the Skyflo log

An AI Meeting Agent That Speaks and Takes Action

How to choose an AI participant that answers the room and starts work during a call, with a practical first-meeting brief.

Read the full page:Skyflo in your meetings
·3 min read·meetingsai-agentsvoice

An AI meeting agent that speaks and takes action needs two things: a voice in the call and access to the work the room wants done. Joining a meeting is only the beginning. The useful question is what happens when someone asks it to investigate, change or show something.

With Skyflo Meetings, a Google Meet, Zoom or Microsoft Teams call connects to one existing mission. A spoken request becomes work on that mission. Its agent can use the tools and context available to it, and the mission can keep working after Skyflo leaves the call.

Decide which kind of help you need

A transcript helps you recover what was said. A private copilot helps one person prepare an answer. A speaking participant answers the room. An agent that executes work also needs a workspace, tools and a clear permission boundary. Products can combine these capabilities, so evaluate the actual workflow rather than the label.

Speaking agents are already a category. Otter describes a voice-activated meeting agent that answers questions and completes tasks. Harvest connects an existing agent to Google Meet. Skyflo's approach connects the conversation to a mission that also holds the work, its history and its approvals.

Give the call a useful first task

Start with a question that has a checkable answer. For example:

Check the launch brief against the current product documentation. List unsupported claims and draft corrections. Leave publication for review.

Before the call, put those documents in reach of the mission. Add the goal, constraints and decisions already made to the meeting brief. Choose whether to save a text transcript. Then use Bring Skyflo into a call, paste the meeting link and choose Join call. The host admits Skyflo as a participant.

During the discussion, address Skyflo by its selected voice name. Ask what it found, redirect the task or give it the next piece of work. A useful response separates a checked fact, a proposed change and a result still waiting for verification.

Evaluate the work and the controls

Use your first call to check four things:

  • Can it answer from the material you supplied?
  • Does a request produce an inspectable output in the mission?
  • Can the room interrupt or stop its participation?
  • Can you follow unfinished work after it leaves?

Everyone admitted to the call can direct the attached mission. Choose its scope and access accordingly. The room can allow an eligible approval once or reject it; wider permission changes stay in Skyflo on the host's Mac.

Skyflo records no audio or video. While it listens, what is said is sent from the host's Mac to OpenAI. Saving a local text transcript is a separate choice, off by default. Requests acted on remain in the mission.

Start with one meeting

Choose a working session with a small, useful deliverable. For engineering, try investigating a blocker during the call. For a customer conversation, try a prepared sales workflow.

The Meetings page has current requirements and pricing. The launch essay explains how meeting participation works in more detail.

An AI Meeting Agent That Speaks and Takes Action | Skyflo