Start with the job to be done

The working assignment is to design a differentiated fraction mini-lesson while keeping assessment judgment with the teacher for ten-year-old learners with mixed confidence. That sentence is narrower than “use AI for teachers.” It identifies a deliverable and a reviewer. Write a definition of done with three layers: the output must satisfy the audience's need; factual or functional claims must be traceable; and a named person must own the final decision. Adapt the route to learning, not the destination. AI drafts examples and supports; the teacher remains responsible for accuracy, context, and consequential decisions.

Start by separating tasks. The model may draft, classify, transform, compare, or suggest. It may not silently approve, publish, grade, deploy, cite, or consent on someone's behalf. For this assignment the authoritative material is learning target, prior knowledge, reading level, allowed representations, misconception patterns, time, and accessibility needs. Anything absent from those inputs is either an explicit assumption or an unanswered question.

Use this prompt as a realistic starting contract:

Learning target: compare fractions with unlike denominators. Create a three-minute explanation using a pizza model, one worked example, four practice items from easy to challenging, and an answer key naming likely misconceptions. Use fictional learners only. Keep the target fixed and do not make grading or placement decisions.

Notice what the prompt does: it states the setting, limits the output, names forbidden behavior, and requests evidence that can be reviewed. It does not ask the model to “make it amazing.” If a constraint matters, make it testable. Replace “be accurate” with a source boundary, formula check, test command, rights ledger, or approval step.

A useful response would look like this: An age-appropriate explanation, a correctly solved example, practice that measures the same target, and a misconception note about comparing denominators directly. That description is intentionally observable. “Looks good” is not acceptance. The operator must solve every item, confirm support versions assess the same target, check reading load and cultural assumptions, and have the teacher approve classroom use. Keep the source material beside the draft so review means comparison, not memory.

Run the work through target → draft → differentiate → solve → review → teach. Stop when an authoritative input is missing, a high-risk claim lacks evidence, private material cannot be safely removed, or the proposed action exceeds the permission granted. Escalation is successful workflow behavior, not model failure.

Common framing mistakes are wrong answer keys; difficulty changed by changing the target; accidental answer reveal in hints; biased examples; automated grading or discipline recommendations. Prevent them by writing a one-paragraph job card: user, decision, deliverable, source of truth, constraints, reviewer, and stop condition. This card becomes the anchor for every later prompt.