Kostenlose Claude-Kurse (Anthropic Academy): welche sich lohnen — und in welcher Reihenfolge

“Free AI courses” usually sounds like marketing. In this case it's true: Anthropic, the company behind Claude, runs its own learning platform — the Anthropic Academy — with roughly 20 free, publicly available courses. You sign up with an email address and learn at your own pace, no credit card, no subscription. So the question isn't whether it's worth it, but which courses will actually help you — and in what order to take them, so you don't start with the wrong one. This guide sorts that out: honestly, without hype, and with an eye on freelancers and small businesses that want to build AI skills cheaply.

What the Anthropic Academy actually is

The Anthropic Academy is the official learning offering from Anthropic, the company behind the AI assistant Claude. The courses run on a learning platform called Skilljar — you don't even need a Claude account for it, an email sign-up is enough. There are currently around 20 public courses, all free, all self-paced. The range runs from the absolute basics (“how do I even use Claude?”) to deeply technical developer courses. That's exactly why a selection matters: not every course is meant for everyone, and some of it is decidedly niche.

When it comes to how long each course takes, Anthropic stays vague: the course pages state no reliable total duration. Reckon pragmatically with “several hours” for the larger foundational courses and considerably less for the short introductions. Firm minute counts would be guesswork — and making up numbers is exactly what we don't do here.

The certificate — honestly framed

This is worth a close look, because it's easy to get the wrong impression. After finishing a course (some include a small closing quiz), you receive a certificate of completion — issued through the platform in Anthropic's name, with a unique verification URL you can add to LinkedIn, for instance. It's a nice record that you worked through the course.

But: it is proof of completion, not an accredited exam. It confirms that you took part and finished the course — not that you earned a tested qualification. Don't confuse it with a different, clearly separate programme: the “Claude Certified Architect – Foundations” certification. That's a separate, paid professional exam (an exam fee of around 125 US dollars) that tests real knowledge of Claude Code, the Agent SDK, the API and MCP — aimed at people making production-grade architecture decisions. The two have nothing to do with each other. The free Academy certificates are learning records; the paid certification is a genuine exam. Keeping the two cleanly apart spares you false expectations later — and helps you spot providers who imply more with the word “certified” than is actually behind it.

The courses genuinely worth taking

Of the roughly 20 courses, only a portion pays off in everyday work. Here are the ones we'd recommend — each with a one-line note on what it gives you:

• AI Fluency: Framework & Foundations — the most important starting point: it teaches the mindset for collaborating responsibly and effectively with AI (more on that shortly). Suitable for literally everyone, from newcomer to heavy user.
• AI Capabilities and Limitations — explains the underlying mechanics: why AI hallucinates, where its knowledge boundaries lie, how its “memory” works. This tells you when you can trust an answer and when you can't.
• Introduction to Claude Cowork — for anyone who works a lot with files and documents: Claude not just in chat, but directly on your projects. It includes its own chapter on security.
• Claude Code in Action — hands-on training for anyone who codes or wants to automate development workflows. Real commands to follow along with rather than pure theory.
• Introduction to Subagents — shows how to delegate tasks to specialised helpers that work in isolation and return only what matters. The principle that keeps larger AI systems manageable in the first place.
• Introduction to Model Context Protocol — for the technically curious: how to connect Claude to your own tools and data sources over open interfaces.

What you can safely skip: the very extensive, multi-hour API development course, if you're not a developer — it goes deep into programming. Likewise the cloud-specific niche courses (Claude via Amazon Bedrock or Google Cloud's Vertex AI): only relevant if you build on exactly those platforms. And the many thematic “AI Fluency for …” spin-offs (for educators, students, non-profits and so on) are specialised versions — for most people the base framework is enough.

The thread that ties it together: the 4-D framework

If you take away just one thing from the whole Academy, make it this mental model. The course “AI Fluency: Framework & Foundations” centres on four core skills that all begin with D in English — the 4-D framework:

• Delegation — deciding which parts of a task you hand to the AI and which stay under your own judgement. Not everything that can be delegated should be.
• Description — describing clearly and with context what you want. A good prompt isn't a magic spell but a precise brief. Clarity beats trial and error.
• Discernment — judging the AI's output correctly: assessing quality and usefulness, spotting bias, noticing missing context, checking facts.
• Diligence — care and responsibility. You remain accountable, you verify, and you carry the consequences.

The heart of it is the fourth skill, diligence. It captures the decisive insight in one line: you hand off the task — but not the thinking, and not the responsibility. The AI does the work; the judgement about it stays with you. That's precisely the difference between someone who uses AI carelessly and someone who uses it with confidence. Internalise this stance and you avoid the classic beginner mistakes — adopting unchecked results, trusting the model blindly, offloading responsibility onto the machine. How to verify AI output systematically and keep a human in the loop at the critical points is something we go deeper into in spotting and safeguarding AI errors.

Recommended order — by user type

So you don't start with the wrong course, here are three paths. They aren't an official Anthropic curriculum but a plausible order drawn from the course descriptions — a recommendation, not a rule.

1. Beginner (little or no AI experience):

1. Claude 101 — orientation and core features
2. AI Fluency: Framework & Foundations — the mindset plus the 4-D framework
3. AI Capabilities and Limitations — understand why and when AI makes mistakes

2. Regular user (already uses Claude, but isn't a developer):

1. AI Fluency: Framework & Foundations — worth it even with prior experience
2. Introduction to Agent Skills — extend Claude with reusable skills
3. Introduction to Claude Cowork — if you work a lot with documents and files

3. Technical / coding (you code or use Claude Code):

1. Claude Code in Action — the foundation for everything else
2. Introduction to Subagents — delegate tasks cleanly to helpers
3. Introduction to Model Context Protocol — connect your own tools and data sources

If you're planning your first agent, our guide building AI agents yourself is the logical next step. And if you want to go deeper on writing good instructions — the “description” part of the framework — you'll find the practical side in prompt fundamentals for businesses.

The honest hurdle

There's one catch that has to be stated plainly: every course is in English. Video, text, exercises — all of it. If your English is solid, you'll get through fine; the language is clear and not overly academic. If English is hard for you, factor that in — the content is worth it, but it takes more effort. As of now, there's no German-language version.

Why we can frame this

A word on transparency: at Forge we use exactly these building blocks — subagents, skills, Claude Code, Cowork — in our own systems, every day. That's why the recommendation above isn't theory: we know from practice which of these concepts genuinely carry weight in day-to-day work and which are more decoration. This is not a paid mention or an affiliate deal — the Academy is simply a good, free way in, and we're telling you where your time is best invested.

Conclusion

The Anthropic Academy is a rare case of “free and still good”: around 20 courses, no cost, at your own pace. You don't need all of them — three to five well-chosen ones are enough. Start with “AI Fluency: Framework & Foundations,” because it delivers the stance everything else rests on: hand off the task, but not the thinking. After that, choose by need — understand, apply, or build technically. The certificate is a nice record of completion, no more and no less; the real professional exam is a separate, paid matter. And the only real hurdle is the English language. If you want to build AI skills cheaply and soundly, this is an excellent starting point — and the concrete question for you: which of the three paths will you take first?