Applied Research Computer Use

Job Summary

To support the development of multimodal agents, the full-time Applied Research - Computer Use position will build post-training and evaluation systems for agents operating across various interfaces, managing workflows and experiments in a flexible work environment.

Key responsibilities

• Develop multimodal environments from partner workflows while ensuring safety and reliability
• Train agents to handle interface changes and recover from failures without restarting
• Conduct evaluations to differentiate genuine task completion from plausible behavior

Required qualifications

• Experience with multimodal agents, browser or desktop automation, and reinforcement learning
• Hands-on experience with model training, evaluation, and experimental loops
• Strong experimental judgment and ability to adapt based on evidence
• Proficiency in Python and ability to create shareable artifacts for research
• Comfort in addressing ambiguous problems and communicating progress with partners