Marco Jahn

Most multi-agent systems start with direct API calls between agents—then debugging becomes impossible, one slow agent blocks the chain, and replaying a failed workflow means re-running everything.

Modern event-driven architectures still separate cleanly into "queues" and "streams," but real workloads no longer do. Requirements like replay, ordering, high throughput and more, increasingly appear across both patterns, making the wrong service choice costly to unwind.

Building resilient serverless applications requires more than relying on platform features alone. As systems scale, we must implement proven architectural patterns for distributed systems.

While serverless architectures can be optimized for cost, throughput and latency, it is not possible to optimize for all factors simultaneously. Organizations must prioritize based on workload requirements.

In this chalk talk, learn considerations for building resilient serverless applications by understanding business requirements. Starting with a simple serverless architecture, explore how nonfunctional requirements drive continuous enhancement and design of architectural patterns and improvements.

Observability is more than monitoring your workload—you also need logs and traces for full visibility into your applications. Many serverless workloads consist of components distributed over multiple accounts and teams.