Event-Driven Architecture
System design where components react to events (webhooks, queue messages, database changes) instead of polling, enabling real-time and decoupled workflows.
Last updated: April 26, 2026
Definition
Event-driven architecture (EDA) inverts the polling pattern. Instead of services asking each other for updates on a schedule, they emit events when things change and other services react. The classic patterns are: pub/sub (one publisher, many subscribers), event streaming (Kafka-style ordered logs), and webhooks (HTTP callback per event). For AI agents, EDA is the right design when an agent should respond to real-world signals (a customer creates an order, a PR is opened, a payment fails) rather than running on a schedule. The trade-off: better latency and decoupling, but harder debugging because the trace lives across many services.
When To Use
Pick EDA when latency to event matters (under 1 minute), when many services need to react to the same signal, or when polling would impose load on the source system. Use scheduled batch for everything else.
Related Terms
Building with Event-Driven Architecture?
I've shipped this pattern in real production systems. If you want a second pair of eyes on your architecture, that's what I do.