Event-driven architecture is often presented as the answer to scalable and decoupled system design. In practice, however, moving to an event-driven model introduces a new set of problems: duplicate event delivery, inconsistent state, failed consumers, replay issues, retry storms, and poor visibility into what the system is actually doing. This talk explores what it really takes to build event-driven systems in Python that can survive production. Rather than focusing only on theory, the session will cover practical patterns and techniques that engineers can apply immediately, including: - designing idempotent consumers so duplicate events do not corrupt state - handling retries safely without creating cascading failures - using dead-letter queues and failure isolation effectively - applying the outbox pattern to reduce consistency issues between databases and message brokers - thinking through ordering, delivery guarantees, and eventual consistency - adding observability through logs, tracing, and metrics so event flows can actually be debugged The talk will discuss common trade-offs in event-driven design and show how Python can be used to build systems that are not only asynchronous and scalable, but also predictable and resilient. Attendees will leave with a clearer understanding of what works, what usually goes wrong, and how to approach event-driven design in Python with a production mindset.