Andreessen Horowitz has led a $43 million Series A investment in Deeptune, a startup building "training gyms" where AI agents learn to handle real workplace tasks like accounting and customer support by practicing in simulated digital environments. This funding signals a major shift in how AI companies are approaching agent training, moving away from static web data toward interactive, task-specific learning environments. What Are AI Training Gyms and Why Do They Matter? Deeptune creates high-fidelity reinforcement learning (RL) environments, which are essentially realistic simulations of workplace software where AI agents can practice multi-step tasks. Think of it like a flight simulator for pilots, but for artificial intelligence. "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," explained Tim Lupo, Deeptune's cofounder and CEO. "What we build are essentially the flight simulators for AI doing work across the economy". These training environments simulate the day-to-day workflows of roles like accountants, customer support representatives, and DevOps engineers, allowing AI agents to learn how to navigate popular workplace software such as Slack, Salesforce, and other ticketing, finance, and monitoring tools. The company has already built hundreds of these training gyms for leading AI labs, and its environments have contributed to recent advances in agents' computer-use capabilities, moving beyond simple question answering to handling multi-step workflows on real software. How Is This Different From How AI Models Train Today? The shift Deeptune represents reflects a broader transformation in AI development. Historically, large language models (LLMs), which are AI systems trained on vast amounts of text, have relied primarily on static web-scale data for training. However, as Marc Andreessen and other investors have warned, 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's approach solves this problem by creating synthetic, interactive environments where models learn through reinforcement learning, a training method where AI systems learn by taking actions, receiving feedback, and optimizing their behavior. "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," explained Marco Mascorro, partner at Andreessen Horowitz. This represents a fundamental shift in AI infrastructure. Rather than scraping more of the public internet for training data, companies can now create rich, task-specific experiences for models by having them practice inside realistic enterprise environments. "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 told Fortune. Ways Deeptune's Technology Is Reshaping AI Agent Development - Realistic Workflow Simulation: The platform simulates actual workplace software and multi-step tasks, allowing AI agents to learn how accountants, lawyers, and software engineers work in real digital environments rather than learning from abstract text data. - Scalable Training Infrastructure: Deeptune has built hundreds of training environments for leading AI labs, creating a scalable platform that major research organizations can use to train and evaluate agent behaviors reliably and at scale. - Task-Specific Learning: The company's approach enables models to master anything that can be distilled into an environment, from editing videos to building financial models in Excel, creating specialized AI agents for enterprise workflows. - Reduced Data Dependency: By using synthetic environments instead of relying on human-annotated data or public web scraping, Deeptune addresses the growing scarcity of high-quality training data that AI labs face. Why Is This Market Exploding Now? The 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 data cited in the funding announcement. This explosive growth reflects the urgent need for better training infrastructure as AI agents move from research labs into production environments handling real work. Major AI labs are reportedly considering spending more than a billion dollars on such environments, and data-labeling incumbents are racing to build out their own offerings. The investment from Andreessen Horowitz, alongside co-investors 776, Abstract Ventures, and Inspired Capital, validates that RL environments are becoming a critical infrastructure category for the AI industry. The round also attracted notable angel investors, including Noam Brown, a researcher at OpenAI; Brendan Foody, CEO of Mercor; and Yash Patil, CEO of Applied Compute. These investors bring deep expertise in AI research and agent development, signaling confidence in Deeptune's approach. Who Is Building This and Where? Deeptune's roughly 20-person team is based in New York and includes engineers and operators from leading AI and enterprise software companies, including Anthropic, Scale AI, Palantir, Hebbia, Glean, and Retool. Lupo frames New York as a deliberate choice and recruiting advantage: "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". The team's pedigree reflects the specialized talent required to build realistic workplace simulations and train cutting-edge AI agents. Lupo frames the company's core mission as solving the defining problem of the next five years: "how can you make models work not just in fixed exams, but in the messy, real world...that's what we work on here". What Does This Mean for the Future of AI Agents? Deeptune's funding and the broader shift toward reinforcement learning environments signal that the next phase of AI development will focus on agents that can handle complex, multi-step tasks in real enterprise software. Rather than building general-purpose models that answer questions, companies are now investing heavily in specialized agents that can actually perform work, from managing customer support tickets to analyzing financial data. This shift has profound implications for how businesses will deploy AI. Instead of using AI for narrow tasks like summarization or classification, enterprises will increasingly use AI agents to automate entire workflows. Deeptune's training gyms are the infrastructure that makes this possible, allowing AI labs to build and refine agents before they're deployed into production environments where they handle real business processes.
