How to set up Claude Projects for a business team

Which Claude Projects do you mean?

Projects in the Claude app are workspaces. Each one has its own files and instructions, and every chat in the project reads them. They are open to all users, and free accounts can have up to five. Anthropic explains them in What are projects?

Projects in Claude Code do more. A project is one ongoing conversation where Claude coordinates a stream of related work. You tell it what needs doing, and it starts a thread for each task. The threads run in parallel, and you can check on them from your phone. As of October 2026 they are in public beta on the Pro and Max plans, and not yet on Team or Enterprise. If you don't see Projects in Claude Code, there is a waitlist. The full reference is Anthropic's Projects documentation.

This guide is about the second kind, set up to run part of a business rather than a codebase.

The business we set it up for

An owner-led e-commerce and warehousing business in B.C. Leads came in by phone, web form and email. Nobody logged missed calls, and 73 follow-up emails sat drafted and never sent. The first fix on the table was to hire a salesperson. We set up a team of AI agents in Claude first. The full story, with every screen, is our November teardown.

Step 1: Connect your systems first

The agents are only as useful as what they can see. Connect your email first (Outlook and Teams, for this client), then your main business system. Put Claude on your phone so the reports reach you.

This client had a year of groundwork behind the setup: email connected in April, their ERP and a memory layer in August. That groundwork is why the team took about an hour to set up. In the director's words:

“The reason why we could do it so quickly is because of all the backend infrastructure we already built.”

Step 2: Create the project and name one lead

Create a new project in Claude Code. Only the name is required. Add the files, folders or repository the work needs; this client connected a shared repository so the lead had somewhere to keep plans and read handovers. You don't need to write code: every prompt in this setup is plain language.

The director named the lead conversation Alex and gave it one goal:

“You are my Chief of Staff, managing the agents under you on executing [the company] GTM. Your focus is to anticipate needs, understand the working engine. Manage our named agents by function to execute on our critical path and keep me updated.”

Then one rule for her: talk only to Alex. “Assume all you can see is Alex.”

Step 3: Set the approval rule once

The director told Alex early on that nothing goes out without her. When she handed it the missed leads, Alex repeated the rule back:

“Your earlier rule still stands: nothing gets sent, nobody gets called and no phone settings change until you approve.”

In a new setup, put a rule like this in the project instructions. Every new thread starts from them, so a rule stated once reaches all of them. They hold up to 16,000 characters. Treat them as instructions, not locks: read what the threads do in the first week.

Step 4: Ask the lead to build the team

She didn't set up the roles one by one. She sent Alex one message:

“Alex needs to define threads/named-agents to own GTM actions so they are not dependent on me end to end with PM, Dev mgr, Engineer, QA, DevOps, Site CRO, Sales Manager, Demand Gen Lead, Product Lead, Operations Lead, Reporting Lead, Strategy lead.”

Alex wrote a charter giving each role what it owns, what it decides alone and what it brings to her. It recommended starting five roles first; she started all 12. Over the next half hour she added Google Ads, SEO, Meta Ads, ERP and HR leads, and Alex added a definition-of-done lead that only accepts work after QA passes it.

Step 5: Hand over your old chats

A project can't read another chat's history. So each old chat wrote a handover file the lead could read, then she archived it. A good handover checks every open item against the live system, not memory, and groups each item under the role that now owns it, with its goal, where it stands, what was prepared and what “done” looks like. The full handover prompt is in the teardown.

Step 6: Ask for two daily reports

“I need two reports daily. Critical Path in the morning on what I need to focus on and what you need decisions on. And at noon my Sales and margin numbers and projections to target.”

Alex set both up for weekdays: the critical path by 8:30 a.m. and sales and margin at noon. In between, the team works.

Step 7: Run it from the Overview

The Overview pane sorts every thread by what it needs from you:

Most threads in “waiting on you” aren't stuck. As Alex put it: “Threads show ‘waiting on you’ when their last message lists a decision.”

What changed

On the day it went live, the leads list showed six leads the director could call right away, ranked, with what each one had asked for. We are measuring the rest over 30 days: leads per week by channel, the share answered within one business day, callbacks made, and whether the sales hire is still needed. We'll report those numbers, not guesses.

What we'd tell another owner

• Connect the systems first. The team is only as good as what it can see.
• Talk to one lead. Let it route the work to the roles.
• Set one approval rule. Nothing goes out until you say so.
• Ask for recommendations, not tasks. “Need a recommendation on…” gets you options with a pick.
• Ask for reports on a schedule. Decisions come to you instead of you chasing them.

Every prompt and template from this setup, including the handover prompt and the sales-manager instructions, is in the teardown on our newsletter.