Choosing Where to Look Across Agent Terminals

This is a summary of the full Korean builder diary.

When I sat down to explain agent mosaic, I started with a question rather than a feature name: with several agents working, which terminal should I look at first?

It is easy to keep watching the last terminal I opened. New output there can make the whole workflow feel busy. But one visible output does not tell me what is happening in the other terminals. I want to choose where to look before trying to read every detail.

Look across, then choose one

In Marblo's Fleet agent view, choose In progress, then Mosaic. The view shows actual terminal output from multiple agents together. Mosaic has been included since v3.0.43.

The previews are read-only, with up to nine cells. When there are more agents, the view indicates the overflow count. Headers show a task title, or the agent name when there is no title, together with model and agent labels. Long titles may be shortened.

To inspect one agent, click the preview's header to open its individual view. Selecting a header and pressing Enter or Space makes the same selection. I think of these as two steps: choosing what to inspect, then reading that output closely.

What I still need to check

This is a way I propose to observe the work, not a claim that an overview verifies it. Seeing output move is different from reading its meaning, and reading output is different from checking the result. I want my explanation to keep those steps separate.

For the demonstration, I plan around four harmless example tasks. That is a filming choice, not a usage metric or a performance claim. The point is to make the selection flow legible, rather than fill every available cell.

One line: an overview is a starting point for choosing where to look next.

Which terminal do you inspect first when several agents are working? If these builder logs are useful, subscribe for the video entries too.