Silicon Valley bets big on ‘environments’ to train AI agents

image via TechCrunch

September 21, 2025, 7:22 PM

• •RL environments are crucial for training AI agents to perform multi-step tasks in simulated workspaces.
• •Major AI labs are investing heavily in RL environments and seeking third-party vendors for high-quality solutions.
• •Startups like Mechanize and Prime Intellect are emerging to lead the RL environment space, with data-labeling companies also expanding their focus.
• •RL environments enable AI agents to operate within simulated applications with tools and computers.
• •There are challenges in scaling RL environments and preventing AI reward hacking.

AI researchers and labs are increasingly focused on reinforcement learning (RL) environments, which simulate real-world tasks for AI agents, to overcome the limitations of current AI agent technology. This has led to a surge in demand for RL environments, creating a new market for startups specializing in their development, with significant investment from major AI labs and data-labeling companies. While the potential is vast, challenges remain in scaling these environments and preventing reward hacking, raising questions about the long-term viability and impact of RL environments on AI progress. The success of RL environments could be pivotal in advancing AI capabilities and creating more general-purpose AI agents capable of performing complex tasks.

Read original article