Member of Technical Staff - RL Environments

Role overview

Cohere is building foundation AI models and enterprise products designed to solve real-world business challenges. This role focuses on creating reinforcement learning environments that realistically simulate enterprise work, enabling the company to train and evaluate AI agents. You will design these environments, run agents within them, and continuously improve both the agents and environments based on performance data and customer feedback.

Key responsibilities

Design and build new reinforcement learning environments that target different agentic capabilities across various industry areas.
Train and evaluate AI agents within these environments to measure their performance and identify areas for improvement.
Integrate all components of the evaluation pipeline including tasks, data, tool implementations, and reward verifiers.
Collaborate with modeling and product teams to identify gaps in agent performance and implement improvements to both agents and environments.
Partner with external vendors to develop high-quality, expert-built reinforcement learning environments and create tools to maintain standards for tasks, data, and verifiers.
Automate the process of discovering model capability gaps and establish systematic approaches to measure agent performance during evaluation and training phases.

Requirements

Experience engineering agents and optimizing them for specific industry applications.
Demonstrated ability to analyze agent trajectories in depth, identify failure points, and address them through model training or engineering adjustments.
Strong focus on measuring agentic capabilities and developing repeatable, systematic measurement processes.
Experience defining what constitutes success for AI agent outcomes, translating this into verifier implementations, and tuning reward design accordingly.
Background designing and running annotation workflows that surface insights into agent performance and verify data quality.
Experience building synthetic data pipelines to support evaluation and training at scale.
Preferred experience with reinforcement learning training, including scaling, troubleshooting, and environment tuning.

What makes this role worth considering

Cohere operates as a remote-friendly organization with offices across Toronto, London, New York City, San Francisco, Montreal, Paris, Berlin, and Seoul. This London-based position offers flexibility to work remotely or from an office location with no minimum in-office requirement. The company provides comprehensive benefits including six weeks of paid vacation, full health and dental coverage, parental leave top-up, a weekly lunch stipend, education and learning support, and a home office setup budget.