Staff Software Engineer (Environments Infrastructure)

About this role

• Anthropic’s Environments organization builds and maintains the infrastructure that improves Claude’s capabilities through reinforcement learning
• That includes the frameworks researchers use to build environments and the infrastructure responsible for running them
• The team’s mission is to productionize research
• You’ll embed with research teams, get up to speed on how they work, and design the frameworks and APIs that let them move faster, building systems the team can understand, own, and maintain themselves
• Scope also includes keeping production RL runs healthy, maintainable, monitored, and easy to triage
• You’ll be a strong fit if you have deep expertise in Python, a refined sense of taste for API and framework design, and good intuition for how complex systems fail, especially silently
• It’s a bonus if you’ve built and operated a stateful distributed system, such as a workflow engine, actor framework, or durable-execution runtime, where correctness depends on getting shared state and recovery right
• You should be comfortable diving into messy research code, finding the abstractions that matter, and improving them incrementally while researchers continue to build on your work
• You should also be comfortable using AI tools to accelerate your own development, but have an impulse towards deep verification
• Design widely used APIs, frameworks, and abstractions that other engineers and researchers build on, making correct usage the default and ruling out entire classes of errors structurally
• Own the platform layers that sit beneath every environment, including the agent runtime
• Build the tooling that lets environment owners understand, debug, and maintain their environments in production without needing an infrastructure engineer in the loop
• Embed with research teams on a rotational basis, work directly in their codebases without slowing down the research they support, and transfer ownership when you rotate off
• Anticipate silent failure modes and ...