Sometimes the answer is the wrong thing to give
A correct answer to a student's question can still be a teaching failure. A question about which method to use on an assessed problem is generally a request for the answer. If you give that answer, you take away the exercise that the assessment is meant to test.
Other questions need a direct answer right away, such as when the deadline is, which reading addresses a particular topic, and what a term means. Treating those as teachable moments would only be obstructive.
The difference is the kind of question. Administrative and definition questions are answered. Questions that form part of the assessed work receive a prompt, a reference, and the route to the answer rather than the answer itself.
The agent answers in this way and stays within the course material. Any question it cannot answer is passed to the educator rather than made up. All assessment-related exchanges are visible to the teaching team.
How the agent is built in Agent Studio
Course material is provided with the assessment brief, because the brief tells the agent which questions constitute the exercise.
The second prompt skill holds your department's position on collaboration and assistance, so the boundary the agent upholds is the one you have stated rather than one that has been invented.
The system prompt asks for simple answers to administrative and definitional questions. For assessed-work questions, the response is based on the related material and includes a follow-up question.
The system prompt does not allow responses based on general knowledge. Anything that falls outside the course material is passed to the educator unchanged.
Students contact the agent through a Slack or Discord node. The Agent Chat node then continues the conversation in a durable session, so a follow-up request remembers what has already been asked.
All exchanges classified as assessment-adjacent are recorded in Google Sheets and visible to the teaching team, which is why the boundary remains auditable.
If a request relates to grades, deferrals, or a student's circumstances, a Human in the Loop node passes it directly to the educator without altering it.
What this agent is built from
- Prompt skills: course material plus the assessment brief that defines the boundary.
- Prompt skills: your department's stated policy on assistance and collaboration.
- System prompt: plain answers to administrative questions. Material and a prompt on assessed work.
- System prompt: nothing answered from general knowledge. Unknowns go to the educator.
- Slack or Discord (Communication): where students reach the agent.
- Agent Chat (Util): durable sessions, so follow-ups keep their context.
- When an ActionFlow is enough: If answering recurring administrative questions from a fixed source is the work, build the workflow. That sequence removes most of the volume.
Frequently asked questions
Start building AI workflows
Create a free account, open a template or a blank canvas, and run your first ActionFlow.
Newsletter
Get product updates
New nodes, agents, and product notes. We send mail only when we have something worth opening.