Claude Is One App Now. Clear Out the Old Setup, Keep the Facts.
Chat and Cowork are becoming one Claude. Here is what to keep, what to clear out, and the one line to add to every big request.
By John Martines · October 23, 2026 · 6 min
Since September 16, Claude's Chat and Cowork have been merging into one app. The long instructions and style files many offices wrote for older models can now get in the way. Here is a short, safe clean-out, and the habit that replaces it.
Here is an example. Maya runs the office at a heating and cooling company. Last spring she spent an afternoon setting Claude up properly. She filled the instructions box with 600 words about the company, the tone she likes and the words she hates. She added a writing-style file she found online and a folder of notes on how to answer customers.
This week she asks Claude for a two-line text to a customer: the technician is running 20 minutes late. She gets back three formal paragraphs, a sign-off, and a reminder of the company's service area. Every rule she wrote is being followed. None of them were written for a two-line text.
If that sounds familiar, the fix is not more instructions. It's fewer.
On September 16, Anthropic started merging Claude's Chat and Cowork into one Claude. There's one message box, and Claude decides whether your request is a quick answer or a bigger job that keeps running in the cloud. It's rolling out to Pro and Max accounts first, with other plans after. On those plans there's no switch to flip: your account moves over, and it can't switch back to the old separate options. On Enterprise plans, an owner turns it on.
Your chats, projects, artifacts, connectors and skills come with you. Claude can use your local files, its built-in browser and your screen only while the desktop app is open.
There's also a new setting in the message box that controls how much Claude does on its own. Auto, the default on Pro, Max and Team, keeps working without stopping to ask about each step, with safety checks before each action. Manual, the default on Enterprise, asks before it takes actions.
How it got here:
- July 24, 2026: Anthropic engineer Thariq Shihipar wrote that the team had been "overconstraining" Claude Code with its own instructions, and advised: "Keep your CLAUDE.md lightweight." (CLAUDE.md is the instructions file developers give Claude.)
- September 16, 2026: Chat and Cowork start merging into one Claude.
- October 8, 2026: Claude's Docs, Slides and Design came out of beta on every plan, including Free.
Most of the setup people built this year was written for older models that needed a lot of hand-holding. The newer models don't. Anthropic's own prompting guide for Claude Fable 5 warns that skills written for earlier models are often too prescriptive and can make the output worse.
Think of it like a new hire's first day. A one-page sheet with the prices, the service area and who to call for what is a gift. A 40-page binder on tone, banned words and how to sign off is something they'll follow exactly, even when it makes no sense.
What Claude can't guess is still worth writing down: your prices, your policies, your service area, who approves what. Rules about how to write are the part to cut.
| Keep | Clear out | |---|---| | A few lines of facts: what you do, who your customers are, two or three hard rules | Long paragraphs about tone, style and words to avoid | | A Project for each real job, holding the real documents (price sheet, policies) | Writing-style skills and files made for older models | | Memory, for everyday work | Folders of instruction files nobody has opened in months |
1. Read your instructions box. Open Settings and find "Instructions for Claude." (If you used Cowork, its global instructions now live here too.) Copy the whole thing into a document first so nothing is lost. Then cut it down to the facts Claude can't guess. Aim for something that fits on your phone screen. If an answer gets worse, put that one line back.
2. Turn off old style skills for a week. If you added skills about writing or tone, switch them off and work normally for a week. Turn one back on only if you actually miss it.
3. Keep one Project per real job. A Project for quotes holds the price sheet. A Project for hiring holds the job description. Facts in, rules about tone out.
4. Say the goal, then have Claude say it back. Instead of steps, give Claude three things: what you want, who it's for, and what good looks like. Then end every big request with this line:
> Before you start, tell me in your own words what you think I'm trying to do, and ask me anything you need to know first.
If Claude gets your goal wrong, you find out in ten seconds instead of after the work is done. We have a short tip on this habit: [Let Claude ask the questions first](/tips/let-claude-ask-first).
5. Check Auto and Manual. Auto is fine for drafting and research. For anything that sends a message, spends money, deletes something or reaches a customer, switch the conversation to Manual, or keep a person approving the last step.
6. Use Memory on purpose. Memory is on by default on Free, Pro and Max (Settings, then Memory). On Team and Enterprise, an owner has to make it available first. For a one-off job where your history doesn't matter, start a new chat, click the plus sign in the message box and turn Memory off before you send the first message.
7. Ask for a page instead of a file. When you want something you'll keep updating or pass around, like a checklist or a one-page plan, add "Make it an HTML page" to your request. Claude builds it as an artifact in your account instead of a file in your downloads. It starts private to you. The Share button sends a link, and everyone who opens it needs a Claude account, so it's a tool for your team, not your customers.
- Does your instructions box fit on one phone screen?
- Is every line in it a fact Claude couldn't guess?
- Did you try a week without old style skills before deciding to keep them?
- Does Claude restate your goal before it starts a long job?
- Is anything that reaches a customer or touches money on Manual, or approved by a person?
Anthropic is clearly heading toward fewer choices and less setup: one box, Claude choosing how to handle the request, your tools carried along automatically. That changes the skill that matters. A year ago the advantage went to whoever wrote the best instructions. Now it goes to whoever can say most clearly what they want and why.
So keep the facts, cut the rules, and make Claude repeat the goal back before it starts. Maya's late-technician text should be two lines again.
---
Sources (checked October 23, 2026). Chat and Cowork merging, the rollout to Pro and Max, no going back: "Cowork is now Claude" (Claude blog, September 16, 2026) and "Using Claude with chat and Cowork unified" (Anthropic Help Center, support.claude.com/en/articles/16761823). The Enterprise switch: Anthropic Help Center, support.claude.com/en/articles/17444462. Auto and Manual, what carries over, local files only while the desktop app is open: support.claude.com/en/articles/16761823. Docs, Slides and Design out of beta on every plan: "Dashboards and Motion" (claude.com resources, October 8, 2026). "Overconstraining" and "Keep your CLAUDE.md lightweight": Thariq Shihipar, "The new rules of context engineering for Claude 5 generation models" (Anthropic, July 24, 2026). Skills written for earlier models can degrade output: "Prompting Claude Fable 5" (Claude Platform docs). Instructions for Claude: Anthropic Help Center, support.claude.com/en/articles/10185728. Memory settings and turning Memory off for one chat: Anthropic Help Center, support.claude.com/en/articles/11817273.