Claude Academy Is Live: What Anthropic's Free Courses Actually Cover

Almost every team I train, someone asks me some version of "is there a free way to learn this first, before we do the live thing?" Up to now my honest answer was a shrug toward the docs and a couple of YouTube channels I trust. As of last week I have a better answer, and it's coming from Anthropic itself.

On August 20, 2026, Anthropic launched Claude Academy, a free, self-paced course platform covering Claude.ai, Cowork, Claude Code, Claude Tag, and the Claude Platform (the API). It's a rebuild of the Anthropic Academy that had quietly existed since March, hosted on Skilljar, with a handful of courses on the API and Claude Code. The new version is bigger, better organized, and structured around a specific way of thinking about working with AI rather than a feature-by-feature tour.

I went through the parts most relevant to how I teach, so here's what's actually in it, honestly.

What's actually there

Sign-up is just an email, no credit card, and every course ends in a certificate you can put on LinkedIn. The catalog spans beginner to technical: Claude 101 for someone who has never opened claude.ai, all the way to Claude Platform 101, which walks through picking a model, building an agent loop, tool use, and context management for people writing against the API directly.

The course I'd actually point a working engineer toward is Claude Code 101. It doesn't start with "here's how to install it." It starts with agentic loops and context windows as concepts, then covers installing across terminal, VS Code, JetBrains, and the web, and centers the whole thing on Explore → Plan → Code → Commit as the default workflow. That's the right place to start someone. It's the same shape I use to open the Claude Code part of a live session, and seeing Anthropic teach it the same way is a good sign, not a threat to the curriculum.

Underneath all of it sits something Anthropic didn't invent themselves: a "4D" framework for AI fluency, originally research out of Ringling College of Art and Design and University College Cork from 2023-2024. Delegation, Description, Discernment, Diligence. It's the load-bearing idea of the whole platform, more than any single course:

That last box, Diligence, is the one most tutorials skip entirely, and it's the one I care most about teaching. Anyone can learn to prompt. The gap that actually shows up on a Day 2 build is people who can't yet tell when Claude's output is quietly wrong.

What it's genuinely good for

I'll say the useful part plainly: if someone on your team has never touched Claude.ai and needs the absolute basics, Claude 101 is a legitimate place to send them. It's structured, it has quizzes and hands-on exercises baked in rather than just video, and it costs nothing. I'd rather a new hire spend ninety minutes there before their first real session with me than start from zero.

One detail worth knowing before you assign it to anyone: Claude Code 101's hands-on exercises need an actual Claude Code seat, meaning a Pro, Max, or Enterprise plan, or your own API key. It's not a course you can complete on the free tier watching from the sidelines. That's a real cost most write-ups gloss over.

Where it stops

A self-paced course platform, however well built, can't do two things a live room does. It can't watch you build something specific to your actual business and tell you, in real time, that the approach you're taking will fall apart in three weeks. And it can't put your whole team in one room working on your own tools, where half of what you learn comes from a colleague's question you'd never have thought to ask.

That's not a knock on Anthropic's work here, it's genuinely well made. It's just what self-paced courses have always been true of, long before Claude existed: they're excellent at transferring information and much weaker at building judgment under pressure, which is the actual skill the Diligence box in that framework is naming. You don't build discernment by watching a video get it right. You build it by watching your own output be wrong, live, with someone telling you why.

So here's what I'm actually doing with it: when someone on a team I'm about to train says in the intake survey that they've never used Claude Code, Claude Code 101 goes into their pre-work. It gets the vocabulary question out of the way before the session instead of eating fifteen minutes of it. Everything past that, I still teach live, because the parts that matter most are the parts a course can't grade. If you want that for your team, built around the tools you already use, that's what private team training is, and 1:1 coaching is the same thing for one person.

Sources: Claude Academy · Claude Code 101 · Claude Platform 101 · AI Fluency: Framework & Foundations