What is an AI meeting participant?

An AI meeting participant is an AI agent that joins a video call as a peer rather than as a recording tool. It can speak in real time, see a shared canvas or screen, remember context from prior sessions, and take actions inside the meeting — such as logging a decision, drafting a doc, or filing a ticket. The distinction matters: a notetaker observes the meeting; an AI meeting participant operates within it.

How is an AI meeting participant different from an AI notetaker?

An AI notetaker is a passive observer that joins as a bot, transcribes the call, and produces a summary afterward. An AI meeting participant is active during the meeting. It can interject when relevant, answer a question about an earlier decision, or take action on a request without waiting for a post-meeting workflow. Notetakers produce a record. Participants close the loop in real time.

Are AI meeting participants legal under wiretap laws?

That depends on consent, jurisdiction, and architecture. The May 20 Otter motion-to-dismiss hearing is the first federal test of whether ECPA applies to AI bots that join meetings via calendar auto-join. AI meeting participants that operate as native platform features — with explicit consent flows and no calendar-bridge bot — generally have a cleaner compliance profile than third-party bot-based notetakers.

What should teams do this quarter to prepare?

Three things. First, stop measuring AI value by transcript quality and start measuring it by in-meeting outcomes (decisions logged, follow-ups assigned, alignment achieved). Second, audit the bots currently joining your meetings — Microsoft's May 2026 bot detection rollout will surface them anyway. Third, choose a meeting surface that can host an AI meeting participant natively: voice, vision, memory, and action capability inside the conversation.

AI Meeting Participant: The Next Wave Beyond Notetakers

Q: Will AI meeting participants replace human teammates?

No. The pattern shipping in 2026 is multiplayer AI: multiple humans plus one or more AI participants sharing the same visual and conversational context. The AI participant takes a defined role — facilitator, scribe-and-decider, specialist consultant — that augments rather than replaces the humans in the call. Meetings without humans become async docs. Meetings with humans get AI as a peer, not a replacement.

The AI meeting participant has arrived — and the AI notetaker era is ending faster than anyone predicted. After eighteen months in which 75% of professionals adopted an AI notetaker and meeting volume dropped just 25%, the bot-in-the-lobby model has maxed out. In the last seven days alone, four signals landed inside one industry: OpenAI and Anthropic both launched enterprise AI service joint ventures on May 4. Anthropic shipped Claude Opus 4.7 with ten pre-built financial agents twenty-four hours later. ServiceNow announced its "AI Action Fabric" with pay-per-agent-action pricing on May 5. And on May 11, a16z, General Catalyst, and Accel poured $20M into Ciridae with the most quotable thesis of the year: "Most AI tools are built for one human and one model in a private workspace, but almost none of the output is shared, aligned, or contextualized across a team."

That sentence is the obituary of the AI notetaker.

The next wave is the AI meeting participant — an agent that does not lurk in the lobby with a "recording" badge, but speaks, sees, remembers, and acts inside the conversation. This piece is about what an AI meeting participant actually is, why the category shift is happening this quarter, and what teams should change before the rest of their stack does.

The notetaker era ended last week

For two years, the AI meeting participant arrived disguised as a notetaker. A bot joined your Zoom or Meet, recorded everything, and produced a transcript that nobody read. That model is being dismantled in three places at once.

First, Microsoft confirmed on May 11 that Teams Copilot will continue to require a visible transcript for post-meeting recall. The "magic AI listening silently in the background" experience that enterprise buyers were promised in 2025 is not shipping. If the AI cannot work without a transcript, the transcript is the product — and that is a notetaker, not a participant.

Second, Microsoft Teams bot detection rolls out to Targeted Release tenants this week (MC1251206). External bots get an "Unverified" label in the lobby. Organizers must explicitly admit them. Admins can auto-block them at the tenant level. Every bot-based notetaker — Otter, Fireflies, Read AI, Fathom, tldv — is one IT policy away from silently failing inside Fortune 500 calls.

Third, Krisp launched VIVA 2.0 at Twilio Signal on May 6 and pivoted hard to voice-AI infrastructure for builders like Vapi, LiveKit, Daily, and Ultravox. The company that defined "clean audio for AI notetakers" has left the human-meeting market for the voice-agent gold rush. They are not betting on the bot in the lobby. They are betting on the agent in the conversation.

When the platform vendor, the IT layer, and the audio infrastructure pivot in the same week, the category has moved.

"75% adoption, 25% meeting reduction" is the canary

Adoption is not the problem. Outcomes are.

Fellow's 2026 State of AI Meeting Assistants survey found 75% of professionals now use an AI notetaker in work meetings. Fireflies alone claims 75% Fortune 500 penetration. Yet meeting volume only dropped by 25% over the same period. The bots are an output-multiplier, not an input-reducer.

Zoom out and the pattern repeats. McKinsey's State of AI finds 88% of organizations have deployed AI in at least one business function, but only 5.5% report measurable financial returns and only 1% of leaders consider their company "mature" on AI deployment. Gallup's State of the Global Workplace 2026 reports US engagement just hit an eleven-year low at 31%, even as desk-worker AI use grew 233% in six months.

This is not an AI problem. It is a participant problem. Today's AI meeting participant arrives after the meeting, in the form of a tidy summary of a confused conversation. The work — alignment, decisions, ownership — was never done in the first place. The transcript just documents the gap.

If you bolt an AI notetaker onto a broken meeting culture, you get more documented broken meetings. The 5.5% of companies that report real returns redesigned the workflow first. They did not buy a Copilot license and hope.

Three reasons the AI meeting participant left the lobby

Transcript theater

There is a Medium post from Traci Levine that crystallized the new mood: "Have you ever walked into a meeting with two people and five bots? That happened to me last week." Rebecca Hinds put it more sharply: "There's nothing that says your time matters more than the other people in the room than sending your bot instead of showing up yourself." The bot-attendee era turned into a passive-aggressive status signal — and the receivers noticed.

The output is what one builder called a "clean summary of a confused conversation." The doc looks great. The team still disagrees about what was decided. Transcript theater is the worst-case version of an AI meeting participant: it documents the problem instead of solving it.

Output multiplier, not input reducer

A 75% adoption rate that only cuts meetings by 25% is a tell. Notetakers do not decide what is a meeting — humans do. Hand a team a perfect summarizer and they will book more meetings, not fewer. The economics of an AI meeting participant only flip when the AI affects what happens during the call: a decision logged, a follow-up assigned, a context surfaced before the question is repeated for the third time this quarter.

Compliance gravity

The Otter motion-to-dismiss hearing scheduled for May 20 is the first federal test of whether ECPA wiretap statutes apply to AI bots in meetings. If Judge Lee denies the motion, every notetaker built on calendar-auto-join inherits the legal exposure. The calculus shifts overnight. The cheapest AI meeting participant is a bot in the lobby. The most expensive one is a bot in the lobby that triggers a $10,000-per-violation ECPA claim.

What an AI meeting participant actually is

Strip the marketing copy and a real AI meeting participant has four capabilities the notetaker does not. None of these are theoretical — every one is shipping in production at some platform this quarter.

Speaks

Voice infrastructure is the unlock. Krisp's VIVA 2.0, with sub-200ms turn prediction and interrupt detection, makes it possible to put a voice agent into a meeting that does not sound like a phone tree. Tavus's AI humans and MindStudio's Pika Me have validated the real-time conversational pattern. The AI meeting participant is no longer a transcription pipeline — it is a voice in the room.

Sees

This is where the canvas matters. An AI meeting participant that can read a shared whiteboard, a diagram, or a screen is one that can answer "what did we just decide about the auth flow" in the moment it matters. Without a visual surface, the agent is limited to what gets verbalized — which, in any real working session, is roughly 30% of the actual content. The agents shipping this year all share a visual loop: see the canvas, hear the room, talk back.

Remembers

Persistent memory across sessions is the second unlock. The state of AI agent memory in mid-2026 is a patchwork — most products store transcript-shaped memory rather than relationship-shaped memory — but the direction is clear. An AI meeting participant that knows what was decided in last week's call, who owns what, and what the team has tried before is a fundamentally different entity from a fresh-context bot that joins, records, and forgets.

Acts

The frontier capability: take an action mid-conversation. File the ticket, send the calendar invite, update the CRM record, draft the doc — without leaving the meeting surface. This is the part the foundation lab JVs are racing to ship. ServiceNow's Action Fabric is one approach. Anthropic's pre-built financial agents are another. The participant who can act is the one who closes the loop.

The collaboration layer is the moat for the AI meeting participant

Here is the most consequential idea in the AI workspace this quarter, from the Ciridae round announcement on May 11: "The collaboration layer becomes the moat." That sentence reframes the entire stack.

The foundation labs build the brains. The vertical agents (Sierra, Decagon, Harvey) do specialized work. But none of those are where humans meet to align, decide, and hand off. That surface — the place where people and their agents converge — is the contested ground.

For two decades, the meeting surface was a commodity. Zoom won by being good enough. Google Meet won by being free enough. Teams won by being integrated enough. None of them were the moat — they were the dial tone.

That has now flipped. With AI participants becoming the unit of work, the surface that hosts them is the most valuable real estate in the stack. Slack, Notion, Asana, Linear — every team-OS contender — is being repositioned this year as a collaboration layer for agents. Even Zoom's AI Companion CX going cross-platform into Google Meet, Teams, and WebEx is an admission that the meeting surface itself, not the video platform, is what holds the value.

The AI meeting participant only works if the surface lets it. That is why the surface is the moat.

From AI behind the meeting to live AI in the meeting

Most workplace AI today is behind the meeting. It pre-reads documents. It writes prep notes. It generates post-meeting summaries. It pings you about action items the next morning. The meeting itself remains a 2019 artifact with a transcription bot taped to the wall.

The shift in 2026 is to AI in the meeting. Real-time. Visible. Participatory.

Multiplayer AI for teams is the architectural pattern: multiple humans plus one or more AI participants sharing the same visual, conversational, and decision context. The notetaker era treated AI as a passive observer. The participant era treats AI as a peer with a defined role: facilitator, scribe-and-decider, specialist consultant, devil's advocate. Different role per meeting type — the way you would assign humans.

A practical implication: meeting templates start to matter again. The kickoff template, the design review template, the customer call template — each implies a different AI meeting participant configuration. Not a different bot. A different role.

What this means for the next 12 months

Four predictions worth holding to.

1. The notetaker category consolidates. Granola, Fathom, Fireflies, Otter, and Read AI cannot all survive a world where Microsoft Teams blocks bots by default, ECPA exposure is real, and the value has moved to in-meeting action. Expect 50% of the category to fold or pivot into voice-agent infrastructure (Krisp's path) within twelve months.

2. The platform vendor takes the participant slot. Microsoft, Google, Zoom, and the emerging native AI-first meeting platforms will eat the AI meeting participant feature inside the surface. Third-party bots survive only as specialized agents (legal, sales, compliance) attached via verified APIs.

3. The canvas becomes table stakes. A meeting platform with no shared visual surface cannot host a participant that sees. Pure-video tools will retrofit a canvas, fail to make it good, and lose share to the video conferencing platforms that natively combine canvas, video, and AI.

4. Pricing detaches from seats. ServiceNow's pay-per-action model is the first signal. Meeting platforms will charge per AI participant per session within the year. Seat licenses will not disappear — but the marginal AI participant unit is what will fund the next stage of product.

The window is open right now

Categories shift when a generation of buyers stops believing the incumbent narrative. That moment is happening this quarter. The notetaker stat — 75% adoption, 25% meeting reduction — is the canary. The Otter trial, the Microsoft rollback, the Krisp pivot, and the Ciridae thesis are the rest of the choir.

Teams that get ahead of this do two unglamorous things. They retire the notetaker-as-feature thinking and ask what an AI meeting participant should do in each meeting type. And they pick a meeting surface that can host one — with voice, vision, memory, and the ability to act inside the conversation, not after it.

The bots in the lobby were never the destination. They were the bridge to here.