Most introductory AI courses are built around capability. They show you what the model can do, walk you through a list of prompt patterns, and leave you with the impression that competence is a matter of knowing the right incantations. This one is built around judgement instead, and that difference is why it is worth your afternoon. The structure comes from a framework Anthropic developed with Prof.
Joseph Feller at University College Cork and Prof. Rick Dakan at Ringling College, organised around four competencies. Delegation is deciding what to hand to a model in the first place, which is the decision most people get wrong before they type anything. Description is communicating the task with enough context to be answerable.
Discernment is evaluating what comes back, including noticing when a confident answer is quietly wrong. Diligence is the responsibility layer, covering verification, attribution and the question of what you are willing to put your name on. Twelve modules, roughly three hours, free, and hosted on Anthropic's own learning platform with a Coursera version alongside it. What impresses me is the sequencing.
The course puts delegation first, which is unusual and correct, because the single largest failure mode I see in people using AI at work is not bad prompting, it is handing over tasks that should never have left their desk and keeping tasks the model would have done better. Once you have a name for that decision you start making it deliberately. Discernment gets similar treatment, and the course is honest in a way vendor material often is not about the fact that fluent output is not the same as correct output, and that the responsibility for checking sits with you. That is a slightly awkward thing for a model company to teach, and they teach it anyway, which counts for something.
The production values are calm rather than flashy, the videos are short, and there are exercises and handouts if you want to actually do the work rather than watch it. The limitations follow from the design. This is not a technical course and does not pretend to be. There is no API, no tool use, no agent architecture, no evaluation harness, and a developer arriving here hoping to learn how to build something will be disappointed within ten minutes.
Anthropic has separate material for that, and the Academy hub is where you should go instead. The framework itself, while useful, occasionally strains for the alliteration, and there are moments where four Ds is doing more work as a memory aid than as a genuine taxonomy. It is also vendor produced, and although I found the bias mild, mostly showing up as Claude appearing in the examples, you should read it knowing who made it. And the assessment is light enough that completion means very little, so the value comes entirely from applying the framework to your own work rather than from the certificate.
There is one more thing worth saying about the timing. A large number of organisations are currently rolling out AI tools to staff with essentially no guidance on how to use them responsibly, and filling that gap with either a policy document nobody reads or a vendor webinar that is really a product demo. This course is a genuinely good answer to that problem, it costs nothing, it takes an afternoon, and it leaves people with a way of talking about the decisions rather than a list of prompts to copy. I would happily hand it to a team.
My 4.3 reflects a well made, honest, appropriately short piece of teaching, held slightly below the top of the range because it is vendor produced, because the technical audience is not served here at all, and because the framework is a useful scaffold rather than the revelation the framing sometimes implies.