How to set up Claude for your company

To set up Claude for your company, you choose a plan, agree safe settings, write instructions that apply to everyone, build skills for the jobs you repeat most, connect the tools people already use, and keep the originals in a library you own. Then you name an AI champion in each team and train people on their own work. We work through those pieces in that order, with your people.

The Claude details in this post come from Anthropic's own pages, as of October 2026.

Setup should also settle who approves a change to a shared skill, and how everyone gets the approved version.

Which Claude plan should you buy?

Anthropic's pricing page lists its Team plan for 2 to 150 seats. A Standard seat is $20 a person a month billed annually, or $25 billed monthly. A Premium seat is $100 or $125 on the same terms. On Team, each person's usage limit is their own, so one heavy user doesn't run everyone else out (Anthropic's Team plan page).

Enterprise has the same $20 seat price billed annually, with a minimum of 20 seats (50 if you buy through Anthropic's sales team). Its seat fee includes no usage, so you pay for each piece of work Claude does at API rates, Anthropic's pay-per-use prices (Enterprise plan page). Should a 40-person company choose Claude Team or Enterprise? works through that choice for one company size.

If some staff already pay for their own Claude accounts, decide what happens to those first. When someone upgrades a personal account to Team, a box marked "Keep your personal account separate" may appear, and left unchecked the account upgrades in place. Anthropic's Team setup guide says content that moves into an organization can't be moved back. What should you do before moving staff from personal Claude accounts to Team? covers the rest.

What should you settle before people log in?

Your IT team or outside IT company keeps sign-in, devices and security. We agree Claude's security settings with them before anything goes live. Without an IT partner, whoever holds Claude's Owner role decides. Team includes single sign-on, which lets people log in with their existing work account. Can your outside IT company (your MSP) run Claude for you? covers where their part ends.

Name two Owners. Anthropic allows one Primary Owner per organization, and only Owners and the Primary Owner can reach billing (managing members). A second Owner means billing isn't stuck when one person is away.

Then read three Team defaults:

• By default, anyone with an email address at your company's domain can find your organization and ask to join it. If joiners are approved automatically and no seat is free, Anthropic says a new seat is bought straight away (find and join).
• Publishing is set to Open, so any member can share a skill with the whole company and nobody checks it first. The other settings are Off, and Requires review, where an Owner approves each skill a member submits before it's published (managing skills).
• Cowork is the part of Claude that carries out tasks with several steps, using your files and tools. On Team, new Cowork tasks start in Automatically approve mode (off by default on Enterprise), so Claude doesn't stop to ask before each step (Cowork settings). Anthropic's Cowork guide still advises staying close when money, files or messages sent in your name are involved.

Each default suits some companies. Someone should choose it and write down why. One you can leave alone: the pricing page says Team does no model training on your content by default.

What goes into company instructions and the first skills?

Organization instructions are a short text Claude reads in every conversation, for everyone in the company. Only Owners and the Primary Owner can set them, the limit is 3,000 characters, and a change can take up to an hour to reach everyone (organization instructions). Because they go into every message, Anthropic advises keeping them short. Put in what should hold everywhere: who your customers are, how you write to them, and how to handle their data.

Skills carry everything else. A skill is a written recipe Claude follows for one job, stored as a folder of instructions and sometimes files, and Claude loads it only when a task needs it (what skills are). Start with the jobs people repeat most, such as proposals or weekly reports. What is a Claude skill? goes further.

Keep the original of every skill in a library your company owns, under version control. Version control keeps a record of each edit, with who changed what, when and why, so you can go back to any earlier version. Claude works from a copy. Amit's post on building a shared prompt library covers how he ran one under version control.

Then connect the tools people use daily, such as email and shared drives. An Owner turns each connector on for the company, each person still signs in to it, and Claude gets only that person's permissions in the tool (connectors). So Claude can't show someone a file they couldn't already open. The exception is a custom connector set up with one shared login, which reaches whatever that login can.

Who helps people, and who approves changes?

Train people hands-on, using the work on their desk that week. Give each team an AI champion too: someone who uses Claude well and whom teammates ask first.

Anthropic's tutorial on scaling Cowork suggests one champion per 25 to 50 users, a monthly check-in with them, a governance review every quarter, and a central team to own the skill library. Without a central team, the champions and a few decision makers share that work. We'd still have a champion in every team, because sales and finance ask Claude different questions. Amit's guide to champion networks covers how those programs run and how they fail.

Take an example: your office manager finds that the skill for chasing invoices gets your payment terms wrong. Shared skills are view-only for the people using them, so she can't edit it. She can write her own fixed version, and with publishing left Open she can share it with everyone. Now the company has two versions of one skill.

So decide during setup who approves a change. We suggest a small group, which we call an AI board and others call an AI committee or council: three to five people, usually a leader, the champions and someone from IT. They meet for about an hour, once or twice a month, and decide which fixes go in. Set publishing to Requires review, and every skill or new version a member submits waits for an Owner to approve it. After launch, the monthly loop takes over.

Some leaders would rather let people start now and agree rules for changes later. That can work for a few weeks. The risk is that private copies pile up meanwhile, and later someone has to pick between them.

Measuring belongs in setup too. Pick up to five jobs and time each one with the person who does it. Without that starting number, nobody can tell later whether Claude made the job faster. The same person can score a few of Claude's answers each month to see whether they're getting better.

Can you set Claude up without outside help?

Often, yes, if you're a firm of 15 or so and one person has the time and interest. Anthropic launched Claude for Small Business on May 13, 2026, a package of connectors and ready-made workflows for the tools small businesses already use. Its install guide says it works on Pro, Max, Team or Enterprise plans. Anthropic released a free online course, AI Fluency for Small Business, the same day.

Help makes more sense when several teams each need their own skills, when nobody has the hours to keep improving them, or when security settings need agreeing with an outside IT company. Amit's AI implementation checklist lists what to check before and after launch.

If you'd like help with any of it, tell us about your team.