Steer a meeting with topics and intent

A long conversation is hard to steer from the scrollback alone. A meeting keeps two structures alongside the chat so you can see the shape of it: topics, which group what has been said, and the intent model, which records what it means.

Topics

Select the meeting title to open topics. Each card is a cluster of related messages with a label, a one-line summary, and a status of active or resolved. Open one to read its messages, and select a message to jump to it in the chat.

The number next to the title is how many topics the meeting has. Topics are grouped automatically as the conversation grows, every few messages, so a meeting that has just started does not have any yet.

Use them for the thing scrollback is bad at: coming back to a meeting an hour later and finding the three minutes where the audience question was settled.

The intent model

The panel on the right of the meeting is the discussion in structured form. It has five tabs, and each holds items you can act on rather than read only:

Tab	What it holds	What you can do
Questions	What is unresolved, and which of those block the work	Answer, close, or add one
Decisions	What has been settled, and what an agent proposes settling	Confirm, reverse, or dismiss
Reqs	Functional requirements, non-functional requirements, acceptance criteria	Add, edit, remove
Scope	In, out, and the parking lot	Add, move, remove
Risks	What could go wrong, and its status	Add, resolve

Eva writes these from the conversation as it goes, and the confidence of each item comes from how clearly it was said. Your edits win: confirming a proposed decision makes it a decision, and dismissing one makes it go away.

The count on a tab is the count that matters for that tab, not the total: open questions, proposed decisions, open risks.

The readiness ring

The ring at the top of the panel scores how well defined the discussion is out of three things:

1. Requirements: at least two, of any kind.
2. No blocking questions: nothing marked as blocking is still open, and there is at least one requirement for them to block.
3. Decisions or scope: at least one decision, or at least two scope items.

Each is worth about a third. It is a gauge for you, not a lock: you can end a meeting at any score. A low ring late in a meeting usually means the same thing every time, which is that nothing has actually been decided yet, and the task drafts it produces will show that.

Making the model do work

Two habits change the output more than anything else:

• State decisions as decisions. “Let me lock in what we have decided” gives Eva something to record. A meeting full of open threads produces drafts full of open threads.
• Answer the blocking questions in the panel. They are the ones an agent said it needs before it can start, and they travel into the task spec if you leave them.

Next steps

• Proposed tasks: what the intent model turns into when the meeting ends.
• Meeting summaries: the written account of the same conversation.
• Meetings: why work starts as a conversation at all.