I’m excited about Claude joining video calls because I can already picture the product work I’d want to get started while the conversation is still happening.
For a while now, I’ve been building skills and workflows that take meeting transcripts and turn them into useful outputs: structured notes, recaps, decisions, and action items. My biggest lesson has been surprisingly simple. A clear objective and a thoughtful agenda do a lot of the heavy lifting.
If everyone knows what we’re trying to accomplish, the conversation gives the AI something useful to work with. If we spend an hour wandering, we’ve mostly created a very detailed record of wandering.
The agenda is part of the input
Before a product discussion, I want to know what decision we need to make, which questions we need to answer, and what we should leave with.
That structure carries through to the recap. It helps distinguish an idea someone floated from a decision the team actually made. It gives action items context. It makes missing owners and unresolved questions easier to spot.
My ideal meeting brief is straightforward:
Objective: What outcome do we need?
Agenda: Which questions get us there?
Decisions: What needs an explicit yes or no?
Output: What should exist when we finish?
Ownership: Who is responsible for the next step?
That is the foundation I’d bring to an AI participant, too.
Now imagine the work taking shape during the call
In its November 12 announcement, Anthropic introduced Claude Meetings in beta, starting with Google Meet for Team and Enterprise plans. Claude can join as a visible participant, work on documents and other artifacts during the conversation, and use connected tools with permission. Read the announcement.
Here’s where my product brain goes.
Imagine a team discussing a new onboarding experience. We agree on the customer problem, define the first release, and work through what success looks like.
While we’re talking, I’d want Claude to assemble a draft requirements document, identify unanswered questions, and propose acceptance criteria. With the appropriate connections and authorization, I’d want to ask it to retrieve the existing requirements and prepare updates for review.
Then we could look at the draft together while everyone is still present.
“Actually, that requirement only applies to new customers.”
“We haven’t agreed to that deadline.”
“That’s an option we discussed. Keep it out of the committed scope.”
Those corrections are valuable. The team gets to inspect how its conversation is becoming work before everyone moves on to the next meeting.
That is the workflow I’m imagining, rather than a claim that every step is already available in the beta.
Make the handoff part of the meeting
The opportunity I see is to shorten the gap between agreeing on something and creating the materials needed to act on it.
A recap helps me remember. A requirements draft helps the team respond. A clearly stated open question gives someone a specific problem to resolve.
For this to work well, I’d want explicit checkpoints. Confirm the decision. Confirm the owner. Mark what remains a draft. Ask before committing a change to a shared system.
Meeting language is messy. “We could” and “we will” can sit two sentences apart. A useful assistant needs the team’s help separating them.
This is also why I’m interested in practical AI agents for small businesses: there is so much useful work sitting between a conversation and the next action.
Holy token consumption, though
That was my other immediate reaction.
Imagine continuous audio understanding, meeting context, connected-tool lookups, and document revisions happening throughout a call. How much processing does that take? When does live assistance earn its cost, and when would a transcript afterward do the job?
I’m especially curious about the economics of true audio modality. But the announcement doesn’t establish that Claude Meetings uses an end-to-end audio-native model, and it doesn’t disclose per-minute token consumption. I wouldn’t turn that curiosity into a cost estimate.
The published beta terms are more concrete: notes, transcripts, and answers without connected-tool use are free on Team and Enterprise plans during beta. Connected-tool answers count toward plan usage. In-call artifacts are free through February 10, 2027.
For my own evaluation, I’d measure how much usable work the team leaves with, how much correction it needs, and how much follow-up effort it saves.
I’m excited to try this with a focused product session: one objective, a clear agenda, and a defined output. Give Claude a useful brief, let the team shape the work live, and see how much further along we are when we hang up.
