AgileRL Lands £6M Seed to Make Reinforcement Learning Accessible for Every Company

London-based AgileRL has closed a £6 million ($7.5 million) seed round led by Fusion Fund, with participation from Flying Fish, Octopus Ventures, Entrepreneur First and Counterview Capital, bringing total funding to date to £6 million. The company is building tooling that makes reinforcement learning — one of the most powerful but operationally complex AI training techniques — practical for organisations without large in-house research teams. AgileRL will use the capital to open a San Francisco office and hire across engineering and go-to-market functions.

Reinforcement learning, in which AI agents learn by interacting with an environment and improving based on feedback rather than labelled datasets, is experiencing a significant resurgence after years in the shadow of supervised learning. It underpins some of the most capable AI systems ever built, but deploying it in an enterprise context has historically required building what amounts to a small AI research lab from scratch: teams of specialist researchers, months of exploratory runs, substantial compute budgets, and custom infrastructure for simulation, reward design, hyperparameter optimisation, distributed training and deployment. Every new use case tends to break the previous setup, forcing teams to start again. That cost and complexity has put RL out of reach for all but the largest technology companies.

AgileRL's platform addresses this with a two-tier product offering. Its free open-source RL framework provides state-of-the-art algorithms at scale and has been downloaded more than 300,000 times. Its managed RLOps platform, Arena, handles the full engineering stack end-to-end, 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 settings manually, the system trains a population of agents simultaneously, identifies the strongest performers, evolves their configuration, and discards underperforming variants — all automatically. The result is a claimed 10x reduction in training time and compute cost compared to standard approaches.

The platform is already in use at research labs and commercial organisations including MIT, Carnegie Mellon, Roblox, IBM, Airbus, JPMorgan and University of Waterloo, spanning applications in defence, robotics, finance and simulation. AgileRL was founded in 2023 by Param Kumar (CEO) and Nicholas Ustaran-Anderegg (CTO). Kumar drew on direct experience building a reinforcement learning system from scratch at his previous company as the founding insight for the business, having seen firsthand how the process consumed engineering capacity disproportionate to the eventual output.

Fusion Fund founding partner Lu Zhang framed the investment as a bet on RL becoming a standard component of the enterprise AI stack, noting that as companies move beyond simple language model applications toward complex autonomous systems, the need for accessible, production-grade RL infrastructure becomes acute. The participation of Octopus Ventures — one of Europe's most active tech investors — alongside Entrepreneur First, which backed the company from its earliest days, signals broad conviction in both the technical thesis and the founding team.

Sources

• Soapbox VC – AgileRL lands a £6m seed round
• HPCwire / AIwire – AgileRL emerges with $7.5M to help companies speed up RL by 10x
• The AI Journal – AgileRL raises $7.5M to take RL into the enterprise