AgileRL

AgileRL is a London-based machine learning company founded in 2023 by Param Kumar (CEO) and Nicholas Ustaran-Anderegg (CTO), with backing from Entrepreneur First. The company was founded to solve a structural problem in enterprise AI: reinforcement learning — the training technique behind some of the most capable AI systems ever built — requires so much specialist infrastructure that only the largest technology companies can deploy it. Building an RL programme from scratch means assembling teams of expensive researchers, months of exploratory compute runs, and custom pipelines for simulation, reward design, hyperparameter optimisation, distributed training, and deployment. Every new use case tends to break the previous setup. AgileRL was built to change that.

The company operates a two-tier platform. Its free open-source RL framework has been downloaded over 300,000 times and provides state-of-the-art algorithms at scale. Its managed RLOps platform, Arena, handles the complete engineering stack from environment validation and evolutionary hyperparameter optimisation to distributed multi-GPU training and one-click deployment. Arena's key differentiator is evolutionary hyperparameter optimisation: rather than training a single agent and tuning manually, the system trains a population of agents simultaneously, identifies the strongest performers, evolves their configuration, and discards underperforming variants automatically — delivering a claimed 10x reduction in training time and compute cost. Customers include MIT, Carnegie Mellon, Roblox, IBM, Airbus, and JPMorgan. In January 2026, AgileRL raised a £6 million seed round led by Fusion Fund, with participation from Flying Fish, Octopus Ventures, Entrepreneur First, and Counterview Capital, and is opening a San Francisco office.