Skip to content
Claude Academy

Thesis

Why we run this live instead of recording it once.

Recording it once would be cheaper to make and far more profitable to sell. Here is the argument for not doing that.

The bottleneck was never information

Every fact in this curriculum is publicly available and free. The documentation is good. There are decent videos. None of that is the constraint, and treating it as the constraint is why so much AI training fails to change anything.

The constraint is the gap between reading that something is possible and having it working in your own environment, where the paths are different, the permissions are wrong, and the example assumed a project structure you do not have. That gap is where most people quietly stop.

Claude stopped being one thing

A couple of years ago there was a chat box, and knowing how to use it meant knowing how to write a decent prompt. That is no longer the shape of it. There is now a terminal agent, a desktop surface, an SDK, hosted agents, a protocol for connecting tools, and an API underneath all of it.

Most people learned one of those surfaces, usually the one they stumbled into first, and assumed it represented the whole. The expensive mistakes come from that assumption: building something laboriously on one surface that another handles natively.

Learning at lunch does not compound

The realistic pattern for a busy operator is twenty minutes here, a blog post there, an experiment abandoned when something urgent lands. That produces a scattered set of half-configured tools and no working system.

Self-paced courses are supposed to fix this and mostly do not, for a reason that has nothing to do with quality: without a fixed date, there is never a moment when not doing it costs anything. A scheduled cohort is a forcing function, and that is a real part of what you are buying.

Why it has to be your own work

Training on a sample project teaches you the sample project. The moment you turn to your own repository, with its particular history and its undocumented assumptions, the mapping falls apart and you are back at the start.

So the exercises here run against what you actually own. It makes the sessions harder to run, because twenty people arrive with twenty different environments. It is also the only reason anything survives past the week.

And why there is an assessment

Attendance certificates are worthless because everyone knows how they are earned. The signal is dead, which is a shame, because there is a real need for a way to show you can do this rather than that you watched someone else do it.

A credential is only worth what it costs to fail. So the assessment is a build, judged by a person, on your own context, and it can be failed. That is the whole design.

That thesis has a shape.

It is a two-day live cohort, capped at twenty, ending in an assessment you can fail.

Next cohort: Sep 15 – Sep 16, 2026 · 4 of 20 seats left