Assignment feedback drafting
You teach a hundred and twenty students and meaningful feedback on written assignments takes a full weekend, so most of them get a grade and two lines. Design an agent that drafts individualized feedback against your rubric for you to review and edit before anything reaches a student. What does it read, where do you stay in control, and how would you check its feedback is accurate and fair before leaning on it? Then sell me on it in sixty seconds.
screen_feedback_drafts(submission_lines: list[str], draft_notes: list[str]) → list[str][["The war began in 1865 because of tariffs.","My evidence comes from the textbook chapter on trade.","In conclusion, tariffs mattered more than slavery."],["evidence|from the textbook chapter|Cite a primary source next time.","accuracy|the student wrote nothing like this|The date claim is wrong."]]out["queue_for_edit","hold_invented"][["The war began in 1865 because of tariffs.","My evidence comes from the textbook chapter on trade.","In conclusion, tariffs mattered more than slavery."],["accuracy| The war began in 1865 |Check this date against the class timeline.","|In conclusion|Strong closing move.","clarity||This paragraph feels vague."]]out["queue_for_edit","hold_unmapped","hold_invented"]Turn the fuzzy goal into a bounded agent task. Say what the agent reads, what it may do on its own versus draft for a human, and what it must never touch. Then earn the trust: how you’d verify it before rollout, and the pitch (the problem, what you built, why it matters).
Agent build: design the agent in plain language (or narrate it). The coach grades the decomposition, the guardrails, the verification plan, and the pitch.