Multica

With the first agent everything is clear: one terminal, one task, you watch. By the third it tips over. Each one lives in its own window, forgets everything when the session ends, and a growing part of the day goes into keeping track: who is working on what, what did it actually run, is it done.

Multica is the layer above that - a board where an agent is entered as the assignee like a colleague. Anyone working with Hermes knows its kanban view; Hermes is one of the twenty runtimes here. The Go part is what pays off, because that is where the operational questions sit as packages of their own: who gets assigned a task, which machine it runs on, what is retried after a failure, and how a run is logged so it can be replayed afterwards.

That puts the same question here as in the building block on Tiny Agents, one floor up: there, which tool an agent gets to see - here, which agent gets to take a task. Worth the look as a template even without adopting it.