Staff Software Engineer, Kubernetes Platform

Anthropic's Kubernetes Platform team runs the control plane behind the company's AI compute fleets, which span several cloud providers and datacentres. This staff engineering role in London owns scheduler behaviour, control plane scaling and the cluster services that training and research workloads depend on. The post is hybrid, with at least a quarter of time expected in the office.

What you would work on - Extending the Kubernetes scheduler for accelerator fleets, including custom plugins that handle gang scheduling and topology awareness - Pushing apiserver, etcd and controller-manager well past the scale they are normally asked to handle - Designing and running core cluster services such as service discovery - Writing controllers, operators and custom resource definitions to encode platform behaviour - Working directly with research and training teams to understand what their workloads actually need - Raising and following up feature requirements with the underlying cloud providers - Taking part in on-call rotation and incident response

What they are looking for - Substantial production experience building and operating distributed systems, not just designing them - Fluency in at least one systems language such as Go, Python, Rust or C++ - Kubernetes knowledge that goes far past everyday use, ideally inside the scheduler, controllers or very large clusters - Ability to debug through the whole stack, from API behaviour down to node and network level - A record of building systems that stay correct and reliable under pressure - Clear written and spoken communication, since much of the work crosses team boundaries

Stack: Kubernetes, kube-scheduler, apiserver, etcd, controller-manager, client-go, controller-runtime, Kueue, Volcano, Slurm, GCP, AWS, GKE, EKS, NCCL, eBPF, cgroups, Linux kernel, Go, Python, Rust, C++

Compensation: The employer publishes a range of £325,000 to £485,000 GBP.