AI Agents: Keeping Systems (and Costs) in Check

Can your AI agents survive production AND your budget? In this episode, Yaron Schneider, CTO and co-founder of Diagrid and creator of the foundational open-source tools Dapr and KEDA, unpacks what it takes to build AI agents that are durable, cost-aware and enterprise-ready.

The episode explores Dapr's approach to shifting focus from infrastructure management toward developer-friendly APIs. It discusses microservices architecture, arguing against "microservice gamification" and advocating for right-sized services and event-driven designs.

A central theme addresses distinguishing genuine autonomous agents from basic function calls. Key infrastructure requirements for agent systems include:

• Durable state recovery
• Identity controls
• Orchestration capabilities
• Supported patterns like the outbox pattern and event-driven architecture

Schneider predicts a shift from cloud-native to "AI-native infrastructure" where tools and code are generated automatically based on objectives, requiring enhanced observability and reliability beyond traditional microservice models.