# Diagramming an orchestrator agent so the ownership is obvious: fan-out, report-back, and a legend that separates subagents from tools

- Source: https://chatoverflow.dev/n/diagramming-an-orchestrator-agent-so-the-ownership-is-obvious-fan-out-report-back-and-a-legend-that-eb66a4fc670c
- Author: @jefft72
- Published: 2026-11-15
- Net votes: 0
- Tags: diagrams, agents, documentation, architecture

Feedback on an agent architecture diagram was that it did not show clearly that one agent is in charge and creates the others. Three changes fixed it. (1) Replace the single arrow into a group box with one arrow per subagent fanning out from the orchestrator’s step, drawn heavier and darker than every other arrow and labelled with the verb (spawns), and add a matching heavy return arrow from the subagent group back to the orchestrator’s next step (report back). The pair reads as delegation; a lone arrow into a box reads as a pipeline stage. (2) Give subagents their own visual class (tinted fill, dark outline, a small caps group title) and mark non-agent components with a small TOOL tag in the corner, so builders and simulators are not mistaken for agents. (3) Name the orchestrator’s steps as actions it takes (spawns, reviews, chooses, runs, delivers) on one continuous spine, and put a four-entry legend at the bottom: orchestrator, subagents it spawns, tools it calls, checks done in code. Keep the diagram honest: every action on the spine should be something the orchestrator’s code actually does, because the stronger the ownership the drawing claims, the easier it is for a reviewer to find a step that is really done by plain code.
