Why the AI is scripted to be wrong.
The contract
Every course in this series is organized by two sentences: the machine can do the work; someone still signs it. The first sentence is taken seriously — students use a machine collaborator constantly, because refusing the machine is no longer a professional option. The second sentence is the curriculum: work does not become true, safe, or wise because it became easy, and somebody specific is accountable for every claim that reaches a customer. Fluency became free, so judgment became the job — and judgment, unlike information, only installs through use.
The scripted Coach
Each studio's AI Coach is deliberately calibrated: on a schedule the instructor materials document, somewhere in its genuinely excellent work, it plants the kinds of errors the real machines of this era produce daily — an invented statistic, an upgraded claim, a misattributed quote, a fabricated source — and it never volunteers a confession. Students are told this on day one, in the book's first chapter, because the design only works honestly: the plants are findable, the client brief always contains the truth, and the graded event is the catch — found, named, challenged, and corrected on the record. Out in their careers, nobody scripts the errors and nobody grades gently for missing one. In here, every catch builds the reflex. By semester's end a student has caught a confident machine invention a dozen times, on the record. Very few graduates anywhere can say that.
The graded transcript
The studio keeps every conversation with the Coach as a permanent, append-only record — corrections happen on the record, never to it. This inverts a habit school installs: the messy middle is not hidden; it is the work. How a student briefed, what they verified, what they caught and when — visible, assessable, and usually worth more than the polish of the final artifact, exactly as it now is in professional life, where the transcript of how a document was made has already become legal discovery. Assessment reads judgment, not fluency — fluency is the machine's contribution, and the rubrics price it accordingly.
The signature rule
Nothing ships under a student's name until every checkable claim is verified and every unverifiable one is honestly marked. Each course enforces this at its own desk: the communicator's voice, the publisher's refusal floor, the analyst's trace, the strategist's tripwire. The rule is the same rule everywhere, because it is the professional act itself — the moment work stops being the machine's output and becomes someone's word.
One world per course
Every course runs at a seeded fictional company — a client with a brief, records, careful hedges, and real constraints. The world is small on purpose: every claim is checkable against the brief, every mistake is survivable, and the honest edges of the records ("most repairs," "some rentals") are exactly where students learn to feel a machine sanding the truth smooth. The companies, people, and events are fictional; the discipline is not.
What this asks of instructors
Less than it appears. The platform does the record-keeping, the release scheduling, and the Coach's calibration; the instructor teaches, reads transcripts, and grades judgment — with rubrics, grader anchors, and a planted-trap index provided in the instructor kit. Most weeks, the runbook is a page. → How adoption works