Mechanize, the reinforcement learning environment startup barely a year old, has reached a $750 million valuation, according to The Information. The company builds what the industry calls “RL gyms” – simulated environments where frontier AI models learn to code, debug, and deploy software by doing it over and over, getting scored on performance each time.
The valuation reflects something bigger than one startup’s fundraise. AI labs are hungry for high-quality training data, and RL environments have become the critical infrastructure for building the next generation of AI agents.
Why RL Gyms Matter Now
Training large language models on static text has hit diminishing returns. The next leap, AI that can actually do things, not just talk about them, requires reinforcement learning. Models need environments where they can attempt tasks, fail, get feedback, and improve.
That’s what Mechanize builds. Their environments let AI models carry out real software engineering work:
- Building features from scratch
- Deploying applications
- Debugging unfamiliar codebases
Automated graders score each attempt, providing the reward signals that drive learning. It’s the difference between reading about swimming and actually getting in the pool.
The Investor Roster
Mechanize has attracted a who’s-who of tech investors. Nat Friedman and Daniel Gross (former GitHub CEO and AI investor, respectively), Stripe co-founder Patrick Collison, podcast host Dwarkesh Patel, and Google Chief Scientist Jeff Dean have all backed the company. The startup reportedly offers engineers salaries north of $500,000 to build these environments.
The company has already been working with Anthropic on RL environments, and other frontier labs are likely customers given the industry-wide push toward agent capabilities.
A Market Taking Shape Fast
Mechanize isn’t alone. The RL environment space has exploded over the past year:
- Mercor pivoted into RL infrastructure and was eyeing a $10 billion valuation based on $450 million in annual revenue
- Prime Intellect, backed by Andrej Karpathy, is building an open hub for RL environments
- AgileRL raised $7.5 million to speed up reinforcement learning for enterprises
Venture firm Wing estimates roughly 20 seed- to Series A-stage companies are competing in this space right now. By 2030, that field will narrow to three to five market leaders.
What stands out here is the speed. Mechanize was founded in early 2025 with an audacious stated goal of “automating all work.” In practice, co-founder Tamay Besiroglu started with RL environments for coding agents, a narrower but enormously valuable wedge. Reaching $750 million in roughly a year signals how urgently labs need what these companies are building.
What This Means
The AI industry’s bottleneck is shifting. It’s no longer just about compute or model architecture. It’s about training signal, giving models structured, scoreable environments where they can learn to act, not just predict. Companies that control the best RL environments will hold serious leverage over the labs building frontier models.
For practitioners, this is worth watching. The quality of RL environments directly shapes how capable the next wave of AI agents will be. And at $750 million for a company this young, investors are clearly betting that this infrastructure layer is foundational.
More details are available in the full report from The Information.