The workshop is one day, indoors, seated around a shared workbench. Nobody is asked to code. Nobody is asked to trust a black box. Teachers leave with three artifacts they can put in front of a class on Monday.
The morning: workbench tour
We open at the workbench, which is the same piece of glass described in a tour of the teachers workbench. The opening hour is not a lecture. It is a slow walk across the three panes, with every teacher clicking through to the underlying node whenever a number moves. The vocabulary is the real technical vocabulary of active inference: generative model, prediction, update, gate (Class E, following Parr, Pezzulo, and Friston, 2022, Active Inference, MIT Press). We use the words as they are used in the literature, and we point at the on-screen behavior that each word names.
By the end of that hour, every teacher in the room can do one small thing: pick a number on the workbench, ask "where did that come from," and follow it back to the input, the belief, and the last update that changed it. That is the whole promise of the tour (Class C, workshop configuration).
Late morning: the gate design lab
The second block is a lab, not a talk. Each teacher writes one gate for a lesson they already teach. A gate is a small named check with a plain-language reason string, and a decision to pass or fail. We work in pairs. One teacher writes the gate. The other teacher tries to break it with a realistic student input.
The gates that come out of this session are not our gates. They are the teacher's gates, in the teacher's voice, for the teacher's classroom. A middle-school science teacher might write a gate that refuses any answer without a cited observation. A high-school English teacher might write a gate that refuses any summary that does not name the source passage. The pattern is the same across grade bands, and we have separate write-ups for gate design for middle school and for higher grades.
The lab ends when every teacher has one working gate, tested against at least two adversarial inputs, and a written reason string a student could actually read out loud.
Early afternoon: honesty policy drafting
After lunch we draft the classroom honesty policy. This is a one-page document that the teacher will sign, the school leader will sign, and the students will read on day one. It names what the tool is allowed to do in the room, what it is not allowed to do, and what the student is expected to declare on any submitted work. The AI-authorship fence sits in the middle of this page, and it stays in the teacher's voice.
We do not hand out a template and ask people to fill in the blanks. We draft the policy in the room, with the school leader present when possible, and we test each clause against a scenario the teacher brings from their own classroom. The policy that leaves the room is a policy the school can defend to a parent, to a department chair, and to a student who wants to know why a rule exists (Class C).
Late afternoon: the follow-up cadence
The last block is short and administrative. We set the follow-up cadence in writing. The default is three touchpoints: a two-week check-in on the initial gate in use, a six-week review of the honesty policy against actual classroom events, and a three-month audit of the workbench against the teacher's evolving lesson plan. The cadence is written into the same document as the honesty policy, and the school leader countersigns.
Nothing about this cadence is a subscription. It is a partnership rhythm, and either side can end it in writing. The partnership page has the current pricing, the current dates, and the school-leader intake form.
What teachers leave with
Three artifacts, all on paper and all in the teacher's voice: one working gate, one honesty policy signed by the room, one follow-up cadence with dates on it. The workbench is a working hypothesis on an attainable path toward General Natural Intelligence, natural not artificial, and the workshop is where teachers learn to inspect it. Do not take the claim on faith. Test the gate in your own classroom, read the reason strings out loud, and tell us where it fails.