Full Job Description
Build evaluation environments and benchmarks that measure whether AI agents can perform realistic company work and meet or exceed human performance. This founding engineer role sits at the intersection of agent evaluation and production readiness, with direct influence on how agent competence is assessed.
What You'll Do
• Design synthetic company environments with realistic email, records, and files for agents to use over extended periods.
• Build robust graders and evaluation systems that can assess agents, including when their performance may exceed human reviewers' expertise.
• Develop benchmarks that reflect real-world production performance.
• Find and address evaluation failures, including reward hacking and benchmark gaming.
• Simulate software tools such as inboxes and CRMs to support realistic agent interactions at scale.
What We're Looking For
• Experience building environments or benchmarks used by frontier AI labs to train or evaluate agents.
• Hands-on experience designing and implementing AI agent graders or evaluation systems.
• Experience simulating software systems that can withstand agent tool use, and building synthetic environments for production-like evaluation.
• A track record of identifying flaws in widely used benchmarks and detecting and fixing reward hacking in graders.
• Startup or early-stage company experience, strong competitive problem-solving experience in areas such as math, programming, debate, or games, and a collaborative approach to ambiguous, fast-paced work.
• Strong conviction in the potential of AI agents to perform computer-based work.
Compensation & Benefits
Base salary is $150,000 to $300,000 annually, with meaningful founding equity. Visa sponsorship availability requires confirmation, as the supplied details differ.
Location
On-site five days a week in San Francisco, California, United States.
