Designing AI into self-directed peer learning groups

Independent peer learning groups have the potential to enrich the learning experience: a space where peers explain ideas to each other, compare interpretations, and collectively make sense of new concepts. However, in practice, there is a common challenge to sustain this self-directed approach to learning: nobody in the group is facilitating.

We see this all the time. Put an experienced facilitator in the room, and the group's conversation is enriched with high levels of energy and engagement. This is the work of an effective facilitator, who knows when to probe or challenge an assumption, introduce a new or different perspective, ask for evidence, draw out a quieter participant, or encourage a group to think more deeply on an issue. Remove the facilitator and conversations can become less structured, harder questions are side-stepped or completely avoided, and groups converge earlier on their answers. All of this reduces the perceived value of self-directed peer learning, and in response participation declines. Over time, the group quietly stops meeting.

I have been thinking about this process a lot, and how AI might be used to improve the effectiveness of peer learning groups, using the value that skilled facilitation creates. A lot of what makes a skilled facilitator is not subject matter expertise, although this is sometimes needed. I suggest it is willingness. Willingness to ask: "what evidence supports that?" or "does anyone see this differently?". This willingness costs a facilitator nothing socially; that is their job. But this may not be the same for a peer. Bringing challenge to a peer learning group, and knowing how to do this, may bring perceived risk to the friendship, the working relationship, or concerns that it may change the tone of the room, and make the group feel uncomfortable.

Peer learning groups therefore do not just lack facilitation process; they lack someone who can afford to spend the social capital it takes to drive good outcomes. I think this is a challenge we see repeatedly across a number of different learning environments, including executive education and higher education, where peer learning is sometimes quickly disregarded in programme design discussions. This feels like a missed opportunity, and one where AI could provide an effective solution to better sustain peer learning groups.

A recent Harvard Business Review article by Gabriele Rosani and Elisa Farri, based on research across more than 300 managers in 35 organisations, argued that most teams already use AI individually but rarely bring it into the actual work of meeting, deciding and reflecting together. When used well, they say, AI can sharpen thinking before a group meets, challenge assumptions while it is meeting, and help turn what was discussed into something reusable afterwards.

Their two conditions for it working: intentionality — a leader has to deliberately choose to change how the group works, not just let a chatbot wander in — and craft: the prompts and structure have to be designed, not improvised.

Rosani and Farri are writing about workplace teams, not peer learning groups. That extension is mine, not theirs, and I have not tested it. But I am curious to learn if this idea could be applied to improve the efficacy and outcomes for self-facilitated peer learning groups.

Seeing the peer learning group as a workflow

Here is how the process could integrate AI.

To complete the process, each person would leave the meeting with one commitment, that could include returning to the source material or additional personal reflection.

None of this makes the AI a tutor; it has no expertise to offer. Its only job is to keep asking the questions an effective facilitator would ask, at the moments a skilled facilitator would ask them.

Which raises the real question: if AI's job is holding process and asking the hard questions, does this simply swap one dependency — no facilitation at all — for another: facilitation that the group cannot run without?

This is a key design question, suggesting that an AI version should actually become quieter, with reduced dependency, as the group grows more comfortable with self-directed facilitation. Building an AI version that introduces a new dependency is simply automating the solution but not addressing the problem. That is the harder version of Rosani and Farri's intentionality: not just deciding to bring AI into the process, but deciding on purpose. In this case that purpose should be to grow the peer group's capability to facilitate its own learning, and to sustain this over time.

---

Reference

Rosani, G. and Farri, E. (2026) 'AI can enhance every stage of teamwork — under two conditions', Harvard Business Review, 7 September. Available at: hbr.org/2026/09/ai-can-enhance-every-stage-of-teamwork-under-two-conditions (Accessed: 9 September 2026).