A live demo of the stateful loop: a fresh, empty context picking up an in-progress journey (6:11).

The interaction

  1. He types just teach and tells it his current status in plain language: he can now complete the cube except for the corner cycle, which hasn't reached memory (6:11).
  2. Because the context is empty, the agent starts by checking the state of the teaching workspace (6:11).
  3. It reads the solve cards to see how far he's got (6:35).
  4. It diagnoses the problem rather than restating it: the concept behind the corner cycle algorithm is solid — what's missing is muscle memory (6:35).
  5. It reads an earlier memorization lesson to match the house style before writing the new one (6:35).

What this demonstrates

  • The agent has effectively been turned into a teacher, and you work with it as you would any one-to-one teacher: you tell it what problems you're having, and it designs lessons to solve them (6:35, 7:00).
  • The diagnosis is expressed in real teaching vocabulary — it invokes the zone of proximal development and prescribes "one algorithm, pure muscle memory work" (7:00).
  • Because everything lives on the file system, he can clear the context and every subsequent teach run still has what it needs (8:04).
sequenceDiagram
    participant U as Learner
    participant A as Agent (empty context)
    participant FS as Teaching workspace
    U->>A: "teach" + current status
    A->>FS: read mission, lessons, solve cards, learning records
    FS-->>A: state of the journey
    A->>A: diagnose the gap (concept vs muscle memory)
    A->>FS: read an earlier lesson for house style
    A->>U: new lesson targeted at the gap
    U->>FS: (report progress → new learning record)