If you’ve been using Claude for a while, there’s good news: Projects just got a lot more capable.
You might remember the old version. A project was a folder where you dropped some files, wrote a set of instructions, and kept related chats together. It was helpful for keeping things organized, but the problem was that you still had to manually prompt each chat inside it.
So Anthropic rebuilt it inside Claude Code. A project is now one ongoing conversation that can start several threads at once. Each thread is its own Claude session working on one specific task, and they all report back to that main conversation (coordinator) when they’re done.
Think of it as an agent orchestrator that can deploy and manage multiple threads at once so you don’t have to. Instead of feeling confused about the progress of each thread, you can clearly see how each one is progressing while your main agent orchestrator keeps you updated across them all.
Now that projects have been expanded, they include more features: the instructions every thread follows, a memory of the decisions you’ve made, a library of the files and artifacts Claude produces, routines that run on a schedule, the model each thread uses, and how much of your token usage it’s spending. You can check on it from the desktop app, the web, or your phone.
Here’s a glimpse of how it works:
As explained in the video, the new Claude project runs in the cloud rather than locally on your computer. This means your computer doesn’t need to be awake for the agent to execute any tasks. You can ask it to create slides, check your email, review Slack messages, manage your calendar, build artifacts, connect to other apps, and more.
That’s a lot of new pieces, so in this post I’ll walk you through each one using a single example: my content workflow. One folder with everything Claude needs to know about my newsletter, one project, and a topic that becomes a newsletter draft, a LinkedIn post, and a batch of Substack Notes.
The content workflow is just the example, though. If your work has several connected pieces, like weekly research, client deliverables, or course material, you can follow the same steps.
Let’s start with what’s actually inside a project.
What’s Inside a Claude Project
Before we build anything, here’s a quick tour of the features. You’ll use each of them in the walkthrough, so it helps to know what they can actually do for you:
The project conversation. This is the one chat you talk to. You tell it what needs doing, and it decides what to answer right there and what to hand off as a thread. Think of it as the person running the project. It keeps track of every thread it started and reads what they report back, without following every step they take.
Threads. These are the workers. Each thread is a full Claude session with its own task, and several can run at the same time. They run in the cloud by default, so they keep going after you close your laptop (you can also ask for a thread to run on your own computer when a task needs your local files or tools). Threads are also where your connectors work. The project conversation doesn’t have any, so anything that needs Gmail, Notion, or Drive gets done in a thread. An Overview pane next to the conversation shows which threads are working, which are done, and which are waiting on you.
Instructions and memory. Instructions are the standing rules every new thread starts with, so you only explain things once. If you’re familiar with Claude Code, this is your
CLAUDE.md. Memory is what Claude learns as you work: a decision you made on Tuesday, a mistake you don’t want repeated, how often you want updates. Claude writes these notes when you ask it to remember something (and sometimes on its own), and you can read and edit them in the project settings.The Library. Every file you add and everything the threads produce ends up here. In my content workflow that means the outline, the draft, the LinkedIn post, and the Notes all sit in one place instead of scattered across chats.
Routines. Work you want done on a schedule, like a Monday morning news scan for post ideas. Ask for it in the project conversation and the routine lives inside the project, running as its own threads.
Models and usage. You can pick a different model and effort level for the project conversation and for the threads. This is important because every thread is a full session, so a project uses your token faster than a single chat does. The usage view shows exactly where the tokens went, broken down by thread and by model.
Your phone. Projects work in the Claude mobile app too. You can check on threads, open what they made, and send new instructions from anywhere.
Who can use it right now
The new Projects are in public beta on the Pro and Max plans and still rolling out, starting with accounts that have used Claude Code’s cloud sessions. Team and Enterprise plans don’t have it yet. You’ll find it under Projects in the sidebar in the Claude Code tab of the Claude desktop app, or in the mobile app. If it’s not there yet, you can join the waitlist.
I’ve been using this for a week, and I can see how it could help more people build a centralized hub for their projects—one that can be accessed anytime, even through a cloud session, without requiring your computer to be awake. All without the complexity of using Claude Code, because the whole experience feels like using Claude Chat, just with more advanced features.
The Example We’ll Run
The example in this walkthrough is a project that runs my content workflow. I give it a topic, and it turns that topic into an outline, then a newsletter draft, then a LinkedIn post and a batch of Substack Notes from that draft.
Everything the project needs to know about my newsletter lives in one folder:
newsletter-operation/
├── CLAUDE.md what to read for which task, the rules that never bend
├── about/
│ ├── newsletter.md what the newsletter is, offers, current topics
│ ├── audience.md who reads it and what they're stuck on
│ └── voice.md how it should sound, what's banned
├── guides/
│ ├── newsletter-post.md
│ ├── linkedin-post.md
│ └── substack-notes.md
├── examples/ a published post, so Claude can hear the voice
├── ideas/ post ideas waiting to be written
└── output/ where drafts, LinkedIn posts, and Notes landI picked this workflow because it shows off what a project is good at. Some of the work can happen at the same time: researching the topic and checking what I’ve already written about it don’t depend on each other, so two threads can handle them at once. Other work has to wait its turn—for example, the LinkedIn post and the notes need the finished draft first.
The same pattern can also apply to any type of work. For weekly research, a few threads can each dig into a different source at once, and the summary waits until they’re done. For a client deliverable, one thread can read the meeting transcript while another runs research, followed by reporting, and then a final thread generates the slide deck from it.
When you set up your own, you’ll start from your own folder, and two things will help:
Your project instructions. A project has a set of standing instructions that every thread reads before it starts, and they work a lot like a
CLAUDE.mdfile: they tell Claude what the project is for, where things live, and what rules to follow. Writing them from scratch is the part most people get stuck on, so I’ll give you a shortcut for that.Your folder. You don’t need anything technical here: just the folder you already work from, with your notes, guides, and past work in it. When you add it to a project, Claude copies it into the cloud, and every thread works from that copy. That’s what lets the project keep running after you close your laptop.
(If you already keep your work in a GitHub repository, you can connect that instead. Most people reading this won’t need to, so I’ll stick with a plain folder.)
What the Project Looks Like Mid-Run
Here’s what the project looked like in the middle of the run.
The new project can be accessed inside the Claude Code tab. The left sidebar remains the same, but the middle screen is where your main project chat happens. This is where the agent orchestrator can deploy multiple threads, which can be accessed in the right-side pane. We’ll go through this in detail later.
Before we go further, a few notes.
🚨 Since this requires understanding how to build a folder structure as part of your context management that Claude can access, I suggest starting by reading these posts:
The Complete Guide to Building Your First Agentic System (That You Can Actually Trust)
From Blank Folder to Working System: How to Set Up Any Project in Claude Code
The Complete Guide to the Context Folder That Changed How I Work With AI Agents
The Ultimate Guide to Turn Claude Into Your Brain’s Most Valuable Co-Worker
From there, you’ll understand how a folder structure can be used to build context that AI can follow.
Projects also use your plan faster than a regular chat. But I’ll show you some tips on how to reduce token spend, how to read the usage view, and which settings to change so a project doesn’t eat your whole week.
What’s Inside This Walkthrough
The starter folder: the same structure as my content folder with every file ready for you to fill in. Add your own details, connect it to a project, and once your instructions are in place, every thread knows what it’s working on from the start.
🎁 My
CLAUDE.mdgenerator skill: point it at your folder and it reads what’s there and writes your project instructions for you. Keep it as theCLAUDE.mdin your repo so every thread loads it automatically, or paste it into your project settings (they hold up to 16,000 characters).The full walkthrough, prompt by prompt: briefing the project and watching it open threads, working with artifacts in the Library, teaching it with memory, steering from your phone, and setting up a routine. Every prompt I used, plus what to check when a thread comes back.
Usage settings that keep your plan intact: how to read the usage breakdown and the model and effort choices I’d make for each kind of thread.
🎁 Video walkthrough of the whole run, from an empty project to a finished draft, LinkedIn post, and Notes.
By the end, you’ll have your own project running in the cloud: hand it a topic on Monday, check in from your phone, and come back to drafts waiting in the Library, built from your own folder and your own rules.
Let’s set it up.
🚨 The full video walkthrough is available at the end of this post. I recommend reading the written post before watching the video 🚨
Step 1: Create Your Project
Open the Claude desktop app, go to the Code tab, and select Projects in the left sidebar. Then click New project.
A small window pops up with three fields.
Name. The only required field. This is how the project shows up in your sidebar, so keep it short and specific. I called mine AI Maker Content.
Goal. One line describing what the project is working toward. Claude in the project conversation uses it to decide what to work on, so make it concrete. Mine:
Turn each week's topic into a newsletter post, plus a LinkedIn post and a batch of Substack Notes from that post, with me approving each piece before it's final.It’s optional, and you can change it later in the project settings, but I’d fill it in.
Context. This is where you give the project something to work from. Click Add and you’ll see four options:
Add repository: a GitHub repository
Add a file: a single file
Add a folder: a folder from your computer
Add a Google Drive folder: a folder from your Drive
We’ll start with Add a folder, because it’s the fastest way in.
One catch: the project won’t take an empty folder. You need a folder that already has real work in it. If you used the old Claude Projects, the files you uploaded there are a good start: your notes, briefs, past drafts, style guides, whatever you kept feeding Claude. If not, gather a handful of files from the work you want this project to do. For my content workflow, that’s my audience notes, voice guide, format guides for each platform, a published post, and a few post ideas.
A few things to know about adding a folder:
It’s a copy. If you edit a file on your computer afterward, the project still has the old version until you upload it again. To update it, open the Library tab, click Add, add the file again, and choose Replace when it asks.
There are limits. A folder upload takes the first 100 files, up to 200 MB, and skips any single file over 30 MB. That’s plenty for a folder of Markdown files.
Hidden files don’t come along. That includes a
.claudefolder, so any skills stored inside your folder won’t be in the project yet. If you rely on a skill, enable it on your Claude account level instead. Fortunately, if you’re a paid member, you can access Maker Labs plugins, so you’ll be able to use all the skills I include there.No CLAUDE.md yet? That’s fine too. I’ve added it in the Maker Labs plugins, so we’ll create one in Step 2.
Click Create project.
Your first look at the project
You’ll land in the project conversation: the one chat you’ll talk to for everything in this project. The Overview pane on the right is where threads will show up once there’s work happening.
If this is your first project, don’t be surprised if Claude starts talking before you do. On a brand-new first project, it takes a turn on its own and may post Setup recommendations or suggest a few threads it could start.
Don’t send it any work yet. Every thread you start from here reads the project’s instructions first, so it’s worth setting those up before the first real task. That’s the next step.
Step 2: Build Your Project Instructions
Now that the project exists, click the gear icon in the project header to open Project settings, then go to Memory. You’ll see a box called Project instructions, with a note underneath: “Like a CLAUDE.md: instructions and rules you write that every new thread reads and follows.”
Yes, it works like CLAUDE.md, as the brain of your project. Don’t know what CLAUDE.md is yet? Read this post.
You have a choice: write your instructions in that box, or write them in a CLAUDE.md file inside your folder. Both work the same way, but I put almost everything in the CLAUDE.md for two reasons.
It lives in your folder, meaning it travels with you: the same file works in a local Claude Code session and in ChatGPT as AGENTS.md.
And you don’t have to write it yourself, because you can ask Claude to do it for you.
Now that you understand the difference, let’s build your CLAUDE.md.
Generate your CLAUDE.md with the skill
The skill is called claude-md-generator. It scans your folder, reads the files that explain your work, and figures out four things before it writes anything:
Work: the repeated job the folder exists for (in my case, turning a topic into a newsletter post and its social pieces)
Input: the material the job starts from (my audience, voice, and format files, plus the idea I’m writing about)
Output: what should exist when the job is done (a draft, a LinkedIn post, a batch of Notes, saved in the right folders)
Key Standard: the one requirement the output has to pass (for me, the voice rules and the banned list)
The good news isclaude-md-generator has been updated into Maker Labs plugin 🎁
Here’s how to run it:
Install Maker Labs plugins into your Anthropic account
If you’ve already installed it, check for updates.
Open Claude inside your project.
Ask Claude to run this: Maker Labs’s
claude-md-generatorand send it.
I wrote about this skill in details in my previous post. Install it first before continuing this post.
The skill will help you create your first CLAUDE.md file, which serves as the main set of instructions for your agent to follow consistently across multiple threads and sessions.
Keep a few lines in the project instructions box
There’s one reason not to leave the box completely empty. According to Anthropic’s docs, project instructions go to every thread and to the project conversation, the one that decides what becomes a thread. Your CLAUDE.md gets loaded by the threads. So rules about how the project should run belong in the box, where the project conversation is sure to see them.
Here’s everything I put in mine:
