Exclusive: Andreessen Horowitz leads $43M Series A for Deeptune’s AI agent ‘training gyms’

(SeaPRwire) –   AI startup Deeptune has secured $43 million in a Series A funding round to develop its “training gyms” for artificial intelligence agents, has learned in an exclusive report. The investment was led by Andreessen Horowitz, with participation from 776, Abstract Ventures, and Inspired Capital, alongside angel investors such as OpenAI researcher Noam Brown, Mercor CEO Brendan Foody, and Applied Compute CEO Yash Patil.​

The company develops high-fidelity reinforcement learning (RL) environments that replicate the daily workflows of professionals like accountants, customer support representatives, and DevOps engineers. This allows AI agents to learn how to handle complex, multi-step tasks within widely used workplace applications such as Slack, Salesforce, and various ticketing, finance, and monitoring tools. “We essentially build simulations of digital work that look like the workspace of an accountant or a lawyer or a software engineer,” cofounder and CEO Tim Lupo explained to .

Lupo compares current AI models to pilots who have “only ever read books or watched tutorials.” “You wouldn’t have a pilot who has only ever read books or watched tutorials fly a plane. You would put them in a flight simulator,” he stated. “What we build are essentially the flight simulators for AI doing work across the economy.”​

Deeptune’s strategy mirrors a wider transition in artificial intelligence, moving from training on static, large-scale web data to conducting large-scale reinforcement learning within synthetic, interactive settings. This trend is visible in recent work on tool-using agents at Microsoft and OpenAI’s computer-using agent. ResearchAndMarkets forecasts that the global reinforcement learning market, encompassing tools and environments, will expand from approximately $11.6 billion in 2025 to over $90 billion by 2034. 

“Rather than relying mainly on data annotated by humans, models are now learning by interacting, executing rollouts, taking actions, and earning rewards in dynamic environments that act like a playground,” Marco Mascorro, a partner at Andreessen Horowitz, told . “Deeptune has constructed a platform that facilitates this shift, enabling leading labs to train and assess these behaviors reliably and at scale. Tim and the team possess a profound understanding and have extensive experience collaborating with frontier AI research labs on these challenges.”

Deeptune reports it has constructed hundreds of these training gyms for top AI labs, and its environments have played a role in recent progress in agents’ ‘computer use’ abilities—advancing from basic question-answering to managing multi-step processes on actual software. “We were the first company to build an environment a bit over a year ago, and no one really knew if it was going to work,” Lupo shared with . “We now know that they work exceptionally well.” He added that any task that can be represented in an environment—“from editing a video to building an LBO in Excel”—is a skill an AI can acquire.

This growing demand is establishing RL environments as a prominent new infrastructure sector. Major labs are reportedly contemplating investments exceeding a billion dollars in such environments, while established data-labeling companies are quickly developing their own competing solutions. According to ResearchAndMarkets, the global reinforcement learning market, which includes tools and environments, is expected to grow from about $11.6 billion in 2025 to more than $90 billion by 2034.

This comes as investors, including Marc Andreessen, caution that AI firms are depleting their supply of high-quality human data, and studies suggest publicly available web data for training could be used up within the next ten years. Deeptune promotes its simulated workspaces as a method for generating rich, task-specific experience for models—by having them train within realistic corporate settings instead of just gathering more data from the public internet. “I think this will become the core focus of data in general: how can we create really realistic environments that look like the enterprises that [models] might be deployed into,” Lupo remarked.​

The company’s team, which consists of roughly 20 people who work in-person, is headquartered in New York. The team includes engineers and operators with backgrounds at Anthropic, Scale AI, Palantir, Hebbia, Glean, and Retool. Lupo describes the New York location as a strategic decision and a competitive advantage in hiring: “If you want to be in New York and you want to work on frontier AI or AGI, Deeptune is one of only a couple places you could join, and probably the only early stage place you could join,” he said. “The defining problem of the next five years is, how can you make models work not just in fixed exams, but in the messy, real world…that’s what we work on here.”