The extensibility is genuine. When an order-placed event already exists, adding an inventory update, a loyalty calculation and an analytics record means adding three subscribers and modifying nothing. In a request-response design, each of those would be another call inside the ordering code.
What you trade away is traceability. There is no call stack showing what happened in response to an event, so understanding a flow means reading subscriber registrations across services. Without distributed tracing and a documented event catalogue, teams reach a point where nobody can confidently say what a given event triggers — which is when the decoupling starts costing more than it saves.