AI startup Deeptune has raised a $43 million Series A to build what it calls “training gyms” for AI agents, Fortune has learned exclusively. Andreessen Horowitz led the round, joined by 776, Abstract Ventures, and Inspired Capital, with angels including OpenAI researcher Noam Brown, Mercor CEO Brendan Foody, and Applied Compute CEO Yash Patil.
Deeptune creates high‑fidelity reinforcement learning (RL) environments that simulate the day‑to‑day workflows of roles like accountants, customer support reps, and DevOps engineers, so AI agents can learn to navigate multi‑step tasks across popular workplace software such as Slack, Salesforce, and other 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 told Fortune.
Lupo likens today’s 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 said. “What we build are essentially the flight simulators for AI doing work across the economy.”
Deeptune’s bet reflects a broader shift in AI from training on static web‑scale data to running large‑scale reinforcement learning in synthetic and interactive environments—a direction seen in recent agentic RL work on tool‑use agents at Microsoft, and OpenAI’s computer‑using agent. The global reinforcement learning market, including tools and environments, is projected to grow from roughly $11.6 billion in 2025 to more than $90 billion by 2034, according to ResearchAndMarkets.
“Instead of depending primarily on human-annotated data, models are learning through interaction, running rollouts, taking actions, and receiving rewards in dynamic environments that function like a playground,” Marco Mascorro, partner at Andreessen Horowitz, told Fortune. “Deeptune has built a platform that enables this shift, allowing top labs to train and evaluate these behaviors in a reliable and scalable way. Tim and the team have a deep understanding and experience working with frontier AI research labs on these problems.”
The company says it has built hundreds of these training gyms for leading AI labs, and that its environments have already contributed to recent advances in agents’ ‘computer use’ capabilities—moving beyond simple question answering to multi‑step workflows on real 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 told Fortune. “We now know that they work insanely well.” According to him, anything that can be distilled into an environment—“from editing a video to building an LBO in Excel”—is something AI can learn.
That need is turning RL environments into a hot new infrastructure category, with major labs reportedly considering spending more than a billion dollars on such environments and data-labeling incumbents racing to build out their own offerings. The global reinforcement learning market, including tools and environments, is projected to grow from roughly $11.6 billion in 2025 to more than $90 billion by 2034, according to ResearchAndMarkets.
As investors, including Marc Andreessen, warn that AI companies are ‘running out’ of high‑quality human data and studies project that public web data for training could be exhausted within the next decade. Deeptune pitches its simulated workspaces as a way to create rich, task‑specific experience for models—by having them practice inside realistic enterprise environments rather than simply scraping more of 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 said.
The roughly 20‑person, in‑person team is based in New York and includes engineers and operators from Anthropic, Scale AI, Palantir, Hebbia, Glean, and Retool, according to the company. Lupo frames New York as a deliberate choice and a recruiting edge: “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.”
