As multi-agent systems scale beyond simple request-response patterns, native publish/subscribe semantics become essential infrastructure.
The Agent-to-Agent Protocol currently uses a request-response model: Agent A sends a message to Agent B and waits for a reply. This works well for direct delegation but becomes inefficient when multiple agents need to react to the same event.
Consider a shopping assistant scenario:
Without pub/sub, the price-monitoring agent must individually notify each interested party, manage delivery confirmations, and handle failures. With pub/sub, it publishes once to a "price-alerts" topic, and subscribers receive the event automatically.
The feature request outlines several key capabilities:
Design tension: A2A is intentionally minimal — adding pub/sub primitives significantly expands the protocol surface. The community will need to weigh simplicity against capability.
Implementing pub/sub in a decentralized agent ecosystem raises interesting challenges:
Broker architecture: Who runs the message broker? Options include:
Discovery: How do agents find topics of interest? This connects to the ongoing Agent Card discussions — perhaps topic catalogs become part of agent capabilities.
Security: Who can publish to a topic? Who can subscribe? The identity and authorization discussions (see AIAR Identity Extension) become even more critical with broadcast messaging.
Event-driven patterns unlock several use cases that are awkward with request-response:
The proposal is in early discussion. Given the scope of the change, expect:
For implementers building multi-agent systems today, this is a signal of where the protocol is heading. Even if native pub/sub takes months to land, designing with event-driven patterns in mind will ease future migration.
Follow the discussion: github.com/a2aproject/A2A/issues/1593