Two shipped agent products, and both are request/response: send, wait, render the answer. A stateful agent graph gives you none of that shape. It emits node transitions, partial state, tool calls awaiting approval, interrupts, and checkpoints you can rewind to — and that is the data your UI has to render. This course is about reading the graph well enough to know which surface each part of it demands, before anyone asks you to design one.
Start lesson one: A graph is not a chain of calls →
The whole spine is registered from day one so lesson numbers never shift. Module one is ready; ask your teaching agent for the next module once the reading-a-graph work is in ARTIFACT.md.
Opening somebody else’s graph definition and knowing, before you write a line of UI, what surfaces it demands.
The difference between a review gate you designed and a review gate the runtime enforces.
What persists between supersteps, and what that persistence lets a UI offer that a request/response app never could.
One framework taught you the vocabulary; the job is reading whichever stack a portfolio company already picked.
Agent frameworks move fast, and most of the documentation here is written by the vendor that sells the framework. Every claim on these pages links to the page it came from, with a date — if a source has moved on, check the resource list and tell your teaching agent. The API is expected to drift; the primitives are what the course is actually teaching.