AI Matey is Adam Grenier’s applied research studio, focused on teaching AI systems useful behavior through real-world signals. It combines lab-grade model training work—including reinforcement learning, training environments, reward design, and evaluations—with two decades of product experience. The studio’s approach starts from how people behave in real products: what they edit, reject, retry, purchase, save, correct, and trust. Those signals can be shaped into training tasks, rewards, and evaluations that connect model performance to real customers and workflows, rather than relying on benchmarks alone.

AI Matey builds simulated environments and agent training systems for long-horizon decisions, adaptation, memory, resource allocation, and multi-agent behavior. It also develops tools that turn human expertise and changing work practices into scalable training tasks and evaluations. Its product-grounded research addresses challenges such as humor and taste, identity preservation, factual accuracy in rewriting, culturally grounded research, and audio quality. Projects include DadBench, a benchmark and reinforcement-learning case study for dad jokes, and Slay the Sprint, a simulated product-management environment.