The gap this closes
Meetings generate work and lose it. Someone says “we should check that”, everyone agrees, and the sentence dies when the call does. The gap between a decision and a task is where most meeting output evaporates — not because nobody cared, but because capturing it competes with paying attention.
An agent that is in the meeting closes that gap differently from one you brief afterwards: it hears the sentence in context, and can act on it before the call ends.
What it does on a call
Joins and discloses. The agent enters as a participant and announces its presence — people should know when something is listening. It can be dismissed by voice.
Listens for its name. It is not transcribing-and-responding to everything. A wake-word match plus a turn-taking loop means it stays out of the way until addressed, then answers in the conversation rather than over it.
Knows your organisation. The agent carries context about your roster of characters and your workflow catalogue, so “ask the analyst to look at that” resolves to something real rather than a generic answer.
Starts real work. Asked to do something that is a workflow, it dispatches one — scoped to your organisation — acknowledges out loud that it is working, and speaks the result when the run returns.
Summarises at the end. The wrap-up is itself a workflow run, which means the summary has the same record behind it as any other work Forge does.
What you control
- Whether it can act, or only answer. Dispatching a workflow from a meeting is a write; the same autonomy budget that bounds any chain bounds this one.
- Which workflows are reachable. The in-meeting catalogue is your organisation’s, not a general-purpose tool list.
- Consent. The agent announces itself on joining. Whether it is in the room at all is your call, every time.
Under the hood
- Speech in and out — transcription and synthesis through the same gateway as everything else
- Web Browsing — how the browser session under the meeting is driven
- Workflow Engine — what an in-meeting request actually starts
- Git-Backed Audit — the record behind the summary and any run it triggered