Stefan is a Principal Engineer at AWS on the Amazon MQ team where he obsesses over a large fleet of RabbitMQ and ActiveMQ brokers. He has been an advocate and teacher of messaging patterns and distributed systems design for well over a decade. When his fingers are not on a keyboard, they’re likely wrapped around the handlebars of his mountain bike, holding on for dear life while hurtling down one of the North Shore’s world-famous trails.
Every agent framework is quietly reimplementing broker capabilities from scratch. Routing? Reimplemented per-framework. Retries? Each one rolls its own. Discovery, fan-out, backpressure, dead-lettering, durable execution: all reinvented in application code, inside a single process, with predictable results. Message brokers have solved these problems for decades, but today’s brokers weren’t designed for consumers that are stateless, non-deterministic, and expensive to invoke. Both sides need to evolve. This talk identifies the five properties a message broker must take on to serve agentic workloads natively, so agent frameworks can stop pretending to be infrastructure and get back to being applications.
Key Takeaways:
Target Audience:
Amazon proposed using the AMQP messaging protocol as a transport for MCP at the inaugural MQ Summit in 2025. Ken Liao’s talk and demo sparked significant interest and follow-on discussion at last year’s summit leading to consensus to engage the AMQP community and ambition to extend the reach of the OASIS standard. Join Clemens, Stefan and Richard for an update on what’s been going on behind the scenes to take this vision to the next level and, most importantly, learn how you can get involved.
Key Takeaways:
Target Audience: