CNCF: infrastructure engineer migrated 60+ Kubernetes resources in 30 minutes with the help of an AI agent

The CNCF blog published a case study on April 23, 2026, showing how an AI agent can significantly accelerate infrastructure migration. Tianyi Zhang from Alibaba Cloud described a process in which 60+ Kubernetes resources were migrated from Ingress NGINX to Higress in 30 minutes, including complete validation.

Context: why migrate from Ingress NGINX?

Ingress NGINX has for years been the de-facto standard for managing inbound traffic in Kubernetes clusters. It is a CNCF project that sits in front of services as a reverse proxy, handling routing, TLS termination, and basic load balancing.

However, as more organizations serve LLM models through their infrastructure, traditional ingress controllers are showing limitations. Traffic to LLM services has specific characteristics — variable response length, streaming, high cost per token — that classical rate limiting does not cover well.

What does Higress bring and why is it AI-native?

Higress is a gateway based on the Envoy proxy and Istio components, developed within Alibaba Cloud and contributed to the CNCF community. The key difference from Ingress NGINX is its built-in AI functionality.

Higress natively supports token-based rate limiting — throttling by number of LLM tokens rather than by number of requests. It also has semantic caching, where responses to semantically similar queries are served from cache, saving calls to an expensive LLM. For infrastructure that serves AI applications, these are significant optimizations for cost and latency.

What does 30 minutes for 60 resources mean?

In traditional work, migrating 60+ Kubernetes ingress resources would require careful annotation mapping between the two systems, manual testing of each route, and verification of TLS certificates. A single engineer would spend a day or two on that.

According to Zhang’s report, the AI agent handled mapping and generating new manifests, ran validation through a dry-run, and confirmed functionality. This is a signal that AI agents are moving from the experimental phase to production readiness for infrastructure teams, with direct implications for MTTR and operational costs in DevOps organizations.