Without a router, choosing the right process is someone's job at creation time. A person picks a template, or a rule matches on fields like type or tags. That works until the categories blur: a "task" that is really a code change needing review, a "bug" that is actually infrastructure. A router turns that choice into a step of its own, done by the agent with the work in front of it, and recorded.
The pattern has three parts. A triage step confirms what kind of work this is. A decision step offers a small number of exits, each naming the sub-workflow it starts, such as feature delivery, a code change with review, a small known task, or an operations change. After the chosen sub-workflow finishes, control returns to the router, which can run closing steps that apply to all work, like capturing what was learned.
Routers keep each downstream workflow small and specific, because none of them has to handle every case. They also make the routing auditable: the decision records which exit was taken and why, so a reviewer can see that a change went through the review path and not the quick one.