Latest / Future of Work Tech with Fexingo: Remote Tools, AI Productivity, and Workplace Software / How AI Is Automating Your Internal Meeting Agendas
Transcript
- Lucas: If you've ever walked into a meeting where the agenda was clearly thrown together five minutes beforehand — bullet points out of order, no time allocations, that one item that's been on the list for six weeks — you know the feeling. The meeting starts with five minutes of everyone trying to figure out what we're actually here to decide. Luna: And sometimes you never figure it out. I've been in hour-long meetings that were basically an expensive group chat. Lucas: Exactly. Well, there's a growing push inside companies to hand that agenda-writing job over to AI. And I don't mean just a shared Google Doc with a template. I mean a generative model that digests your project status, recent emails, last meeting's notes, and spits out a structured agenda with suggested timeboxes and even recommended attendees. Luna: I've seen a few startups offering that. But it feels like the sort of task that sounds easy until you try to do it well. An agenda isn't just a list of topics — it's a negotiation about what matters most to whom. Lucas: Yeah, and that's exactly the friction. But a mid-sized SaaS company called TierOne — about 400 employees — recently ran an internal pilot. They trained a GPT model on eighteen months of meeting transcripts, project management updates, and past agendas. The goal: have the AI draft the agenda for weekly team stand-ups and monthly cross-functional reviews. After three months, they measured a 70 percent reduction in the time managers spent prepping agendas. That's not trivial. Luna: Seventy percent — that's like taking a two-hour prep job down to about 36 minutes. But how good were the agendas? Did the teams actually use them, or did they just get rewritten? Lucas: That's the interesting part. In the pilot, about 85 percent of ai generated agendas were accepted with minor edits. The main edits were around prioritization — the AI sometimes put a low-urgency item ahead of a critical deadline, because it didn't have a feel for the political temperature. But structurally, the agendas were more consistent than the human-written ones. Every item had a stated goal — 'decide on vendor' versus 'discuss vendor options' — and a time limit. Luna: That consistency piece is big. One of the biggest complaints in meeting surveys is that agendas are vague or missing entirely. If an AI can guarantee that every meeting has at least a decent structure, that's a baseline improvement. Lucas: A survey from a workplace analytics firm last quarter found that almost 40 percent of knowledge workers said they would trust an ai generated agenda more than one written by their manager. The reason? They felt the AI was less likely to skip over uncomfortable topics — like a project that's behind schedule or a budget overrun — because it doesn't have the social pressure to avoid conflict. Luna: That's a wild stat. But also kind of sad — that people trust a machine to be more honest than their own boss. But I get it. We've all seen agendas that conveniently omit the one thing everyone's actually worried about. Lucas: Right. So there's a real use case here. But the technology still has blind spots. Agenda-writing depends on context — not just the facts, but the relationships. The AI doesn't know that the VP of Engineering is sensitive about headcount discussions, or that the CEO wants to keep a new partnership under wraps until next week. If you feed it all the internal data, you might get an agenda that reveals something prematurely. Luna: So there's a governance question. Who decides what data the AI can see? And how do you prevent it from including items that shouldn't be on the table yet? Lucas: That's exactly the problem. The companies that are doing this well are using a tiered access model. The AI only ingests data that's already broadly shared — project status from the PM tool, last meeting's notes from a shared drive. It doesn't look at private messages or confidential docs. And then a human — usually the meeting organizer — reviews and curates. So it's not full automation. It's more like an AI assistant that drafts, and the human edits. Luna: That sounds sensible. I can see this becoming a standard feature inside platforms like Notion or Asana. Actually, I think some already have basic agenda templates, but not the generative piece. Lucas: They're heading there. Microsoft has been testing a Copilot feature that suggests agenda items based on your calendar and recent emails. But the big leap is when the AI can synthesize across multiple sources — not just your inbox, but the actual project status from Jira, the previous meeting transcript, the shared goals document. That's where TierOne's pilot got interesting. Luna: You know, I want to loop back to something you said earlier about trust. If 40 percent of workers trust AI agendas more than their manager's, that suggests a deeper issue with how meetings are run. Maybe the real value of AI here isn't just efficiency — it's forcing more transparency. Lucas: I think that's exactly right. The AI has no ego, no political agenda. It surfaces the things that the data says are important. If the project is two weeks behind, the AI will put that as item one with a red flag. A human manager might bury it at item four and spend the first twenty minutes on low-stakes updates. Luna: But is there a risk that ai generated agendas become too rigid? I've been in meetings where the agenda was so tight that any interesting tangent was cut off. Some of the best ideas come from going off-script. Lucas: That's a real concern. The TierOne pilot actually found that teams using AI agendas reported a slight decrease in 'creative detours' — but they also reported shorter meetings. The trade-off is real. The question is whether you value flexibility or focus more. For some meetings — like a weekly status update — you probably want focus. For a brainstorming session, you might want looser structure. Luna: So maybe the AI should be smart enough to adapt based on the meeting type. A status meeting gets a tight agenda; an innovation session gets a loose one. Lucas: Exactly. And that's where the next generation of these tools is heading. Some are already experimenting with 'agenda modes' — you tell the AI the meeting type, and it adjusts the level of detail and flexibility. The challenge is training the model to know the difference. Because even within the same team, a 'strategy review' can mean very different things from week to week. Luna: It's like the AI has to learn the culture of each team. That's a pretty personalized training requirement. Not all companies have the data or the willingness to do that. Lucas: No, and that's the barrier. A small startup might just use a generic template and that's fine. But for larger organizations, the ROI comes from customization. TierOne spent about six weeks fine-tuning their model. That's a non-trivial investment. But if it saves hundreds of hours across the company per quarter, it pays off. Luna: Let's talk about the elephant in the room: job displacement. If AI takes over agenda writing, what happens to the admin staff or junior employees who used to do that? Is this another task that gets automated away? Lucas: In the TierOne case, no one lost their job. The time saved was reallocated to more strategic work — analyzing the outcomes of meetings, following up on action items, preparing deeper research. The agenda prep was often something managers did themselves, not a dedicated role. So it's more about shifting their focus. Luna: That makes sense. But I wonder about the long-term. If AI gets really good at this, will we eventually have AI that not only writes the agenda but also runs the meeting? That's a different conversation. Lucas: We're not there yet, and I'm not sure we should be. The human element in meetings — reading the room, sensing when to push or pause — is still something AI is terrible at. But for the prep work? Absolutely. And I think that's a good thing. Meetings are expensive. The average mid-level manager's time costs the company about $100 an hour. If you can save ten minutes per meeting per person, that adds up fast. Luna: Okay, so let's be practical. If someone listening wants to try this, what's the first step? Do they need to build a custom model, or are there off-the-shelf tools? Lucas: Start with what you already have. If you use Microsoft 365, test the Copilot agenda suggestion feature. If you use Google Workspace, there are third-party add-ons like Clockwise or Motion that already generate basic agendas from your calendar. For something more custom, you can use a tool like Mem or Notion AI — you give them context, they generate a draft. You don't need a full custom model unless you have very specific needs. Luna: And the key is to review and edit. Don't just accept the AI's output blindly. Use it as a starting point. Lucas: Exactly. Treat it like a smart intern who did the first draft. You still need the human touch to catch the nuance. But if you're spending more than 15 minutes prepping a meeting agenda, you're probably overthinking it. Let the AI handle the grunt work. Luna: I think the biggest takeaway for me is that this isn't just about saving time — it's about making meetings more intentional. When the agenda is clear and structured, everyone shows up knowing what's expected. That alone can transform meeting culture. Lucas: Yeah, I think that's the real opportunity here. We've been complaining about bad meetings for decades. Maybe the fix isn't another meeting about meetings. Maybe it's an AI that forces us to be more disciplined about what we're actually trying to accomplish. Luna: Well said. And if any of our listeners try it, let us know how it goes. We'd love to hear real-world stories — good or bad.