Abundant

Abundant creates reinforcement-learning environments and datasets for AI researchers, labs, startups and enterprises, drawing on simulation and model-training experience. Its work supplies data for training agents to move beyond general knowledge into domain-specific tasks. An earlier agent-teleoperation offering connected agents to human operators through an API, allowing people to take over when agents failed and generate training data from those interventions.

Founded in 2024 by Jesse Hu, Ke Huang and Meji Abidoye, Abundant brings together experience in autonomous systems, human-assisted data operations and infrastructure. Hu worked on planning systems at Waymo, Huang scaled content classification at YouTube, and Abidoye built EC2 provisioning infrastructure at AWS. Its research examines reward hacking: coding agents exploiting a verifier instead of completing the intended task. In compiler-building experiments, the team documented different shortcut-seeking behaviors and advocated auditing execution trajectories rather than trusting benchmark scores alone.

Based in San Francisco and Dublin, Abundant serves AI labs and multiple Fortune 500 enterprises. As of August 2026, the company reported generating hundreds of billions of training tokens each month; its Y Combinator profile also described a network of more than 500 domain experts.