Kubernetes Podcast from Google

Brandon Royal, a Google Cloud product manager behind Agent Substrate, joins Tim Hockin, a Kubernetes founder and longtime infrastructure engineer. They explore why AI agents need new primitives, from reusable workers and multiplexed actors to checkpointing, resumable sessions, networking, identity, and policy. They also discuss Agent Sandbox, higher-order scheduling, and the roadmap toward production-scale agent workloads.

Dipesh Rawat, a software developer at IBM and long-time Kubernetes contributor, served as Release Lead for v1.37. He recounts his path from enhancement teams to release leadership. Topics include the Gharwal release theme, 67 tracked enhancements, notable deprecations and removals, scheduling and AI/GPU improvements, KAML stabilization, and upgrade and cgroups v2 guidance.

Niki Manoledaki, Staff Platform Engineer at Grafana Labs and core maintainer of Project Kepler, talks GreenOps and measuring energy and carbon for Kubernetes. She describes Project Kepler’s rewrite, challenges of measuring on VMs versus bare metal, and how AI changes sustainability metrics. Short, practical dives into observability, platform tools, and why accuracy versus estimates matters.

Natali Vlatko, SIG Docs co-chair and Open Source Architect at Cisco, and Kat Cosgrove, SIG Docs technical lead and Kubernetes Steering Committee member, explain the new Kubernetes AI usage policy. They discuss why AI cannot sign CLAs or be listed as co-authors. Conversation covers handling low-quality AI PRs and comments, meeting note-taker bots, and paths for newcomers to contribute to docs and releases.

A deep dive into running one-pod-per-user sandboxes at scale on Kubernetes. They cover pod churn, Cilium dataplane quirks, and strategies for warm pools versus cold starts. Container runtimes and gVisor tradeoffs get attention, along with snapshotting, dynamic registries, and persistence patterns for user state.

Ryota Sawada, a software engineer at Numtide and release lead for Kubernetes 1.36 with a finance and cloud-native background. He talks about the Haru theme and logo creation, the release’s large scale and GA of user namespaces. Conversation covers new Workload API and pod group ideas, workload-aware scheduling for AI/ML, and progress on Dynamic Resource Allocation.

Lucy Sweet, Staff Software Engineer at Uber who led the move of massive compute fleets to Kubernetes. She talks about why Uber migrated from Peloton, making services portable, traffic strategies and canaries. Blue-green node pools, DNS as a critical dependency, stateful migration challenges, and tools like Grail and kubegpt also come up.

Amie Wei is a Sr. Solutions Engineer at HashiCorp and was the winner of last year's GKE Turns 10 Hackathon. It was Amie's first time entering a hackathon and she ended up bringing the prize home with a Cart-To-Kitchen AI Assistant. Do you have something cool to share? Some questions? Let us know: - web: kubernetespodcast.com - mail: [email protected] - twitter: @kubernetespod - bluesky: @kubernetespodcast.com News of the week Kubernetes 1.36 codename Haru is here LLM-d is a CNCF Sandbox project CNCF Certification Advancement & Recertification Experience program (CARE) Agones is a CNCF Sandbox project Links from the interview Amie Wei on LinkedIn GKE turns 10 Hackathon Hackathon winner announcement Online boutique sample Bank of Anthos sample The Cart-to-Kitchen AI Assistant on GKE

Drew Hagen, a Senior DevOps engineer at Medtronic and release lead for Kubernetes 1.35, explores the Timbernetes theme symbolizing resilience. He shares insights on leading the release, the significance of coordination, and critical enhancements like in-place pod resource updates. Drew discusses scheduling improvements beneficial for AI and edge computing, security updates, and the retirement of NGINX Ingress. He also addresses important deprecations and upgrade warnings, offering a glimpse into the future of Kubernetes and its evolving applications.

Janet Kuo, a Staff Software Engineer at Google Cloud and key figure behind the Kubernetes AI Conformance program, dives into the new AI conformance initiative. She highlights how AI workloads differ from standard Kubernetes conformance, emphasizing their unique demands for stateful operations and hardware. Janet explains the introduction of the Dynamic Resource Allocation (DRA) API, designed to improve accelerator management for complex AI needs. She also shares insights on future standardizations and the importance of consistent platform capabilities.