Setting the AI policy for a task

Every task can declare what AI is allowed to do on it, and whether an apprentice must disclose using it. This is authored once, on the task, and enforced everywhere the app offers AI assistance.

The shape

{
  "aiPolicy": {
    "mode": "guided",
    "allowedCapabilities": ["explain", "hint", "quiz", "coach"],
    "solutionGenerationAllowed": false,
    "disclosureRequired": true
  }
}

{
  "aiPolicy": {
    "mode": "guided",
    "allowedCapabilities": ["explain", "hint", "quiz", "coach"],
    "solutionGenerationAllowed": false,
    "disclosureRequired": true
  }
}

• •mode — the overall stance for this task (currently guided in practice).
• •allowedCapabilities — which of the four apprentice-facing task-detail modes are switched on: explain, hint, quiz, and coach. Leave a mode out and it shows as disabled rather than hidden, so the apprentice can tell it's a deliberate choice, not a bug. Unless you say otherwise, a new task gets explain, hint, and quiz — coach is off until you add it.
• •solutionGenerationAllowed — whether the apprentice can ask AI to produce a full solution once they've explicitly scoped a conversation to this task. Most tasks keep this false; you can enable it deliberately for a task where working through a generated solution is itself the point.
• •disclosureRequired — whether Submit for review is blocked until the apprentice records how they used AI (None / Assisted / Generated).

See the five AI assistance modes for what an apprentice actually sees and experiences from explain, hint, quiz, coach, and solution. Mentor-side AI assistance (pre-review and the mentor-packet assistant) is separate from this policy entirely — it's always available to mentors, since it's a tool for the mentor's own judgment, not something this policy needs to switch on or off.

Authoring this in apprenti creator

The task editor presents AI policy as chips and toggles — a segmented mode selector, capability chips, and two switches (solution generation, disclosure) — never as a raw JSON blob to hand-edit. Disclosure defaults on for new tasks; you turn it off deliberately, not by omission.

Why disclosure matters

Disclosure is not a ban on AI, and most tasks allow explanation, hints, and quizzes. It exists so a mentor reviewing a submission can trust that what looks like the apprentice's own reasoning actually is — or knows exactly where it wasn't. Requiring disclosure by default keeps that guarantee in place; leave it off only for tasks where AI involvement genuinely doesn't need tracking.

A reasonable default for a new task

{
  "mode": "guided",
  "allowedCapabilities": ["explain", "hint", "quiz"],
  "solutionGenerationAllowed": false,
  "disclosureRequired": true
}

{
  "mode": "guided",
  "allowedCapabilities": ["explain", "hint", "quiz"],
  "solutionGenerationAllowed": false,
  "disclosureRequired": true
}

Loosen it deliberately per task, not globally — the policy lives on the task precisely so different tasks can make different calls.