When success relies on a combination of variables known and unknown, controlled and encountered, you don't get reliably great results by just following a recipe.

I was explosively reminded of this with my first batch of home-fermented soda. When I cracked the top on a bottle to "burp" the gas—using the technique I saw demonstrated online, at the time indicated in the recipe—my strawberry rhubarb soda became a roaring geyser of sticky pinkness. It took my husband and me two days to clean it all off the kitchen, my clothes, and the dog.

Because, of course, she parked her furry butt right there with me in the strawberry downpour.

Folks eager to ferment soda can find lots of recipes spelling out the exact ingredients, timing, and how to recover when things go wonky. You can find lots of meeting recipes too. Both processes use tools and take place within containers: measuring cups and bottles for the one, slides and meeting rooms for the other. And in both cases, you can still fail spectacularly because success requires paying attention to how the living beings in your meeting room or bottle are interacting with the recipe.

People buying AI note takers often hope the bottle will burp itself. They didn't love their meetings before, but now they have transcripts! Somewhere in there, surely, value will bubble up.

What kind of value? This varies based on who you ask, but it typically comes in two flavors.

Meeting transcripts are raw fermentation material. Rich, accurate records create possibilities, but turning possibilities into realized value requires someone who knows what they're looking for.

Some leaders crave delectable insights. But of course, the analytical lens determines what you find. Bring an anthropologist's eye, and you'll surface cultural patterns: what different offices believe, how they signal status, and where the rituals diverge. Bring an I/O psychology lens, and you'll find workforce dynamics: why certain people stop speaking up, or where emerging managers are struggling. Bring an innovation lens, and you may uncover buried product ideas that never escaped a standup.

Other leaders are hungry for operational effectiveness: teams getting more work done more reliably, with less drama. Clear action items, solid decisions, updated project records, and happy green OKRs. This value is realized when the people who were in the meeting walk out knowing exactly what they decided and who is doing what next.

These are different promises. AI note takers fall short when people conflate these expectations and assume that all this value arrives effortlessly via the AI-equivalent of DoorDash for meetings.

From 2010 to 2025, I led a software company dedicated to capturing value from business meetings. In that time, we built the Meeting Performance Maturity Model: a framework that describes what meetings look like as organizations develop more effective practices. The model, along with the templates, consulting frameworks, and the book that followed, came out of watching teams struggle to bridge the gap between the results they wanted and the results they were equipped to get, given what they knew about meeting well.

Here's what we learned about meeting records. At Level 2, the baseline for professional meetings, records provide evidence that expected practices were followed. That's the floor: proof something happened. The practices that move work forward don't appear until Level 3 and above, and they all require humans actively engaging with documented outcomes during the meeting itself.

We've always known this. You probably know this, too.

In formal settings, the chair says, "Read back the decision" or "Strike that from the record." Minutes get distributed for review and approval before they become authoritative. In well-structured team cadence meetings, you see the same pattern: the EOS Level 10 meeting, Coda’s Dory & Pulse, real-time agendas, and Amazon's 6-page Memo. Teams working at an Obeya or a Kanban board update their visual record in the room, in the moment, together. The humans collectively construct shared meaning and agree that what they can all see right there in front of them is worth acting on.

This isn't new. It's documented organizational practice across decades and industries.

I've watched a decade of automated transcription and note-making tools arise. Most got more buzz, more funding, and more early adoption than our more prosaic, human-driven tools did. They also quickly folded once teams realized that the novelty of easy records doesn't improve meeting quality.

The tools that survived are committed to answering a specific question by using transcripts as analytical fodder. They don’t win by promising operational effectiveness. Gong.io is the clearest example: it analyzes transcripts from sales calls to identify what the most successful salespeople say, how much they talk, and which questions they ask, then turns those patterns into playbooks for the rest of the team. Gong mines the transcript data as an organizational dataset to develop better sales strategies. Gong’s value lies in the analysis, not in what happens during individual meetings.

That's a coherent product. It works because it doesn't pretend passive capture equals team alignment.

For executing on strategy and moving projects forward, passive capture isn't enough, and the maturity model explains exactly why. An AI note taker operating silently in the background functions like a classmate taking lecture notes. When you zoned out in class, your friend's notes were a handy reference. But imagine scanning their notebook and spotting a line saying you will buy pizza for the study group. Wow. Um, no. Maybe you suggested it, but you never promised to pay. That's not happening, no matter what their notes say.

AI meeting notes are like that. They look mostly reasonable—there's an odd conclusion here and there, names spelled weird, and that conversation about Jim's new haircut probably didn't need to be in the permanent record—but we aren't bound to anything written there. Also, most AI-generated summaries go unread.

If an action item pops out and no one hears it happen, does it make a sound?

People read the updates and announcements, then change what they’re doing. Recorded decisions turn into action. People are held to account for the tasks assigned to them. Meeting notes that simply document what was said may feel satisfying, because “Oooh, you didn’t lose anything!”, but that's not the same as driving work forward.

This failure can't be resolved by training AI on a better recipe or through more clever automation. The only way to ensure the humans in a meeting walk out with a shared understanding of what they decided and what they'll do is to engage the humans in documenting those results.

To be fair, the tools are evolving. Microsoft's Teams Facilitator agent now generates real-time notes that everyone in the meeting can co-author simultaneously, stored in Microsoft Loop and synced to Planner. That's closer to the shared visible record that Level 3+ meetings require. This gives teams infrastructure for active engagement.

But infrastructure isn't practice. And this is where our own software history is instructive.

We built meeting templates that walked teams through structured agendas in real time: timers for each topic, prompts for the kinds of notes to capture, cues for who was doing what. We gave teams exactly the kind of active engagement scaffolding that the maturity model describes. What we found was that teams without an existing foundation in well-structured meetings couldn't use it effectively—not because the tool was confusing, but because they lacked the meeting competency to operate within the structures it offered. Before we added templates and guidance, teams handed this infrastructure produced things like 72-item agendas for one-hour meetings with 30 people. Nothing about that is logistically or structurally sound, and they didn't have enough meeting competency to see that.

The templates helped. The guidance helped. But the underlying problem couldn’t be solved with better buttons or self-serve learning.

The most significant barriers to getting operational value from meetings remain social. You have to run a better meeting, your team has to engage in that meeting, and all the other groups you interact with have to be cool with upping their meeting game, too.

The step where humans learn what structures they need to get reliable results, and then develop the skills to operate within those structures, may be aided and facilitated by AI going forward. It cannot be replaced by it.

Most AI note-making tools don't support active in-meeting engagement well. The collaborative infrastructure is starting to appear in enterprise platforms, but most teams aren't operating at the meeting maturity level needed to use those effectively. Yet.

For most groups, the highest-leverage move available right now is also the simplest: stop the meeting five minutes before the end, let the AI generate its notes, and run this loop together as a group:

This is what good human note takers have always done. AI generates a faster first draft. Then, you need to take a few minutes to own what it says.

Until teams build the meeting competency to use richer structures effectively, this five-minute loop at the end of every meeting can fill the gap between analytical fodder and notes that help get work done.

If you’d like to explore how we can help your organization, please visit our website here.