Kevin Weil's Warp Sciences Raises $215 Million
OpenAI's former science chief raised $215 million for a US chemistry lab that trains robots, with OpenAI's CFO among his angel investors.
Read the storyFounder, Arda; former chief research officer, OpenAI
Bob McGrew is the founder of Arda, which develops AI software for manufacturing, and a former chief research officer at OpenAI. His research record includes reinforcement learning, dexterous robotic manipulation and contributions to large language models. Arda emerged from stealth in October 2026. McGrew also received a commission in the US Army Reserve’s Executive Innovation Corps in June 2025.
Get AI Secret’s free daily briefing, including Bob McGrew and the wider AI industry.

Bob McGrew is the founder of Arda, which develops AI software for manufacturing, and a former chief research officer at OpenAI. His research record includes reinforcement learning, dexterous robotic manipulation and contributions to large language models. Arda emerged from stealth in October 2026. McGrew also received a commission in the US Army Reserve’s Executive Innovation Corps in June 2025.
Sources checked . Selected career background and milestones.
Selected research, leadership and career milestones, newest first.
Accel identifies McGrew as Arda’s founder and announces its emergence from stealth. The company develops AI software intended to connect engineering design changes with factory-floor execution.
Source: Accel ↗The Army commissions McGrew as a lieutenant colonel in its Reserve through Detachment 201, alongside technology executives recruited for their expertise in advanced technologies.
Source: US Army / DVIDS ↗OpenAI names McGrew in the executive-leadership acknowledgments for o1. The credit distinguishes research-program leadership from the specific technical contributor and research-lead lists.
Source: OpenAI ↗OpenAI lists McGrew under additional leadership and resource allocation for GPT-4o, documenting his role in the broader research and launch effort behind the multimodal model.
Source: OpenAI ↗OpenAI demonstrates a robot hand solving a Rubik’s Cube using policies trained in simulation. McGrew is among the credited authors of the research project.
Source: OpenAI ↗The paper studies reinforcement learning for a dexterous robotic hand, using randomized simulation to train policies that transfer to physical hardware, with McGrew among the authors.
Source: Research paper ↗McGrew co-authors a reinforcement-learning method for sparse-reward tasks. The research learns from unsuccessful goal-directed experiences and demonstrates manipulation policies trained in simulation on a physical robot.
Source: OpenAI ↗Briefs, articles and relevant sections, newest first. Explore older coverage below.
More coverage
OpenAI's former science chief raised $215 million for a US chemistry lab that trains robots, with OpenAI's CFO among his angel investors.
Read the storyBob McGrew's Arda raised $232 million for an assembly line agent that cut Anduril's instruction writing from 10 hours to 15 minutes.