Where should an AI agent draw the line between tutoring and doing the work?

The question

How do you configure course Agents to guide student learning through hints and questions without generating finished coursework? Practice varies across disciplines: some instructors allow Socratic questioning only, while others permit worked examples with modified numbers or citation checks but prohibit draft generation entirely.

Useful context

Alkimi includes Learning Assistants and other Student Learning templates configured to ask guiding questions, offer hints, and refrain from completing assigned tasks. Instruction Library sections cover academic tutoring and academic integrity norms, and faculty can add the Academic Integrity Reminder Safeguard where a course Agent will be asked coursework questions directly.

Those Safeguards are prompt-level controls, so it is worth being precise about their reach. They shape the system prompt the Agent runs on, and they stop short of hard-filtering student inputs or guaranteeing compliance. Clear syllabus expectations and thorough testing therefore carry as much weight as the configuration itself.

Share your approach

Please share how your course defines this boundary.

  • Allowed and off-limits assistance. What does your syllabus permit, and what do your Agent instructions rule out?
  • Verification prompts. Which test prompts did you use to confirm that your Agent holds the boundary?
  • Surprises. What unexpected response occurred when students first used the Agent?

Please state your discipline and course level, and keep student names, education records, and graded assignment questions out of your reply.