Claude Cowork and chat have merged: how to prompt the new unified Claude

On 16 September 2026, Anthropic announced that Claude Cowork and the regular Claude chat are merging into one experience, as described on the official Claude blog. Before this, you had to pick: chat with Claude, or hand a bigger job to Cowork. Now it's one conversation, and Claude decides how much work a request needs.

The change is rolling out gradually to Pro and Max plans on web, desktop and mobile, with Team and Free plans to follow, so you may not see it yet. If you do have it, here's what's different and how to prompt it.

What actually changed

• One mode, not two. You no longer open a separate Cowork window for a bigger task. You ask a quick question or hand over a full job in the same conversation, and Claude works out which it is.
• Claude Docs, Slides and Design are built in. These launched in beta on paid plans as part of the same update. Claude can now write a document, build a slide deck you can download as PowerPoint or PDF, or produce a design, directly inside a normal conversation, and each one creates a shareable link you can open from web, desktop or mobile.
• Everything from Cowork carries over. If you were already using Cowork, your chats, projects, artifacts, connectors and skills are all still there, in the same place.
• You can set how Claude checks in. You can tell it to ask before taking action, or to proceed on its own and only check in when it actually needs something from you.

Prompt for the whole job, not a mode

Because Claude now decides how much work a request needs, the biggest change to your prompting habit is to stop asking "quick question or Cowork job?" and just describe the finished result you want:

Goal: [WHAT YOU WANT, e.g. a one-page project plan for the office move]
Starting material: [WHAT TO USE, e.g. the notes and budget pasted below]
Format: [WHAT IT SHOULD LOOK LIKE, e.g. a doc with sections for scope, timeline and risks]
Check in with me: [WHEN, e.g. before contacting anyone outside the team, or not at all unless something's missing]

<material>
[PASTE YOUR NOTES OR DATA HERE]
</material>

The one-page project plan prompt is a good starting point for this kind of request: it already asks for scope, milestones, owners and risks in one message, which is exactly the shape of brief that works well now that there's no separate "Cowork mode" to switch into.

Tell it how often to check in

The check-in setting is new, and it's worth using deliberately rather than leaving on the default. For a task you want to keep an eye on:

Before you start, tell me your plan in one line.
Check in with me before: [THE POINTS WHERE YOU WANT A SAY, e.g. sending anything, choosing between two options, spending time on a part you're unsure about]
Otherwise, keep going and give me the finished result.

For a lower-stakes task, you can drop the check-ins and just ask for the result, which is closer to how Cowork worked before the merge. Either way, say which one you want. Don't assume the default matches what you'd pick.

Use Docs, Slides and Design without leaving the chat

Because these three tools are now part of the same conversation, you can ask for one directly instead of requesting "a document" and hoping Claude picks the right format:

• "Write this up as a Claude Doc I can share with my team."
• "Turn this into a Claude Slides deck, six slides, one idea per slide."
• "Make this a Claude Design I can send as a PDF."

Each one gives you a shareable link, so you can send it on without exporting a file first, or download it (PowerPoint or PDF for Slides) if the other person needs a file.

Follow a task from your phone

Because Cowork's cross-device behaviour carried over into the merged experience, a task you start at your desk keeps running, and you can check on it from your phone later. This is where the weekly personal review and time-blocked schedule prompts work well as something you start in the morning and check on later, rather than a message you have to sit and wait for.

For bigger, slower jobs, such as working through a research question over a few days, the sharpen your research question prompt is worth running first, so the task Claude is working on in the background is the right one before it spends time on it.

A note if you're on Enterprise

Anthropic says Enterprise admins get at least 30 days' notice before anything changes for their organisation, so if your Claude access is managed by your company, don't expect this straight away. Check with your admin rather than assuming the merge has reached you.

Where to go from here

If you're deciding whether to reach for Claude or another model for a task, see which Claude model you should use. For the agent side of things, including how to keep a standing task within its lane, see our guide on what OpenAI learned from shelving GPT-6.1 Astra. All the prompts above are free in the Claude prompts collection, and the prompt builder can turn a rough idea into a full brief for you.

Quick checklist

• Stop choosing a "mode". Describe the finished result you want in one message.
• Set your check-in preference on purpose: ask-first, or check in only when needed.
• Ask for a Claude Doc, Slides deck or Design by name when you want one.
• Start a longer task at your desk and check progress from your phone.
• On a managed Enterprise account, expect the change later than everyone else.

Prompts to try

Browse all 1594 prompts →