Staff Software Engineer, Kubernetes Platform

Anthropic•San Francisco, NY

2d•Hybrid

About The Position

Anthropic runs some of the largest Kubernetes clusters in the industry, with fleets of hundreds of thousands of nodes across multiple cloud providers and datacenters to train, research, and serve frontier AI models. The Kubernetes Platform team owns the Kubernetes control plane that makes those clusters work. We are operating at a scale where the defaults stop working. We own the scheduler and extend it to place topology-sensitive ML workloads across thousands of accelerators at once. We scale the control plane itself — apiserver, etcd, controllers — so it stays responsive as object counts and node counts grow by orders of magnitude. And we build the core cluster services every workload depends on, like service discovery, so they hold up under the same pressure. We make sure the control plane is fast, correct, and always available. Your work will directly determine whether Anthropic can keep reliably and safely training frontier models as our compute footprint continues to grow.

Requirements

Significant software engineering experience building and operating production distributed systems
Proficiency in at least one systems-appropriate language (e.g., Go, Python, Rust, or C++)
Deep, hands-on Kubernetes experience (well beyond "user of”) into scheduler, controllers, apiserver, or operating large multi-tenant clusters
Demonstrated ability to debug complex issues across the stack, from API behavior down to node and network-level root causes
A track record of designing for reliability, correctness, and clear failure semantics in systems other engineers depend on
Strong written and verbal communication; comfort building consensus with internal stakeholders

Nice To Haves

Experience with Kubernetes internals or contributions: kube-scheduler / scheduling framework, apiserver, etcd, client-go, controller-runtime, or similar
Experience building or operating cluster schedulers or batch systems (e.g., Kueue, Volcano, Slurm, or in-house equivalents)
Background scaling control planes or coordination systems (etcd, ZooKeeper, Consul, or large DNS/service-mesh deployments)
Familiarity with ML infrastructure: GPUs, TPUs, or Trainium; gang scheduling; topology-aware placement; collective networking such as NCCL
Experience with GCP and/or AWS, including GKE/EKS internals and Infrastructure as Code
Low-level systems experience such as Linux kernel tuning, cgroups, or eBPF
8+ years of relevant industry experience, including time leading large, ambiguous infrastructure projects

Responsibilities

Own, operate, and extend the Kubernetes scheduler for Anthropic's accelerator fleets, including custom scheduling plugins and policies for gang scheduling, topology awareness, and preemption
Scale the Kubernetes control plane (apiserver, etcd, controller-manager) to support clusters far beyond typical limits, and find the next bottleneck before it finds us
Design, build, and operate core cluster services such as service discovery that every workload in the fleet depends on
Build and maintain custom controllers, operators, and CRDs
Partner with research, training, and inference to understand workload shapes and turn their requirements into platform capabilities
Collaborate with cloud providers on required features and escalations
Participate in on-call, lead incident response, and design processes (postmortems, runbooks, SLOs) that help the team avoid repeating failures