AI Co-Facilitation: How Several AI Tools Support Your Workshop in the Background

A facilitator who takes notes, watches the clock, formulates follow-up questions, and reads the room all at once – that was a multi-tasking act long before AI came along. By now, there's an AI tool for almost every one of those jobs. The problem: most facilitators try them one at a time, and end up either with a messy pile of browser tabs or the disappointment that a single tool just can't do it all.

AI co-facilitation takes a different approach: instead of one tool taking over the workshop, several small, specialized tools work together in the background – each handling exactly one task, orchestrated around a shared agenda. Facilitation stays with you. AI takes on what's easy to automate: preparation, timing, documentation, sparring, follow-up.

The Principle: Division of Labor, Not an All-Rounder

Most AI meeting tools promise to do everything at once: transcribe, summarize, extract action items, even facilitate. In practice, that often results in a tool that does many things adequately – and a black box where nobody quite knows which data goes where.

AI co-facilitation flips that logic. Instead of one all-rounder, you get a stack of individual, swappable building blocks, each responsible for exactly one job:

1. Preparation – AI drafts a first version of the agenda
2. Timing – a tool keeps time and flow in view
3. Documentation – a tool listens in and records what's said
4. Sparring – a tool suggests follow-up questions in real time
5. Follow-up – a tool helps you process the results afterward

The advantage of this split: you can swap out, disable, or replace any single block with a tool you trust more. If one tool fails, or the topic is confidential, you can simply drop that one block without the whole workshop depending on it.

Before the Workshop: AI Drafts the Agenda

The first building block is the most common one: letting AI draft an agenda. You describe the goal, duration, and group size, and the AI returns a first block structure. In Sessionplan, this works directly through the “Create session with AI” option in settings: you copy a pre-written prompt into an LLM of your choice, describe your workshop, and get back a ready-made Sessionplan JSON to import.

We've described exactly how this works, what automation levels exist, and what to watch out for from a privacy angle in a dedicated article: Responsible Use of AI in Sessionplan. Important for co-facilitation: this first step only produces a draft. The real fine-tuning – breaks in the right spot, buffer time for discussion, the order of topics – stays manual work.

One small but effective addition for the prep phase: image generators like Midjourney or DALL-E 3 produce custom images or metaphors in seconds – handy as a visual icebreaker for check-ins (“Which image describes your week?”) or as a cover image for the closing presentation.

During the Workshop: Three Tasks, Three Tools

Sessionplan Keeps Time

Once the workshop is running, Sessionplan itself takes on the first building block of co-facilitation: timing. Live Mode shows the current block, remaining time, and next item throughout the session. You no longer have to watch the clock yourself and calculate in your head whether the discussion still fits the time slot – the tool handles that, while you focus on the group.

This is deliberately the only task Sessionplan itself handles during a live workshop. For documentation and question sparring you need additional tools – for the same reason Sessionplan doesn't have a built-in LLM: those tasks require processing spoken content, and that shouldn't happen unnoticed in the background of a planning tool.

A Second Tool Documents in the Background

For documentation, it's worth looking at open-source tools instead of reflexively reaching for the best-known cloud tool. Two examples:

• Whisper – OpenAI's open speech recognition model, which can run locally, for example through desktop apps like Buzz or directly via whisper.cpp. The recording never leaves your machine.
• Vosk – a lightweight offline speech recognition toolkit that also runs on less powerful hardware and integrates easily into your own scripts.

If you're already looking for a video conferencing tool with built-in AI transcription, similar functionality is now available in open-source alternatives like BigBlueButton or Nextcloud Talk.

If you'd rather use a ready-made, commercial tool instead, Otter.ai or Fireflies.ai are established alternatives that take live notes and automatically generate summaries including action items – in exchange, your data runs through the provider's servers. For a broader overview of open-source alternatives to well-known SaaS tools, see our article Free Alternatives to Expensive Software.

Two more useful helpers beyond pure transcription: Miro AI or the open-source affine.pro automatically cluster sticky notes on digital boards and summarize them as text. And if a recording needs cleanup afterward – removing background noise, cutting dead air – tools like Descript, which lets you edit audio like a text document, or krisp.ai for pure noise cancellation can help.

An AI Sparring Partner Suggests Questions

The third building block is the least standardized, but often the most valuable: a second AI window thinking along in the background. The idea is simple. At the start of the workshop, you open a chat with your preferred LLM and feed it context: the workshop's goal, the group, the agenda from Sessionplan. During the discussion, in quiet moments – say, between two blocks – you type in a few bullet points about what was discussed. The tool then suggests one or two possible follow-up questions you could ask.

An example: in a retrospective, you type in: “Team says deployments take too long, no clear root cause given.” The sparring partner suggests back: “Who on the team last ran a deployment, and what did they notice?” – a question you might otherwise only have asked after the session.

The role split matters here: the tool suggests, you decide. It doesn't see the mood in the room, who's holding back, or whether a topic is better addressed later. Use the suggestions as an extra perspective, not a script, and close the window as soon as it distracts more than it helps.

One variant is especially useful right before a decision: deliberately use the AI as devil's advocate. A prompt like “Take the role of devil's advocate and find three weaknesses in this outcome” surfaces blind spots before a decision becomes final – a critical counter-view that, in a group, often nobody wants to volunteer for.

If you want even more interaction, combine the stack with a tool like Particify: participants ask live questions or vote via smartphone, without you having to interrupt the flow of the session.

After the Workshop: Export as the Bridge

The last building block closes the loop. Sessionplan doesn't store your agenda on a server – instead, the planner's export function lets you download it as JSON. This file contains the complete structure of your workshop: blocks, times, notes, materials list.

That format makes AI-assisted follow-up easy. You can hand the exported JSON, along with your documentation notes – from Whisper, Vosk, or handwritten bullet points – to an LLM and ask it to produce meeting minutes, a list of open items, or a summary for participants. Because the structure is already machine-readable, the AI doesn't have to reconstruct it from a wall of prose first.

If you want to automate this step on a regular basis, our article Responsible Use of AI in Sessionplan also covers the skill-file and MCP-server routes – letting you trigger the export-import loop directly from your own AI tool.

A Practical Example: A Workshop with AI Co-Facilitation

To see how the five building blocks show up in a real agenda, check out our example template: a 90-minute team workshop where every block notes which AI tool comes into play – from the tech check at the start to the export at the end.

Checklist: Build Your Own Stack

1. Draft the agenda with AI and import it into Sessionplan, or adjust it by hand
2. Before the workshop, decide whether and how you'll document it – and inform participants
3. Test an open-source documentation tool before using it in a real workshop
4. Turn on Live Mode in Sessionplan so timing stays in view
5. Optional: prep a second chat with context in case you want question sparring
6. After the workshop: export the JSON and use it together with your notes for follow-up

What AI Co-Facilitation Is Not

AI co-facilitation doesn't replace facilitation experience. No tool notices when a break is needed because energy in the room is dropping, or when a conflict is better handled one-on-one than in the group. The individual building blocks take routine tasks off your plate – time management, note-taking, an extra angle on follow-up questions – and by doing so, free up space for the moments where human judgment actually matters.

Just as much, AI co-facilitation isn't surveillance through the back door. If you're recording in the room, say so beforehand. If you're passing notes to an AI, know in advance which tool runs where and what data it processes. The whole point of splitting things into separate building blocks is that you can always trace exactly what's happening – and drop any block that doesn't feel right for a particular workshop.

Conclusion

AI co-facilitation works best as an interplay of small, swappable tools – not as one tool that takes over the entire workshop. Sessionplan deliberately handles only part of it: AI-assisted preparation and timing in Live Mode, both without a built-in LLM and without server-side storage. Everything else – documentation, sparring, follow-up – you combine with the tools you trust. In the end, the agenda remains the shared foundation: a Sessionplan link that works the same way no matter how many AI tools ran in the background.