Staff Research Engineer, Multi-Agent Scaling

Anthropic is building AI systems designed to be reliable, interpretable, and aligned with human values. The organization brings together researchers, engineers, policy specialists, and business leaders focused on ensuring AI benefits society. As the field advances rapidly toward more capable systems, Anthropic is tackling fundamental questions about safety and control.

Multi-agent teams are beginning to solve complex problems that individual agents cannot handle alone—from refactoring entire codebases to proving mathematical theorems. Understanding how these teams scale is critical research territory. You'll join Anthropic's AI Research and Engineering team to investigate the dynamics of agent coordination: how performance, cost, and efficiency evolve as team size, compute budgets, and task complexity increase. Your work will establish the experimental frameworks and infrastructure that measure and enable large-scale agent collaboration across the organization.

What you'll do

• Design and execute large-scale experiments testing agent team performance, applying rigorous quantitative reasoning to interpret results and distinguish signal from noise

• Investigate how efficiency and capability scale with team size, computational resources, and task horizon, identifying the constraints that limit further progress

• Build robust infrastructure to run very large agent teams reliably, diagnosing and fixing failure modes that emerge only at scale

• Create evaluations for long-horizon problems that remain trustworthy and interpretable under increasing complexity

• Develop monitoring tools, dashboards, and metrics that help researchers understand what large agent teams are doing and why

• Collaborate with research teams throughout Anthropic, enabling them to run experiments on your platform and helping them translate findings into actionable insights

What you'll bring

• Substantial professional experience in software engineering, machine learning, or research engineering

• A track record of owning something meaningful end-to-end: a major system, evaluation framework, benchmark, agent application, or research initiative

• Genuine enthusiasm for both research and engineering, and comfort moving fluidly between exploration and implementation

• Quantitative thinking about intricate systems, with healthy skepticism toward numbers until they're thoroughly validated

• Ability to move from loosely defined questions toward concrete experiments without detailed specifications

• Focus on measurable results, pragmatism, and adaptability in pursuit of impact

• Strong written and verbal communication skills

• Genuine concern for how your work affects society

Nice to have

• Experience designing or operating large-scale distributed systems such as schedulers, sandboxed execution environments, or inference platforms

• Background building evaluations and benchmarks for language models or agents

• Knowledge of scaling laws and empirical research at scale

• Training in operations research, statistics, economics, physics, quantitative finance, or similar fields that model complex systems

What they offer

• Annual compensation between 500,000 and 850,000 USD

• Visa sponsorship available for qualified candidates

• Hybrid arrangement with expectation of in-office presence at least 25% of the time in San Francisco, New York, or Seattle

• Bachelor's degree or equivalent professional experience required