Building the Network for AI Agents — Yan Avlasov, Google

AI agents necessitate a fundamental shift in networking infrastructure, requiring a new semantic layer to manage personalized state, bursty traffic patterns, and complex policy enforcement. Yan Avlasov, a staff software engineer at Google and senior maintainer of Envoy, explains that while traditional proxies prioritize throughput, agentic workloads demand intelligence regarding model selection, cost control, and schema transcoding. Envoy provides a robust foundation for this evolution, though it must adapt to handle high-frequency control plane updates and stateful inference. Furthermore, AI-powered security scanning has transformed vulnerability detection, enabling the identification of sophisticated, multi-bug exploits that exceed human analytical capacity. This shift creates a massive influx of security reports, forcing a transition toward continuous, automated hardening of core systems like Envoy to address path normalization and JSON parsing vulnerabilities in an increasingly agent-driven ecosystem.