You've been the project manager of every Claude project. Now Claude does it.
On this page
Claude Projects used to organize your work.
Now they orchestrate it: you brief one project manager and it runs the team.
👀 This is a sneak peek at a beta feature
🧭 A clear picture of what the new Projects look like, what changes and what stays the same
📬 One project set up with a goal, your files and your standing rules
🧩 A first piece of work handed off and split into threads that report back
🔁 A routine that runs on a schedule while you're away
🙋 For you if you're on Pro or Max and you have ongoing work that keeps producing tasks
🚫 Skip it on a Team or Enterprise plan (not available yet, and projects can't be shared)
🚫 Skip it if you don't see Projects after the check below, and come back when it reaches you
🧭 Check you can open it today
First, check whether Projects (Beta) appears in the Claude desktop sidebar. Anthropic also lists it under claude.ai/code and the Code tab. It’s rolling out to Pro and Max accounts, with earlier access for accounts already using Claude Code in the cloud who don’t have older projects in Claude chat or Cowork.
- If you see Projects: carry on.
- If you don’t: the rollout hasn’t reached your account yet. Join the waitlist and come back; upgrading from Pro to Max doesn’t guarantee access.
- If you’re on Team or Enterprise: skip this page for now.
Your older Claude Projects keep working as they do.
This is a sneak peak into what projects will look like in the future.
🛑 Before you upgrade
📦 Your context moves with the project, so there is nothing to download or back up.
↩️ Anthropic hasn't said whether you can undo an upgrade. Treat it as one-way until you see otherwise.
🔀 Once upgraded, Chat, Cowork, and Code projects merge into one project view, shown below.
What you see after an upgrade: a welcome page that lists where everything went. The numbers here are an example.
🔀 What actually changed? From organizing to orchestrating
Old Claude Projects organized your work, and you still ran every chat yourself.
You were the project manager.
The new Projects give that job to Claude.
You brief one coordinator and it splits the work into threads: workers on their own clean desks, side by side, each reporting back when done.
📤 You hand over the work → 🧩 Claude splits it into threads → ⚡ Threads run side by side → 📬 Results come back to you
🚀 The project is still where your AI employee lives, and this is the version that manages its own team.
⚠️ Be aware:📎 Upload = copy: Claude works from the uploaded copy, not the original file on your computer. ☁️ Cloud by default: Threads run in Anthropic’s cloud (unless you start local sessions on purpose), so they can keep running after you close your laptop. 💻 Want local files? Choose Work locally. Setup is in Advanced.
🗺️ What changed, and where everything is
| Before: organizing | Now: orchestrating | |
|---|---|---|
| Who starts the chats | You, one at a time | You talk to the coordinator, who starts threads for you |
| Work at once | One chat | Several threads side by side |
| Where results go | You copy them out of each chat | The Library tab collects your files and the ones Claude makes |
| What it reads first | Your instructions and files | Instructions and files, plus project notes every thread is given by the coordinator |
| When you leave | Work stops when you stop | Cloud threads keep going when you close your laptop |
| Where you run it | The chat window | Browser, desktop app or phone, and you can check on threads from your phone |
| Who can use it | Shareable on Team plans | One person, no sharing for now |
| Who has it | Everyone | Pro and Max, in beta |
💚 TUB’s favourite feature of the new Projects? The Library.📚 From the files you upload to everything Claude creates — all in one view.
Key terms, in the order you'll meet them:
- Goal: one line on what the project is for.
- Coordinator: the project conversation you talk to (also called the orchestrator). It decides what becomes a thread.
- Thread: one worker on one job, with its own workspace.
- Overview: the pane beside the conversation that shows every thread and what needs you.
- Library: the tab holding the files you added and the files threads made.
- Instructions: standing rules you write once, in the box on the Memory page (up to 16,000 characters). Every new thread follows them.
- Memory: your instructions, plus Auto memory, the notes Claude writes itself. Every thread reads the notes file,
MEMORY.md, and you can tell Claude to change or remove a note. - Routine: a saved prompt that runs on a schedule in the cloud.
- Models: in General, choose the model and effort for the coordinator and threads. Leave the defaults to start.
- Environment: where cloud threads run, with their network secrets, plus the folders on your computer Claude may use. Advanced, see the toggle at the end.
- Settings: the gear in the project header opens General, Memory, Environment, Usage and Debug access.
Now build yours.
🛠️ Create: give it a goal and files
A goal, a few files and a note about your business are all the project needs.
Imagine you post on Instagram and write a Substack. Under each step you'll see what you would type for that project. Build your own, or just read along. At the end there is an optional prompt that sets up the real thing.
Step 1: Open the New project form
In the left sidebar, click Projects, then New project.
Step 2: Name it and write the goal
Type a short name in Name. In Goal, write one line on what the project is for:
Goal line. Change each bracket and delete the "for example" part:
Reach [a number and what it counts, for example: 800 Substack subscribers] by [a date, for example: December 31]: [what this project handles, for example: plan my posts and check my numbers], checked [how often, for example: every Monday].💡 Tip. Pick a number a little above what you do today, so it stretches you. No numbers yet? Make the first goal "find my baseline": plan four weeks of posts and read the results. Keep the goal to one line and put the detail in your instructions.
Step 3: Add your files
Under Context, click Add, or drop files in.
Files to add: Upload anything the work depends on, such as past results, notes, templates or a list of who owns what.
Add key documents about your brand: who you serve, what you sell, how you sound.
Example (content check-in): your brand and voice doc, your content pillars and your latest analytics export.
Step 4: Create the project
Click Create project.
Claude greets you with a few ways to begin, such as Pick up recent work, Make a plan and Use my tools.
Pick one, or skip them and send your own request.
Next, tell it your rules.
📜 Brief: set the standing rules
Standing rules go to every thread before it starts, so you say them once.
Step 5: Open the instructions box
Click the gear icon in the project header, then Memory.
The box at the top holds rules every new thread reads and follows.
Step 6: Paste your rules
💡 Tip. Keep it short: a few clear rules beat a long list. The limit is 16,000 characters.
Paste this. Change each bracket and delete the "for example" part.
Standing rules
You are my [role, for example: weekly content planner].
You are helping [who and with what, for example: me, a coach, plan Instagram and Substack posts].
You sound [how, for example: short and plain].
Always read [your file, for example: my brand and voice doc] before any task.
Always show your source: quote the row or file you used.
Never guess numbers or names. If it isn't in my files, say "not in export".
Never post, publish or message anyone.
Work like this: propose threads and wait for my go-ahead. Run at most three threads at a time.
Good work here looks like [your standard, for example: a plan I could post from without rewriting].💡 Pro move. Make it the best project manager you've ever had. Add this under your rules and edit it to fit how you like to be managed.
Proactive project manager (example)
Act like an employee who wants the promotion.
Be proactive. After every task, tell me what you noticed and what you would do next.
Suggest work you could take off my plate, ranked by time saved. Don't wait to be asked.
Lead with the answer in one or two sentences. Then give the trade-offs, then what you need from me.
Use plain words. If a term needs explaining, explain it in one line.
If you're unsure, say so and tell me what would settle it.The better you brief it, the better it manages, and shaping that is your job.
Claude also keeps its own project notes, and you can ask it to remember or forget something in the conversation.
Want a different model? (optional)
Click the gear icon, then General. Under Models, set the model and effort for the coordinator and for new threads.
Threads do the work, so they start with more effort than the coordinator. Leave the defaults to start.
Now give it the work.
📤 Delegate: hand over the work
One messy request is enough. Claude decides how to split it. Type it in the project conversation.
Step 7: Send the one big request
Paste this in the message box (Ask Claude a question or start a task...), change each bracket, delete the "for example" part and send it:
Hand-off prompt
Here is the work: [describe it, for example: compare last month's Instagram and Substack numbers with my content pillars, find the three posts that worked best, and draft next week's plan]. This is a big job, so split it up however you think is best.Step 8: See the threads start
Claude splits the work and shows each thread as a card under your message, with a title and a status.
Click a card to open that thread.
Example (content check-in): the cards might read Instagram numbers, Substack numbers and Next week's plan. Your project may split the work differently.
👀 Review: check progress and results
Go and do something else. The work keeps running. Coming back later, start here.
Step 9: Open Overview
Threads take as long as the job needs, so come back later.
Click Overview in the project header. A dot on that button means a thread is waiting for you.
The Threads tab groups the work:
- Ready for review
- Waiting on you
- Working
- Landing
- Idle
- Resolved
Step 10: Answer what's waiting
Open a thread's card to read what it did, or to answer its question in the thread's own message box.
To steer one thread, type in its Steer this thread... box.
When Claude needs a decision, it can show a card with lettered options, one marked Recommended. Click one or reply in chat.
Step 11: Open what the threads made
Click the Library tab. The files the threads made are there. Open one.
💚 Why this matters: everything your threads build is in one place. Before, you hunted through old chats.
Example (content check-in): a what-worked file and next week's plan.
The Library tab: your uploads, folders and the files and artifacts Claude makes.
⏰ Repeat: put the work on a schedule
A routine is a saved prompt that runs on a schedule in the cloud, so it runs while your laptop is closed.
Step 12: Ask for the routine
In the project conversation, paste this:
Routine prompt. Change each bracket and delete the "for example" part:
Every [Monday], remind me to add the latest [files or notes, for example: Instagram and Substack numbers]. When I do, run this same work again.⚙️ Advanced: reach more than your files
Environment: where threads run, and what they can reach.
- 1. An API key. Use it when a tool has no connector but offers an API, so a thread can pull your numbers by itself instead of you uploading exports. Open Project settings, then Environment.
- Under Cloud, pick a cloud environment.
- Under Network secrets, click Add secret and enter your key. Values can't be viewed after saving.
- Never paste a key into a message.
- 2. Work locally. Use it when files must stay on your computer. Choose Work locally from the + menu beside the message box. One-time setup:
- In the desktop app, go to Settings > Claude Code. Turn on Use this computer from your phone and claude.ai, add the folder and keep the computer awake. It needs Claude Code v2.1.280 or later.
- Pre-approved folders (optional): in Project settings, then Environment, go to Local. Turn on Let Claude use your device and add a folder under Pre-approved folders.
🧰 Optional: build your own content check-in project
One prompt sets up the example for your own accounts.
Paste it into a new project and answer its questions. It stops after each of four steps and waits for you.
- Interview: where you post, how often, your numbers today and your target, your pillars, your brand doc and your hooks.
- Tools check: which tools Claude can reach and which numbers you have to upload.
- Library: your channel list, a what-worked file, your numbers and your project rules, each shown before it is saved.
- Routine and plan: a weekly or monthly check on your numbers, then your first content plan.
Master prompt
You are the setup assistant for my content check-in project. Your job is to set up the project so it can check my numbers and plan my content. Never post, publish or message anyone. Work in four stops and wait for me after each one.
STOP 1: INTERVIEW
Check what is already in the Library against this list: brand and voice doc, hooks or swipe file, channel list, numbers, what-worked log. Then ask me, in one message:
- Where do I post? Give me the link for each channel.
- How often do I want to post on each one?
- What number am I at today on each channel, and what number do I want by what date? Suggest a stretch target for each and tell me how you chose it. If I have no numbers yet, say so and make my first goal "find my baseline".
- Which metric counts as "worked" for me on each channel (for example saves, replies, opens or new subscribers)?
- Who do I serve, what do I sell, and what should my posts lead to?
- What are my 3 to 5 content pillars?
- How many hours a week can I spend on content, and which formats can I make?
- Do I have a brand and voice doc? If not, tell me to make one first (the Brand Voice Builder skill can do that) and do not invent my voice.
- Do I have hooks or posts I want to copy the style of?
- Which tool shows my numbers (Metricool, Meta Business Suite, Substack, YouTube Studio or another)?
- Where do I plan content (Notion, Airtable, a spreadsheet, nowhere)?
Tell me what is missing and wait.
STOP 2: TOOLS CHECK
For each tool I named, check whether you can reach it from this project. Tell me plainly: connected, not connected, or not sure. Never say you can read a tool you cannot reach.
- If my planner is connected, read it and tell me what you found. Change nothing.
- If my numbers are not reachable, give me two options: I upload an export to the Library, or I paste the numbers into the chat. Tell me which menu to use for each tool. Menu names change, so if a step does not match what I see, ask me what I see.
- Only if I ask for it: tell me whether you could pull numbers by controlling my browser, and what I would have to approve first. Do not start without my approval.
Wait until I confirm my numbers are in.
STOP 3: BUILD THE LIBRARY
Build these files and show each one to me before you save it:
- Channel list: each channel, its link, its goal with a number and a date, and how often I post.
- Numbers: one table per channel with date, title and the main metrics from my files. Use only what is in my files. Where a number is missing, write "not in export".
- What-worked log: my 10 best posts, ranked by the metric I named for each channel. For each one give the date, the title, the number and one line on why it may have worked. Mark that line "guess", because you see the post and not the audience.
- Project rules: my standing rules in this shape, using what I told you: You are... / You are helping... / You sound... / Always... / Never... Show them to me and I will paste them into the project's Memory page myself.
STOP 4: ROUTINE AND FIRST PLAN
Ask me whether I want the numbers check weekly or monthly, and on which day. Then create a routine that does this on that schedule: remind me to add my latest numbers, and when I do, update the numbers file and the what-worked log, and write the next plan.
Then write my first content plan for the next week or month: one row per post with date, channel, format, topic, pillar, hook direction and the row in my numbers that supports it. Respect how often I said I want to post and the hours I have. If my planner is connected, ask me before you put the plan there.
Finish with a short report: what is in the Library, what is still missing, what you assumed, and three blind spots you can see in my numbers.
RULES
Never invent a metric, a date or a post. Quote the row you used. If a file is empty or unreadable, say so and ask me for another. Keep your messages short and plain.Before you run it: have your brand and voice doc and one export from your analytics tool ready. No brand doc? Use the Brand Voice Builder skill first. Want to study competitors instead of your own numbers? The Instagram Content Playbook does that. CHECK: untested end to end. Does the routine get created from a message, can threads read CSV exports from the Library, and can Claude reach Metricool or Meta from a project?
⚠️ Watch Out
What if I don't see Projects in my sidebar?
The rollout hasn't reached your account.
Join the waitlist at claude.com/form/projects and come back.
What if my usage runs out fast?
Every thread is a full Claude session.
Open Usage in Project settings to see what each thread used, and ask Claude to run fewer threads at a time.
In General, Auto-continue when usage limits reset lets stopped threads pick up where they left off.
What if a thread looks stuck?
It may be waiting on a permission prompt inside the thread.
Open the thread and answer there.
What if Claude can't see my file?
Cloud threads can't read files that live only on your computer.
Upload the file to the project, or choose Work locally.
What if Claude says it can't save a note?
Memory may be off in your main Claude settings.
Turn it on, then ask again.
What if the first result isn't right?
First runs usually need a tweak.
Tell Claude what's off in the project conversation, and ask it to remember the fix.
✅ You're done when:
📚 Go deeper
The official documentation, checked October 19, 2026:
Official information checked October 19, 2026.
