5 Capabilities Your Broker Needs for Agentic Workloads

Speaker:
Stefan Moser


Speaker:
Vignesh Selvam


Abstract:

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:

  • Attendees will learn to recognize where agent orchestration frameworks are reimplementing messaging infrastructure,
  • understand the specific demands that agentic workloads place on brokers that today’s designs do not address,
  • and leave with a clear picture of the capabilities their broker must gain to support this new class of consumer.

Target Audience:

  • Messaging engineers, architects, and technical leaders who operate message brokers in production and are starting to see agent-driven workloads arrive on their infrastructure. No prior AI or agent framework experience required.

Tags:
AI, Operating, optimizing and scaling