How do you diagram a multi-agent AI system?

The Router Agent reads the request and picks who gets it. Behind it sit three specialists: the Research Agent searches, the Code Agent writes and runs code, the Analysis Agent works over data. From the user’s side there is one box. Three do the work.

Braces are what put the specialists in a column instead of a row. All three land in the column after the router, top to bottom in the order you typed them, and that layout says something a chain cannot: these are alternatives at the same level of the system. A reader who assumes the agents run in sequence will ask why analysis has to wait for research, and with the three stacked in one column that question does not come up.

One thing the braces get wrong, and it is worth knowing. In the animation the request goes to all three specialists in the same pass, because braces mean everything at once, and a router usually picks one, so the picture shows a broadcast where the system does a dispatch. If you want it to route, write the same line with bars, |Research Agent, Code Agent, Analysis Agent|, and each pass goes to a single specialist, cycling through them in order. Boxes and column stay where they are. Only the arrows change.

I think routing is the first structural change most single-agent systems need. Any prompt that has to cover research, code and analysis ends up mediocre at all three, so you split it, and each specialist gets its own instructions and its own tools. After that the router is the part you have to get right, since every request goes through it and a misroute hands the job to a specialist with the wrong tools.