Andreessen Horowitz has led a $43 million Series A in Deeptune, an AI startup building what it calls "training gyms" — reinforcement learning environments where AI agents can practice complex, multi-step tasks before being deployed in real workplaces. The round also included 776, Abstract Ventures, and Inspired Capital, with angels from OpenAI, Mercor, and Applied Compute.

Flight Simulators for AI

Deeptune's core idea is simple: AI agents trained only on static data are like pilots who've only ever read books. CEO Tim Lupo likens his company's environments to flight simulators — safe, high-fidelity replicas of real work that let agents learn from doing, not just reading.

The company has built hundreds of training environments simulating the daily workflows of accountants, customer support reps, DevOps engineers, and lawyers — complete with realistic versions of Slack, Salesforce, ticketing systems, and financial tools. Agents run through these simulations, take actions, and receive rewards, building competency before touching real data.

According to Deeptune, its environments have already contributed to advances in agents' "computer use" capabilities — moving AI beyond simple Q&A to navigating real software interfaces.

A Hot New Infrastructure Category

The funding reflects growing conviction that RL environments are the next major AI infrastructure layer. Major labs are reportedly considering spending over a billion dollars on such environments, and incumbents in data labeling are racing to build their own. The global reinforcement learning market is projected to grow from $11.6 billion in 2025 to over $90 billion by 2034.

a16z partner Marco Mascorro said models are shifting from human-annotated training data toward "learning through interaction" — and Deeptune is building the playground where that happens.