Claude Cowork Merged: Anthropic Combines Chat and Cowork, Launches Docs & Slides

So, which Claude were you supposed to use today? The chat window for quick questions, or Cowork for the bigger jobs? If you hesitated, you’re not alone. The split confused plenty of people, and Anthropic just deleted it. Claude Cowork and Claude chat are now a single unified Claude: one conversation that decides for itself which capabilities your request needs. Alongside the merge came two new beta tools, Claude Docs and Claude Slides, plus Claude Design folded directly into conversations.
The company’s own explanation is blunt: people kept saying the frustrating part was deciding where a task belonged, and work started in one place didn’t carry into the other. So the choice is gone. Bring a quick question or a report due at noon; Claude takes it from there, even after you’ve closed your laptop. Nothing you built in Cowork is lost: tasks, projects, artifacts, connectors, and skills all carry over intact.

What the Claude Cowork Merger Actually Changes

Until this merge, Claude had two front doors. Chat was the familiar window for questions and drafts. Cowork was the agentic workspace, launched earlier this year as a research preview, where Claude could plan multi-step work, read and edit files, browse the web, and keep running in the cloud after you closed your laptop. The friction was never capability. It was triage. You had to classify your own task before starting it.

Now the triage is Claude’s problem. From any conversation, the model figures out whether a request needs a quick answer, a longer Cowork-style work session, or Design’s visuals, and it draws on the context, skills, and connectors already sitting in that conversation. The same thread can open with a question, expand into a project, and hand you finished files. For the machinery behind that kind of behavior, the planning loops and permission scopes that separate an agent from a chatbot, the same architecture is what our complete guide to AI agents breaks down in detail.

Three mechanics matter most. First, continuity: your existing Claude Cowork chats, projects, artifacts, connectors, and skills are exactly where you left them. Open the app and pick up where you left off. Second, check-ins: by default Claude asks before taking an action, and you can loosen that to longer stretches between check-ins, but you keep the final say. Third, scheduling: a recurring job like a weekly report can run on its own, and everything Claude makes lives at one shareable link you can open on your phone.

Claude Docs and Claude Slides Arrive as Betas

The merger is only half the announcement. Anthropic also shipped two new creation tools, both in beta on paid plans. Claude Docs is a document you write together with Claude: create sections, edit in real time, leave comments for Claude on specific parts, then export to Google Docs or Microsoft Word. Claude Slides drafts the deck: tweak individual elements, present straight from Claude, share a link, or download as PowerPoint or PDF.

Honestly, exporting straight to Word and PowerPoint is the smartest detail in the whole launch. Most offices aren’t abandoning their tools of record, and Anthropic clearly knows it. The beta framing stays deliberate too: enterprise admins choose when these tools turn on inside their organizations, with at least 30 days’ notice before anything changes.

Design Quietly Stops Living Apart

Claude Design, the visual website-and-prototype builder that launched as its own surface in April, now works inside any conversation. Ask for a website or a prototype and the visuals come back in the same thread. Prefer the standalone version? Per the announcement, it keeps working as before. Nothing forced.

This is the least dramatic of the three changes and possibly the most telling: the same logic that merged Claude Cowork into chat now pulls Design into the conversation too. No side quests to other tabs.

The 1.2 Million Sessions That Explain Why

Press releases explain what changed. Anthropic’s own usage data explains why the merge was inevitable. In an analysis of 1.2 million anonymized Claude Cowork sessions across more than 600,000 organizations, the largest usage category was business process and operations, pulling scattered updates into a single report, building onboarding checklists, reconciling spreadsheets, at 33.4%. Content creation and copywriting came next at 16.4%. Software development? Just 8.7%.

Read that twice. The agentic workspace built on the same architecture as a coding agent turned out to be mostly a tool for the work around the work: the connective stuff every office job involves but nobody lists on a resume. Anthropic is careful about the numbers themselves, the sample was rate-capped, so these are shares of sampled sessions rather than absolute volumes, and the labels came from an automated classifier, not human review.

Even with those caveats, the shape is unambiguous, and it’s why Claude Cowork didn’t survive as a standalone product. People weren’t reaching for Claude Cowork to do the core of their jobs. They were reaching for it to assemble and structure information so they could apply their own expertise. A merge that lets any conversation escalate into that kind of work without a mode change is the product catching up to how it was already being used.

The One-Way Trip: Rollout, and How to Tell Where You Stand

The rollout is staggered, and it doesn’t come with a switch. Pro and Max plans get the merged Claude Cowork first, on web, desktop, and mobile, over the coming weeks, for existing and new subscribers alike. Team and Free plans follow later, with no date attached. Enterprise admins get at least 30 days’ notice and decide when the new betas activate inside their organizations. There’s nothing to turn on. Anthropic moves your account for you.

Two accounts on the same plan can move at different times, so the fastest way to check is the message box: if it still shows separate Chat and Cowork options, your account doesn’t have the new experience yet. And once Claude chat and Cowork merge on your account, there’s no going back, Notebookcheck, which covered the rollout closely, reports that the old separate options don’t come back. Your work carries over, but the split is gone for good.

Reading through the 226-comment Hacker News thread on the announcement, I could see why the reaction has been genuinely split. One commenter, Sherveen, worried that merged modes blur a real distinction, the conversational chat experience handles research-type questions very differently from an agentic work mode, and collapsing them risks, in his words, “the regression of think-first-act-later AI UX.” Another, binlog, argued the opposite: presenting a toggle is “a terrible user experience,” and “the model itself should figure out how much power it needs to run the query.” Which of those resonates with you probably predicts how you’ll feel about the merge.

A few Claude Cowork migration details worth knowing: old Cowork tasks now sit in Recents alongside your chats, the Global instructions setting moved into Settings > General, and the web-search toggle is gone, Claude simply searches the web when it helps. Incognito chats still open in the previous experience, which means no file creation or code execution there.

Claude vs ChatGPT Work: Same Race, Different Starting Lines

The obvious comparison is OpenAI’s ChatGPT Work, launched in July as a chat-plus-Codex agent for office workers, powered by the GPT-5.6 model family. Both companies are betting that the next phase of competition isn’t benchmarks but workflow, who owns the window where work gets done. Fortune framed Anthropic’s move as a step toward an AI “superapp”: one place where casual chat evolves into documents, slides, designs, and autonomous agents. The superapp race has its ironies, some of the same industry figures pushing faster agents have also signed calls for restraint; for that story, see our coverage of the AI slowdown pledge.

The comparison, in plain terms:

The differences are real but shrinking. Claude’s edge is export honesty, Word, PowerPoint, Google Docs, and PDF out of the box, no proprietary format lock-in, while Work’s edge is breadth: spreadsheets, websites, and the Sites beta. Underneath, though, the bet is identical: stop making the user pick the tool, let the model route the work. Simon Willison, who has tracked this space for years, called the merge “hopefully good news” for anyone who, like him, was “increasingly confused at Cowork v.s. Claude v.s. Claude Code,” and added his hunch that “figuring out what this actually means in terms of features and surfaces is still going to take quite a bit of work.” That last line is the one to keep. The announcement is a paragraph; the product is a maze. For the broader three-way breakdown across the leading chatbots, see our Claude vs ChatGPT vs Gemini comparison.

What to Actually Do With the New Claude

If the merged Claude Cowork has reached your account, three things are worth doing in the first week.

One-way migrations always make me a little twitchy, and this one is no exception, if the routing guesses wrong on something important, there’s no toggle back to the old split. That’s the real price of the simplification: you trade a small, annoying choice for trust in the router. And trust, with a Claude Cowork session running in the cloud after you’ve closed your laptop, is the whole ballgame.

• Enterprise admin? You get 30 days’ notice and choose when the betas turn on, set your policy before they do.
• On Team or Free? The update arrives later, which is quietly good news: the rough edges get documented first.
• Keep internal Claude guides updated. The mode picker and the web-search toggle are gone from the new experience.
• Anything client-facing from Docs or Slides gets a human review. Betas earn trust the same way routers do.

And one trust note: autonomous agents with access to your files and connectors deserve the same skepticism you’d give any new hire with the same access. We’ve covered what happens when AI agents go off-script, our OpenAI misalignment piece walks through six such cases, and the lesson applies here too: start with the default that asks first.

Questions People Are Actually Asking