Cowork task' as a signal to work harder no longer signal anything. Replace that framing with explicit scope, as in the prompt cards above.
- Until version history ships, pick a milestone rhythm — export to Markdown whenever the doc reaches a state you would be upset to lose.
Terms worth knowing
Quiz
Check yourself
0/4- One conversation, no mode to choose: the cost signal moved from the door (opening Cowork) to the prompt. Say explicitly when you want a cheap answer and when you want the full agentic run.
- Four things moved: global instructions → Settings → General; web search → automatic, steer in the prompt; research → /deep-research or +; files → Settings → Storage folder.
- A Claude Doc is a living document with collaborators and comments; a generated .docx is a download. Choose deliberately — they share rules differ completely.
- Design your prompt around Claude's comments: tell it to ask in a comment rather than guess, and you convert invention into a visible question.
- Two limits deserve a workaround today: no version history (export to Markdown at milestones) and charts that never refresh (regenerate after every data change).
- Regulated teams: Claude Docs is unavailable under CMEK, ZDR and HIPAA, and doc activity is not in the Compliance API yet.
Sources & further reading
- Anthropic — Claude Cowork and chat are now one Claude (announcement, September 16, 2026)
- Anthropic Help Center — Claude Cowork and chat are one Claude (rollout, relocations, what carried over, current gaps)
- Anthropic Help Center — Get started with Claude Docs (creation, tabs, comments, sharing matrix, exports, limits)
- Anthropic Help Center — What are artifacts and how do I use them? (the three templates and their plan availability)
- Related on AILmanac: Artifacts · Cowork Scheduled Tasks · Generating Real Files · Projects · Plans, Limits & Billing
Next
- Claude Dashboards & Claude Motion — the two templates added on October 8, 2026, and the Claude Design migration deadline
- Artifacts: Live, Runnable Outputs — the container every template renders into
- Cowork Scheduled Tasks — the recurring-task machinery that carried over intact
- Generating Real Files (docx/pptx/xlsx/pdf) — when you want a download, not a living document