Software Engineer, Compute Infra

Description

The company is working to build artificial intelligence that helps people understand the universe, covering the full stack from compute clusters to models and consumer apps.

The team builds the sandbox infrastructure that gives each agent its own virtual machine. These VMs run coding agents, Grok Bot, reinforcement learning workloads and more. The fleet spans millions of CPU cores across third-party and first-party datacenters.

You will own the cloud platform for agents, including the virtualization layer (Firecracker and Cloud Hypervisor microVMs), control plane scaling and the networking and connectivity layer above it.

Responsibilities

• Re-architect the control plane and scheduler to handle 1 million concurrent VMs per cluster.
• Build a ublk block device that deduplicates and lazy-loads VM data and takes efficient snapshots.
• Design an egress proxy and network isolation that securely filter, monitor and route agent traffic.
• Design resilient systems that run across a mix of cloud and on-premise datacenters.

Requirements

• Bachelor's degree in computer science, engineering, math or a scientific field, or at least 2 years of professional software experience in place of a degree.
• Preferred: deep experience building compute orchestration systems such as Kubernetes or Mesos.
• Preferred: experience building and operating large fleets of VMs or isolated sandboxes.
• Preferred: experience with the Linux kernel and virtualization technologies.
• Preferred: interest in designing the infrastructure primitives future agents will need.