Claude training for teams: what Anthropic gives you free, and what it does not

Start with the free courses, they are good

Most vendors make you pay for the basics. Anthropic does not, and any honest provider should tell you that before quoting.

Its Develop AI Fluency collection at academy.claude.com carries three free courses. AI Fluency: Framework and foundations runs 14 lessons and a quiz, about four hours, and is the general grounding in working with a model. AI capabilities and limitations runs 13 lessons and a quiz, about three and a half hours, and builds a usable mental model of what a language model can and cannot do. Building effective human-agent teams is in beta at 5 lessons and around 45 minutes, aimed at teams moving from one person using AI to several people working alongside agents. All three were free when checked on 26 September 2026.

Send your team there first. It costs nothing, it is written by the people who build the model, and it removes the weakest reason to hire anybody. If a provider's proposal reads like a summary of those three courses, you are being sold something you already have.

Where the free courses stop

Self paced courses are general by necessity. They are built to be correct for every learner, which means they cannot be specific about any of them. Five things sit outside their reach, and all five are yours.

• Your data. What may go into a prompt, what never may, and who decides the edge cases. No external course knows your contracts or your regulator.
• Your admin settings. Who can enable what, what retention is set to, which connectors are allowed, whether staff are on a managed workspace at all. These should be decided before a rollout, not after somebody pastes a client list.
• Your context. A model is only as useful as what it can see. Getting your own documents, systems and house knowledge in front of Claude is engineering work, not a lesson.
• Your tasks. The difference between knowing Claude summarises well and knowing how your team's monthly report gets built with it is several hours of doing it badly first, with someone watching.
• Your policy and sign off. Who checks output before it reaches a client, and what happens when it is wrong anyway.

The UK evidence supports taking the second point seriously. The 2026 SKAI employer guide, produced with Skills England, sets out that baseline AI training should be mandatory before staff use workplace AI tools where organisational data, confidential information, regulated activity, safety or professional judgement are involved. That is a stronger position than most companies currently hold.

What Claude training for a team should cover beyond the basics

Once the free grounding is done, a paid session earns its place by being specific. In practice that means seven things.

• Projects, so shared context lives somewhere the team can reach instead of inside one person's chat history.
• Working with your own documents, and the checking habit that has to come with it. Plausible and correct are different, and the gap is where the incidents happen.
• Where the model is weak. Arithmetic, dates, anything it cannot actually see, and the specific failure of confident invention. Our post on using AI at work goes into the disclosure and checking habits that follow from this.
• Connecting Claude to your own systems, so it answers from your knowledge rather than the internet's. We wrote up the approach in giving your AI a company brain with MCP.
• Claude Code for people who do not write code, which is now a realistic proposition for analysts and operations staff rather than a novelty. See Claude Code training and whether non developers can build software with Claude.
• Admin, retention and workspace settings, walked through with whoever actually owns the account.
• When not to use it, written down, with examples from your own work rather than generic warnings.

If you want the wider view of what Claude does inside a business before you scope training, Claude for business is the overview.

Claude, ChatGPT or Copilot

Worth saying plainly because it changes what you should buy: most organisations end up with more than one of these, and that is a reasonable outcome rather than a procurement failure. Licences are cheap relative to salaries, the tools have different strengths, and staff will use whichever is closest to hand.

So training that only works in one tool ages badly. The SKAI framework makes the same point from the evidence side, warning against provision built on a single tool because the capability expires when the tool does. Teach the transferable judgement, then the tool specifics on top.

Our comparisons are Claude versus ChatGPT for business teams and Claude versus Copilot. Both carry the same disclosure this post does, because we are a Claude partner and you should read them knowing that.

The shape that works for a team

Shorter and more often beats one long day, which is not what most providers sell. The SKAI guide found stackable modules of 30 to 90 minutes are often more practical than longer courses, and recommends revisiting training at three and six months so people can reflect on how their habits actually changed.

Give people paid, protected time. The same report treats that as a condition of the training being reachable at all, and employers named limited availability at 37% as a barrier second only to cost at 42%.

And aim the depth where the gap is. Employers in that survey reported ongoing difficulty with technical AI skills at 67%, against 32% for responsible AI and 10% for non technical skills. The implication for Claude specifically is that the harder half of the work is connecting it to your systems and data, not teaching people to type into a box.

What to ask a Claude training provider

Six questions, including one that is uncomfortable for us.

• What is in this that is not already in Anthropic's free courses?
• Will we work in our own workspace, on our own files, with our own logins?
• Who decides our admin, retention and connector settings, and at what point in the process?
• What do we keep afterwards, beyond slides?
• What happens at three months?
• Is your certificate accredited, and if so, by whom?

On that last one, our own answer: we run an AI certification for individuals which is waitlist only, and it is not accredited by any awarding body. We would rather say that than let it be assumed. Any provider whose certificate is accredited should be able to name the body in one sentence, and several cannot.

Where we sit

Stated once more for clarity: our sister agency Apexure is a partner in Anthropic's Claude Partner Network, so we have a commercial interest in you choosing Claude.

On track record, HFI Institute brought me in as the AI instructor on a multi session programme they built for product managers at Salesforce, covering Claude and Claude Code on the team's own work. The case study sets out what was delivered and by whom.

In the UK we run this as a scoped team session rather than an open course, either live online or in person around Birmingham. The one day AI team workshop is the usual starting point, and AI training in the UK covers the wider formats.