Software Engineer, Compute and Isolation

Baseten; United States; Remote; USD 120,000 - 180,000; Full time

Job summary

Baseten is seeking an experienced Platform Engineer to own the microVM lifecycle for agent sandboxes, including start, stop, suspend, resume, snapshot, and fork. You will optimize the guest kernels and ensure isolation across multi-tenant workloads.

You'll tune host CPU/memory configurations, manage resources, and drive end-to-end startup performance from boot to first command. Contribute to open-source decisions and work across hardware, storage, and networking to keep Baseten Sandboxes fast

Responsibilities

• Own the microVM lifecycle for agent sandboxes: start, stop, suspend, resume, snapshot, and fork
• Build and tune the custom guest kernels sandboxes run on
• Tune host CPU and memory configuration so sandboxes run fast and safely on each type of hardware we operate
• Manage host-level resources and sandbox density on each machine
• Own startup and resume performance end to end, from VM boot or snapshot restore to the first command
• Own the guest and host side of each sandbox: its identity, how it attaches to storage, and the virtual devices that connect it to the network
• Design isolation for multi-tenant workloads where the hypervisor is the security boundary
• Debug virtualization and isolation issues at scale, from userspace through the kernel to the hardware
• Contribute architecture decisions and code back to the open-source projects we build on

Job description

ABOUT BASETEN

Baseten powers mission-critical inference for the world's most dynamic AI companies, like Cursor, Notion, OpenEvidence, Abridge, Clay, Gamma, and Writer. By uniting applied AI research, flexible infrastructure, and seamless developer tooling, we enable companies operating at the frontier of AI to bring cutting-edge models into production. We're growing quickly and recently raised our $1.5B Series F, led by Altimeter Capital, Conviction Partners, and Spark Capital. Join us and help build the platform engineers turn to ship AI products.

THE ROLE

Every agent on Baseten Sandboxes gets its own microVM, where it writes and runs code, calls tools, and keeps state across long tasks. Agents pause and resume constantly, so the machine underneath has to start, stop, snapshot, fork, and resume fast, and keep every tenant isolated while it does. General-purpose virtualization was not built for that workload. The Agent Infra team builds that layer. Sandboxes run in microVMs with custom Linux kernels on bare-metal servers we operate. You'll own how sandboxes start, stop, snapshot, fork, and resume, and how they perform on each host's CPU and memory, down to the hardware. You'll also own the guest and host side of each sandbox: its identity, how it attaches to different storage classes for performance or capabilities, and the virtual devices that connect it securely to the network. In September 2026, Baseten acquired Blaxel, whose microVM sandboxes are the foundation for Baseten Sandboxes. We are building a cloud, not just using one. That means designing for many tenants from day one and thinking through what each decision means all the way down to the datacenter. The team is small and ownership is big: we ship in weeks and months, not multi-year programs. If you've worked deep in KVM and Firecracker, Cloud Hypervisor, or QEMU and want to build the computers AI agents run on, we'd love to talk.

EXAMPLE INITIATIVES

• Blaxel is joining Baseten to build the future of agentic infrastructure
• Securing the open frontier with NVIDIA OpenShell and Blaxel sandboxes
• Baseten Sandboxes documentation

RESPONSIBILITIES

• Own the microVM lifecycle for agent sandboxes: start, stop, suspend, resume, snapshot, and fork
• Build and tune the custom guest kernels sandboxes run on
• Tune host CPU and memory configuration so sandboxes run fast and safely on each type of hardware we operate
• Manage host-level resources and sandbox density on each machine
• Own startup and resume performance end to end, from VM boot or snapshot restore to the first command
• Own the guest and host side of each sandbox: its identity, how it attaches to storage, and the virtual devices that connect it to the network
• Design isolation for multi-tenant workloads where the hypervisor is the security boundary
• Debug virtualization and isolation issues at scale, from userspace through the kernel to the hardware
• Contribute architecture decisions and code back to the open-source projects we build on

REQUIREMENTS

Hands-on experience with