Mechanize

Proximal, founded in fall 2025 in San Francisco, describes itself as a data research lab that builds complex, long-horizon reinforcement learning environments and evaluation data for frontier AI labs and enterprises. It started with coding-agent training data and is expanding into drug design, chip design and legacy software modernization; it also released the open benchmark FrontierSWE.

Fleet (fleetai.com), founded in 2024 and based in New York, builds high-fidelity reinforcement-learning environments that replicate enterprise software such as Salesforce and Excel plus browser/desktop workflows, so frontier AI labs and enterprises can train and evaluate computer-use agents. Founder and CEO Nicolai Ouporov was a founding engineer at Respell (acquired by Salesforce).

Bespoke Labs was founded in 2024 by Mahesh Sathiamoorthy and Alex Dimakis in Mountain View, California. It builds reinforcement learning environments and infrastructure for training, evaluating, and improving long-horizon AI agents for frontier labs and enterprises, using simulated real-world company environments (codebases, emails, Slack logs, etc.) as RL training grounds.

Sepal AI (YC S24, San Francisco, founded 2024) builds training data, evaluation benchmarks and RL environments for frontier AI labs. It was acquired by Mercor in February 2026 for an undisclosed amount.

Founded in 2018 and headquartered in San Francisco, Labelbox is an AI data and agent infrastructure company offering Horizon (RL environments for frontier labs), Recursion managed agents for enterprises, and the Alignerr expert network for labeling, evaluation and human feedback.

Micro1, founded in 2022 in Palo Alto, California, helps AI companies source and manage human contractors for data labeling and model training.