Keith McCandless
7 min readJul 13, 2023


Six Maestros in Search of Creative Design: Liberating Structures and AI LLMs

With contributions from maestros Nancy White, Fisher Qua, Henri Lipmanowicz, Philip Clark, Elise Keith, and Keith McCandless

LS maes-tro | noun | a seriously playful person who uses, invents, and designs interaction with the Liberating Structures repertoire

Figure 1: Are we descending into or ascending out of the fog? Maestros mused that LS & AI LLMs were like looking over the horizon and seeing simultaneously an archangel-in-training and a devilishly clever pixie. [Golden Gate Bridge, photo by Keith McCandless]

A handful of maestros got together on Zoom to compare notes on using AI LLMs with Liberating Structures. There was a strange combination of giddiness and dread in the air. Are we working with an archangel-in-training or a devilishly clever pixie? A single Wicked Question summed up the vibe:

How is it that we are both seduced AND hesitant to use AI in our LS practice?

Like all Wicked Questions, articulating the paradoxical truths of a situation inspires hope of making progress. For this article, we are focusing on progress including and unleashing everyone everyday in creative design of their interactions (e.g,, meetings, classes, launches, retreats, coaching sessions).

While it seems too early to have sharp advice on what to promote and discourage, we have questions and loving provocations to share. Perspectives from six authors are shared below and separately by Nancy White and Elise Keith.

LS and ChatGPT in the Classroom

My interest in the LLM ChatGPT was sparked when I was forwarded a class outline created by a Danish professor named Daniel Russo. He is a new LS user (“infected” by Christiaan Verwijs) keen on sharing LS with students and colleagues. Daniel is researching applications of AI LLMs and he decided to generate the 90-minute workshop flow with ChatGPT. Daniel received very good feedback from students who experienced the workshop. [See the designed agenda at the end of this article].

Daniel was asked to step in for a sick colleague, so he needed a quick solution. He gave me permission to share the class outline while noting that “95% was generated by ChatGPT, with my input being minimal.” When I asked for more detail about the ChatGPT prompts, he shared a succession of three prompts:

Prompt 1 Objective: To engage high school students in an interactive workshop that introduces them to Large Language Models (LLMs), Generative AI, and ChatGPT, and encourages them to brainstorm creative applications and discuss potential ethical considerations.

Prompt 2 This workshop structure ensures that students are actively engaged in the learning process, with opportunities to brainstorm, discuss, and reflect on the potential applications and ethical implications of generative AI and LLMs like ChatGPT. Be more specific on which liberating structure to use.

Prompt 3 This workshop structure, incorporating specific Liberating Structures, ensures that students are actively engaged in the learning process, with opportunities to brainstorm, discuss, and reflect on the potential applications and ethical implications of generative AI and LLMs like ChatGPT. Restructure. Each activity can not last less than 20 minutes (except with introduction and conclusion).

Reflecting on the experience of prompting ChatGPT, Daniel added,

“As you can see, it’s very bad prompting (I also had almost no preparation time). Nevertheless, it’s still surprising (to me at least) on how well it did the job. If I can come up with a suggestion for you guys, I think it would be useful to have a ‘prompting manual’, where ‘ignorant’ users like me can just copy paste prompts and create their own WS based on their specific needs, like a restaurant menu.”

Figure 2: Icons “on the menu” for matchmaking challenges with specific LS

I like Daniel’s suggestion because helping LS users design strings (i.e., logical~creative progressions of multiple LS moving toward a goal) is a central part in my practice. I want to help people address their immediate challenges while learning a new design skill to apply to all future endeavors.

Challenges to developing coaching support is complicated for the following reasons: helping relationships require trust that develops with time and shared experience; the repertoire is large and growing; a limited number of people have direct experience across the whole body of work; gaining experience helps with design confidence but that takes time; and, grokking how one LS generates material and momentum for the next LS involves a shift to more developmental forms of design and evaluation.

Currently, I employ a variety of design aides and coaching strategies. None are perfect. All require a good chunk of time. Here are examples:

  • Working together or sharing leadership on a project
  • Using the LS Selection Matchmaker (often used while inviting co-leadership in client design groups)
  • Employing visual aids in the form of design cards and modular tiles in Google slides
  • Sharing field stories and storyboards with string illustrations from diverse settings
  • Developing “trial balloon strings” or options developed for a client’s design group
  • Offering suggestions via email, social media, or LS Slack, “You might try…” or “Here is something I tried when facing a similar challenge…”

So, more coaching support in the user community–particularly support to help generate options before making design decisions–could add value. Further, LS are very forgiving: new users are often thrilled with being invited to shape next steps with any LS selected. There are many good design choices for the designer/co-leader and navigating small bumps in the road is part of “failing forward.”

Jumping into use is far more important than the refinements that come with practice. Just as Daniel received great feedback from students for his hasty-AI generated class design, new users could expand the depth and breadth of their practice.

Self-Coaching with AI Prompting

Responding to Daniel’s suggestion, below we crafted this AI prompting guide and template for LS users. The prompts are equally useful in a coaching session that does not feature AI LLM use.

Figure 3: A Google doc version of this guide that you can download is here. Additionally, check out a generic guide here ChatGPT: Zero to Power User. Thanks to Silvia de la Torre.

Nancy White and Elise Keith have written more about AI and its use in design and coaching. Highly recommended. Nancy White muses in her blog about how to use AI to advance your LS practice, the essence of “warm electronic communication,” and shares a few experiments to test just how accurate AI LLMs are at the moment. Elise Keith sorts and shifts through worries while recommending AI use for brainstorming and meeting design. Also, check out the Zapier chatbot Elise has pulled together: you can ask for examples of how your group might use a specific LS and generate stories about what may be made possible for the participants. These functions humanize artificial intelligence just a teeny bit.

Wicked Questions and Ethical Considerations to Guide Next Steps

Just as Daniel included ethical considerations in his classroom design, we outline the seductive and troubling aspects of AI LLMs as a LS design aide. As our experience grows, we will draft Wicked Questions to confront these paradoxical attributes.

Figure 4: AI use with LS is simultaneosly seductive/useful AND unattractive/troubling. The paradoxial tensions can be confronted with Wicked Questions… to be developed as we gain more field experience.


Would love to hear about your experience with this archangel-in-training or devilishly clever pixie. Do you have Wicked Questions to share? There is a LS Slack channel (ai-and-ls-design) for people with a shared interest in experimenting with AI LLMs and LS design. You are invited to join the experiment.

Figure 5: Looking way out there, preparing to be surprised. [photo: overlooking the Panama Canal by Keith McCandless]

Daniel’s Class Design

Exploring the World of Generative AI: LLMs, ChatGPT, and Beyond

Objective: To engage high school students in an interactive workshop that introduces them to Large Language Models (LLMs), Generative AI, and ChatGPT, and encourages them to brainstorm creative applications and discuss potential ethical considerations.

Workshop Structure (90 minutes):

1. Introduction (10 minutes)

● Briefly introduce the workshop’s objectives and structure

● Welcome and icebreaker activity: Impromptu Networking

● Prompt: “Why did you choose this workshop?”

2. AI Video and Group Discussion (30 minutes)

● Watch:

● Liberating Structure: Conversation Café

● Students discuss the video in small groups (5–6 people), focusing on their initial thoughts and impressions (20 minutes)

● Each group chooses one key takeaway to share with the class

3. LLMs & ChatGPT Presentation and Idea Generation (25 minutes)

● Examples of how ChatGPT in study environment by [ name ] (5 minutes)

● Liberating Structure: 1–2–4-All

● Stage 1 (8 minutes): Students individually brainstorm ideas for creative applications of ChatGPT and other LLMs

● Stage 2 (9 minutes): Students pair up and share their ideas with each other, refining and expanding on their thoughts

● Stage 3 (15 minutes): Pairs combine into groups of 4 and consolidate their ideas, selecting the most promising ones and prepare a poster

4. Gallery Walk (20 minutes)

● Liberating Structure: Shift & Share

● Each group creates a poster or visual representation of their top ideas (5 minutes)

● Students walk around the room, reviewing and discussing each group’s work in a “shift and share” format, with one member from each group staying behind to explain their ideas (15 minutes)

● Optional: Students can place stickers or tokens next to the ideas they find most interesting or innovative

5. Ethical Considerations and Whole Group Discussion (20 minutes)

● Liberating Structure: Troika Consulting

● Students form groups of 3 (2 minutes)

● Each group member shares an ethical concern related to generative AI and LLMs (e.g., data privacy, misuse, biased algorithms) (6 minutes)

● The other two members of the group provide advice or suggestions on how to address the concern (6 minutes)

● Reconvene as a whole group and discuss the ethical considerations raised by different groups (6 minutes)

6. Conclusion and Reflection (5 minutes)

● Recap key takeaways from the workshop

● Encourage students to continue exploring the world of generative AI and consider how they can contribute to its future



Keith McCandless

Keith is co-developer of Liberating Structures and co-author of the book The Surprising Power of Liberating Structures ...