AI Agents Are Joining the Meeting: Four Things to Agree on Before You Let One In
A team member who is not a person is becoming a normal part of the meeting. On October 7, at its annual WebexOne conference, Cisco announced that AI agents can be added directly to Webex spaces, meetings and calls, where colleagues can @mention them and hand them work. AI agents in meetings are no longer just tools that tidy up a transcript afterwards. They sit in the discussion, hear the context and can act on it. For small businesses and brand teams, the question is no longer whether to use them. It is whether the rules are agreed before they come in. This article covers what changed and the four things we suggest a team settles first.
What happened: AI agents went from personal assistant to meeting participant
Until now, AI assistants were mostly one-to-one. You asked, it answered, and nobody else saw it. Cisco is going a different way. According to Computerworld, users can @mention an agent in a Webex group chat to schedule meetings, take notes and track tasks, and the whole team can see the result and build on it. Cisco's own example: ask an analytics agent to pull usage figures, then ask a presentation agent to update the slide deck with them.
The agents do not have to be Cisco's. Anthropic's Claude Managed Agents can be added to group conversations, and OpenAI's recently announced dots agents are supported too. There are also specialised roles: a personal agent that screens your calls and judges urgency, a translator agent that works in real time across ten languages, and agents that keep working in the background for a long stretch and come back for your approval.
Mind the timeline. SiliconANGLE and Computerworld both note that most of the new features are planned for early 2027, with the translator agent due for general release in November. In other words, now is the time to think the rules through.
Three things AI agents in meetings change
First, AI output goes from private to public. When AI got something wrong before, only you knew. Now an agent's reply in a group space is seen by the whole team and may be used straight away. Mistakes travel faster.
Second, AI hears the whole discussion. To be useful, an agent has to read the context of the conversation. That means the quotes, staffing matters and client details raised in a meeting can all become material it handles.
Third, AI keeps working after the meeting. The call ends and the agent is still updating documents, sending messages and arranging follow-ups. Irwin Lazar, principal analyst at Metrigy, points out that an agent's usefulness depends on access to external data such as CRM and ERP systems. The deeper the connection, the more useful it is, and the more it needs managing.
This is the same shift we described in AI That Works While You're Offline: What to Settle Before Hiring a Persistent AI Agent: AI moving from "acts when called" to "always there."
Four things to agree on before an agent joins
1. Who can see it, and who knows it is there
Computerworld quotes a Cisco spokesperson: agents in spaces are shared with the team, while agents in the call side panel are visible only to the individual user. That difference matters. If an AI the other side does not know about is listening and taking notes during your call with a client, that is a trust problem. Say it up front. In any meeting with external participants, explain at the start that an AI agent is present and what it will do. If they object, turn it off.
2. What data it can touch
List what the agent may read, then list what it must never read: salaries, contracts, unpublished quotes, customer personal data. Cisco says identity, security, governance and visibility controls will determine what agents can access and do. A tool having controls does not mean your team has configured them. Start with the minimum and add access only when needed.
3. What always needs a human to approve
Replying to a message, editing a deck, booking a meeting: each sounds small, but once it goes outside the company it is the company's position. Split the work into two clear groups. What the agent can finish by itself (notes, action lists, drafts), and what a person must confirm (external emails, changes to official documents, anything involving money or commitments). Snorre Kjesbu, Cisco's senior vice president for collaboration, also stresses that human judgment and accountability must stay at the centre. Every agent should have a named owner.
4. What it costs, and how it is charged
Computerworld reports that advanced features such as collaborative agents and the translator agent require an add-on, and Cisco has not disclosed the price. The translator agent includes 50 minutes per user per month, with more requiring a further add-on. Before you roll out, ask which features are in your current plan, which are billed by usage, and how usage caps are set.
How a small team can start: pilot one kind of meeting
You do not need to switch everything on. Pick one regular, low-risk internal meeting, such as the weekly project update, and let the agent do three things only: take notes, list actions, and send reminders afterwards. Run it for four weeks, then ask three questions. Were the notes accurate? Were the actions really followed up? Did anyone hold back because an AI was in the room? The last one matters most. A meeting is only worth having if people are willing to say what they think.
If several AI tools are already running in your company, do an inventory first. Our guide is here: How Many AI Agents Are Running in Your Company?
Conclusion: let AI get the notes right, let people get the conversation right
AI agents in the meeting save the time spent on notes, tidying and chasing. What really matters in a meeting, such as hearing the concern a client did not voice, making the call when people disagree, and owning the outcome, is still human work. That is what TitApeX means by "AI perfects, people move."
If you want to plan how AI tools come into your team, or need your internal systems put into a shape AI can use, take a look at our services or talk to us.
Comments