• AI Fire
  • Posts
  • 🚨 Claude Code Just Got a HUGE Upgrade: Run an Entire AI Team from One Chat!

🚨 Claude Code Just Got a HUGE Upgrade: Run an Entire AI Team from One Chat!

Let AI agents split complex work into smaller tasks. Run multiple tasks in parallel instead of handling everything one by one. I'll walk you through how to set it up, what works, and more tips.

TL;DR

Claude Code projects lets you coordinate multiple AI agents through one cloud-based workspace. A coordinator assigns tasks to parallel threads, maintains shared context, and keeps work moving even when your computer is off.

You can use it for complex, ongoing projects that involve several connected tasks. Each thread handles its own work while the coordinator tracks progress and manages dependencies.

This guide covers project setup, model configuration, memory settings, and cloud environments. You'll also see how to assign tasks, monitor agents, resolve blockers, and review the final output.

Key points

  • Setup: Configure coordinator and thread models, project instructions, and cloud settings.

  • Execution: Claude manages parallel threads while you review progress and handle blockers.

  • Results: Our FlowPilot demo produced a launch report covering four workstreams and 42 tasks across five phases.

πŸ€– What's Your Biggest Struggle With AI Agents?

Login or Subscribe to participate in polls.

Introduction

Anthropic is rolling out Claude Code Projects (Beta), a new feature that lets you manage multiple AI agents working together on complex tasks. Here's what makes it exciting:

  • One AI coordinator breaks down your project and assigns tasks to multiple Claude Code agents.

  • Multiple agents work in parallel, handling research, coding, testing, and more.

  • Tasks keep running in the cloud, even when your computer is turned off.

  • One central workspace lets you monitor progress, review results, and manage your entire AI team.

Imagine assigning one project and having an entire team of AI workers handle different parts while you focus on other things.

I’ll show you how it works, including a real example of building an AI-powered dashboard with multiple agents.

I. What’s New with Claude Code Project?

Claude Code project lets you manage multiple AI agents working together in the cloud, even when your computer is turned off.

You give one AI coordinator a goal, and Claude splits the work into parallel threads. Each thread has its own session, maintains context, and shares information so everyone stays on the same page.

Unlike regular Claude Code subagents that finish after completing a task, sessions in Projects can keep working over time. You can check their progress, give new instructions, or step in when needed.

what-is-claude-code-project

Think of it this way:

  • Without Projects: you're the manager. You open sessions, pass context, check outputs, and restart when something stops.

  • With Projects: you write one clear goal. The coordinator assigns work to threads, tracks progress, and flags when it needs you, so you step in only when something actually needs a human decision.

That's a meaningfully different way of working.

II. Set Up New Claude Code Projects

I'll walk you through creating your first Claude Code project and setting up everything you need to get multiple AI agents working together.

  • Project Setup: Create a project and define its goal.

  • Models & Memory: Configure coordinator, threads, and project instructions.

  • Cloud Environment: Set up cloud access and local permissions.

Step 1. Create Your Project

Open Claude Code and look for Projects (beta) in the left sidebar. If you don't see it yet, check that your account is on Pro or Max, and that you've been admitted from the waitlist.

Once you're in:

  1. Click New Project and give it a name.

    Step 1. Create Your Project
  2. Add a Project Goal describing what you want to achieve. You can also upload relevant files for context. Both are optional at this stage.

  3. Click Create Project.

Projects work best for tasks with multiple connected workstreams, like researching, writing, and planning happening in parallel. For a quick one-off, a regular Claude Code session is usually faster.

For this guide, we'll create a project called SaaS Product Launch Manager.

Learn How to Make AI Work For You!

Transform your AI skills with the AI Fire Academy Premium Plan - FREE for 14 days! Gain instant access to 700+ AI workflows, advanced tutorials, exclusive case studies and unbeatable discounts. No risks, cancel anytime.

Start Your Free Trial Today >>

Step 2. Configure Models and Memory

Click the gear icon at the top of your project to open Project Settings.

Under Models, you'll find two separate settings worth thinking about:

Setting

What it does

Recommended approach

Coordinator Model

Manages task assignment and tracks progress

Use a capable model (like Opus 5.5 or Fable 5.1) at a lower effort level. The coordinator mainly orchestrates, it doesn't reason deeply

Thread Model

Handles the actual work: research, writing, planning

Use the same model at a higher effort level. Threads do the heavier lifting

configure-your-claude-code-project-settings

The coordinator doesn't need to do deep reasoning. It mostly assigns work and tracks status. Threads are where the real thinking happens, so give them a bit more headroom.

These effort levels are a sensible starting point rather than a required setup, so adjust them once you see how your project runs.

Next, go to Memory and add your guidelines under Project Instructions. These rules travel with every thread in the project, keeping outputs consistent even when agents work independently.

For SaaS Product Launch Manager, here's what I'd use:

You are coordinating the launch of a B2B SaaS product. 

Keep all deliverables aligned with the product's target audience and business goals. 

Delegate research, positioning, and marketing tasks to appropriate threads. 

Clearly label assumptions, flag missing information, and maintain consistent messaging across the project.
set-up-project-instructions-and-memory-in-claude-code

You can also enable Automatic Memory so Claude picks up important context as the project progresses. Each new thread then starts with that shared memory already loaded, and anything a thread learns feeds back in for later ones.

Finally, Projects can keep going across usage-limit resets, so a long project resumes where it left off instead of stopping mid-task and waiting for you to restart it.

Check the General settings for the continue-across-resets option and turn it on. Keep in mind that Projects also burn through your 5-hour and weekly limits faster than a single session, so running fewer threads at once helps.

Step 3. Configure Your Cloud Environment

Go to Project Settings β†’ Environment to choose where your Claude Code agents will work. You'll see two sections.

Cloud:

  • Project repositories: connect a GitHub repo if your project involves code

  • Select a cloud environment: choose a pre-configured cloud workspace if needed

For the SaaS Product Launch Manager example, the default cloud environment is fine. No GitHub connection needed.

Local:

  • Let Claude use your device: toggle this on if you want Claude to access files on your computer

  • Pre-approved folders: specify which folders Claude can use without asking each time

step-3-configure-your-cloud-environment

For this demo, I'll leave device access off and pre-approved folders empty, since we're not working with local files.

Use worktrees can stay off for most projects. It's mainly useful when you need separate Git workspaces for different threads working on the same repo at the same time.

That's it for the basic setup! Next, we'll give the coordinator its first task and see how Claude Code projects handles the workflow.

Did This Guide Help You Set Up Claude Code Projects?

Login or Subscribe to participate in polls.

III. Run AI Agents With Claude Code Projects

Now that Claude Code projects is set up, I'll give the coordinator a real task to see how multiple AI agents divide the work and collaborate.

  • Task Assignment: Give the coordinator a clear goal and specific deliverables.

  • Parallel Execution: Claude distributes tasks across multiple AI agents.

  • Progress & Results: Monitor threads, resolve blockers, and review the final output.

Head back to the main SaaS Product Launch Manager workspace and enter your prompt in the chat box at the bottom.

Since this project involves several connected tasks, I'll provide a fictional product, a clear goal, and specific deliverables so Claude has enough context to get started.

We're launching FlowPilot, a fictional B2B SaaS platform that helps small marketing teams manage campaigns, deadlines, and approvals in one workspace.

Target audience: Marketing teams of 5 to 30 people at US-based startups.

Main competitors: Asana, Monday.com, and ClickUp.

Pricing: $19 per user/month.

Launch timeline: 30 days.

I want you to coordinate a complete product launch plan using parallel threads where appropriate.

Please:

1. Research competitor positioning and identify opportunities to differentiate FlowPilot.
2. Develop our positioning, key messages, and unique selling points.
3. Create a 30-day marketing launch plan covering email, LinkedIn, and content marketing.
4. Prepare a launch checklist with priorities, owners, and deadlines.
5. Consolidate everything into a clear final launch report.

Keep all deliverables consistent, identify assumptions, and ask for clarification only when critical information is missing. Use illustrative estimates when verified data isn't available, and label them clearly. Don't publish content or take external actions without my approval.

Once you submit, the coordinator reads the goal and starts assigning work to threads.

You can track everything in Overview, or check the Threads panel to see what's finished, what's running, and what's waiting on you.

monitor-your-agents-and-review-results

In the FlowPilot test, Claude created four workstreams: competitor research, product positioning, marketing planning, and the launch checklist. Most finished on their own, but the positioning thread needed more product details before it could continue.

If you see a Waiting on you status, open that thread, answer the question or provide the missing context, and the thread picks back up.

Once the workstreams are complete, the coordinator consolidates everything into a final launch report.

Before acting on the results: check the sources and assumptions, especially any figures Claude labeled as illustrative estimates. Claude Code Projects does solid work, but numbers that weren't sourced from real data should be verified before they go into a presentation or a decision.

You can also pause the project at any time, or let it resume interrupted threads after your usage limits reset.

IV. Claude Code Project Demo Results

After running SaaS Product Launch Manager, Claude Code projects completed all four workstreams and combined everything into a FlowPilot Launch Report.

The report covers:

  • Competitor analysis of Asana, Monday, and ClickUp

  • Product positioning and key differentiators

  • A 30-day marketing plan across email, LinkedIn, and content

  • A launch checklist with 42 tasks organized across 5 phases

claude-code-project-example-flowpilot-launch-results

The coordinator here spotted where information was missing and asked for confirmation before finalizing anything.

The final report had no remaining blockers, and assumptions were clearly labeled so you know exactly what needs verification before the real launch.

Conclusion

Managing multiple AI agents sounds exciting, until you're the one checking every session, copying context across windows, and restarting things that stopped while you were asleep.

Claude Code Projects changes that. You hand one coordinator a goal, set the rules once, and let the threads handle the work. Our FlowPilot example went from a single prompt to a complete four-workstream launch plan with no manual session management.

If you're working on something complex and ongoing, like a product launch, a research project, or a multi-step content strategy, give it a try. Start with one clear goal, let Claude organize the work, and see how much coordination you can hand off.

If you are interested in other topics and how AI is transforming different aspects of our lives or even in making money using AI with more detailed, step-by-step guidance, you can find our other articles here:

Reply

or to participate.