When: Friday, April 3, 3:00 PM
Where: Tyler 055

Abstract
Every generation of developer infrastructure faces the same core tension: how do you give increasingly powerful systems the ability to act autonomously while maintaining the safety and governance guarantees that organizations require? For decades, this played out in distributed systems — microservices, data pipelines, and platform engineering at organizational scale, where the hard problems were as much about people and process as they were about technology. Now that tension is taking on a new shape: the systems taking action aren’t just human developers and scheduled jobs, but AI agents that reason, write code, invoke tools, and accumulate knowledge over time.

This talk explores what happens when you apply distributed systems thinking to the emerging agent infrastructure space. What changes when your “distributed actors” are non-deterministic AI models? How do you isolate execution for a system that authors its own code? How do you enforce authorization policies on something that can’t be trusted to follow instructions? And what does it take to move agents from impressive demos to production systems that enterprises actually deploy?

Drawing on experience building a global development platform in asset management, leading product for the commercial ecosystem around Apache Airflow, co-founding an agent platform built on WebAssembly runtimes and knowledge graphs, and now working on Amazon Bedrock AgentCore at AWS, the talk will ground these questions in concrete architectural decisions — from tool-call interception and sandboxed environments to memory systems that extract knowledge rather than just store conversations. There will be plenty of room for discussion, especially around the accountability challenges that come with deploying autonomous systems in the real world.

Bio
Ryan Fox-Tyler leads product for Amazon Bedrock AgentCore at AWS — a platform for building agents that are more knowledgeable, take action with safe tools, and continuously optimize. Previously, he co-founded Hypermode, an agent development platform built on WebAssembly runtimes and knowledge graphs, and led product for Astronomer, the company behind Apache Airflow. Before entering the startup and cloud ecosystem, Ryan spent a decade at a trillion-dollar global asset manager, where he built and ran the firm’s first global development platform. His work spans developer experience, distributed systems, and AI infrastructure.