We're looking for a distributed systems engineer to build the infrastructure Plasma's agents run on. You'll design the scheduling, state, isolation, and fault-tolerance layers that keep a fleet of agents working for hours at a time.

You've built and run large distributed systems, such as job schedulers, control planes, or data platforms, and want to apply that to a new kind of workload. The role is both research and product: long-running agents raise open problems in scheduling and recovery, and what you find goes into Plasma's runtime.

What You'll Work On

• Scheduling: Place and run concurrent agents across compute, within the budgets users set

• Isolation: Build the sandboxed environments agents execute in, with limits on network access, data access, and spend

• State & Consistency: Keep agent state and shared knowledge consistent at the storage layer while many agents read and write them at once

• Fault Tolerance: Checkpoint long-running work so a task survives a crashed agent, a failed model call, or a lost node

• Messaging: Scale the transport behind agent-to-agent communication

What We're Looking For

• Experience designing and operating distributed systems in production: schedulers, consensus, replication, queues, or stream processing

• Strong programming skills in Python and at least one systems language such as Go or Rust

• Experience with sandboxing and isolation, such as containers, microVMs, or network policy

• A record of debugging failures that only appear at scale or under concurrency

• Experience running LLM-based agents in production is a plus