Directed Acyclic Graph (DAG)

Think of a Directed Acyclic Graph less like a subway map and more like a corporate approval chain. A staff member can send a document up to a team lead, and the team lead up to an executive, but the executive never sends approval back down to the staff member. Arrows only flow one way, and there's no loop that circles back to the starting point—hence "directed" (arrows point a fixed way) and "acyclic" (no loops).

This structure shows up when AI handles work broken into multiple stages. Say there's a task to research some material, a task to summarize the results, and a task to review that summary—the research has to finish before summarizing can start, and the summary has to finish before review can start. Mapping out this order and these dependencies with arrows makes it clear at a glance which tasks need to happen first and which ones can run at the same time.

This matters especially when multiple agents work simultaneously. If a loop exists, one task can end up waiting forever for another task's result—a deadlock. Removing loops entirely means that problem simply can't occur, by design.

Try writing out any task step by step, then draw arrows next to each step showing what has to finish before it can start. If following the arrows ever leads back to a step you already passed, that loop is a sign something's wrong with how you've ordered things.