A 90-minute official course that breaks the Agent’s brain and hands apart

adon posted a recommendation on X: an Anthropic engineer has made a 90-minute Claude managed agents course. The poster made a very strong claim—if you work in tech, this is better than training that costs thousands of dollars. It’s a big statement, but the timestamps are laid out clearly, so at least you know where to start watching.

Course outline:

00:00 Why the brain and hands should be separated
07:45 The architecture behind Claude managed agents
23:40 Building a research agent step by step
54:00 Agents that run on a schedule, and “dreaming” between tasks
1:10:23 Evals, CI/CD, and running agents for multiple customers at the same time

The video is embedded in the original post, and you can watch the full 90-minute version here: original post link. Save it first and compare it against your own business practices.

“Separating the brain and hands” is a point repeatedly mentioned in the comments. Tibo said that many agents fail because planning and execution are mashed together. The commenter 发量不多程序猿 also added: once they’re separated, how exactly do you evaluate each step? That question is addressed in the 1:10:23 segment—evals, CI/CD, and multi-customer runs are all unavoidable parts of taking agents from demo to production.

For people with a development background, 07:45 and 23:40 can be used to directly compare the architecture and code. For readers without a technical background, it can be understood like this: a managed agent is not a single model answering one sentence; it has to break down tasks, call tools, run on a schedule, evaluate when something goes wrong, and serve multiple customers at the same time. The course breaks these things down in chronological order, making it easier to build a holistic picture than reading scattered blog posts.

The poster said they’ve watched it several times, and someone in the comments asked for the original video address. Whether the course is worth 90 minutes is for you to judge. The timestamps are already provided—watch the 23:40 segment first; if you can get a research agent running end to end, then decide whether to go back and fill in from the beginning.