Live coaching suggestions for human tutors
During online tutoring sessions, an agent suggests next moves to less experienced tutors, who decide what to say to the student.
The use case
Tutoring programs hire many new tutors who lack the instincts of experienced teachers, especially in math. A coaching agent watches the chat of a live session and, when the tutor asks, suggests a guiding question, a hint, or a way to address a misconception, modeled on expert teaching moves. A Stanford study of this approach found students of tutors with the tool were more likely to master topics, with the largest gains for weaker tutors. The tutor stays in charge of the conversation and chooses whether to use, change, or ignore each suggestion. Program leads review session quality.
Example workflow
- Tutor runs a normal chat-based session
- Tutor requests help when a student is stuck
- Agent suggests a guiding move based on the chat
- Tutor adapts or ignores the suggestion
- Program lead reviews sessions and outcomes
Tools
- Tutor CoPilot research
- Claude
- Zoom
Risks and guardrails
- Suggestions may not fit the student
- Tutors may lean on prompts instead of learning
- Session recordings involve minors' data
- Measure outcomes, not just usage
Sources
Written by Agents Lookup. Credits and official pages for the tools this use case names:
- Stanford Tutor CoPilot paper on arXiv: arxiv.org/abs/2410.03017
- Claude (official page): claude.ai (checked 2026-11-09)
- Zoom MCP server (official page): developers.zoom.us/docs/mcp (checked 2026-11-09)