When Tim Lupo and Lukas Schmit launched Deeptune in 2022, they were chasing the AI dubbing market—the kind of localization technology that promised to help content creators reach global audiences. That business, it turns out, was just the warm-up act.
By late 2024 or early 2025, the New York-based startup had executed a sharp pivot into territory that would seem unrecognizable to its earliest backers: high-fidelity simulation environments where AI agents train to handle spreadsheets, customer support tickets, and coding tasks. The shift apparently worked. On March 19, the company announced it had raised $43 million in Series A funding led by Andreessen Horowitz, with participation from 776 (Seven Seven Six), Abstract Ventures, and Inspired Capital.
The round also drew a telling roster of angel investors: Noam Brown, a research scientist at OpenAI; Brendan Foody, CEO of Mercor; and Yash Patil, who leads Applied Compute. That lineup—a mix of AI researchers and infrastructure operators—hints at where Deeptune believes the market is heading.
Training Grounds for the Agent Economy
What Deeptune builds now are what it calls "training gyms"—simulated environments that replicate popular enterprise software like Slack and Salesforce, allowing AI agents to learn through reinforcement learning rather than static, human-labeled datasets. According to Fortune, the company has constructed hundreds of these simulation environments for leading AI labs, though Deeptune hasn't disclosed which ones.
The platform integrates with just a few lines of code and ships with built-in problem sets, datasets, and testing frameworks. It's infrastructure designed for a specific moment: as AI labs race to build autonomous agents capable of operating within real workplace software, they need places to train them. Deeptune is betting it can become the go-to provider of those digital proving grounds.
"We're seeing a shift from static, human-labeled data to dynamic reinforcement learning in interactive environments," Marco Mascorro and Martin Casado of Andreessen Horowitz wrote in their investment announcement. Mascorro now occupies a board seat, according to SEC filings.
Perhaps the most notable aspect of Deeptune's current incarnation is its team composition. The roughly 20-person operation draws engineers and operators from Anthropic, Scale AI, Palantir, Hebbia, Glean, and Retool—a roster that reads like a greatest-hits compilation of recent enterprise AI plays. They work in person from an office on Park Avenue South, a detail that feels increasingly anachronistic in a remote-first era but may reflect the intensity required to pull off this kind of technical pivot.
The Road from Dubbing to Reinforcement Learning
The transformation wasn't gradual. Earlier investor materials from 2023 and 2024 still described Deeptune as an AI dubbing and localization platform for creators—suggesting the pivot occurred relatively recently, likely sometime in 2024 or early 2025.
The company had previously raised around $3 million in seed funding, according to a 2023 Forbes profile, with backing from Seven Seven Six and VaynerFund. An October 2025 job posting indicated total funding had surpassed $5 million before this latest round.
SEC filings show the current offering at approximately $42.2 million, with the first sale occurring January 28, 2026, and involving 19 investors. The filing also notes a $5.166 million SAFE conversion rolled into the round—a common structure when startups are moving quickly and want to clean up their cap tables.
A Bet on Reinforcement Learning Infrastructure
Andreessen Horowitz's investment thesis positions reinforcement learning environments as a critical—and currently underbuilt—layer of the AI stack. As progress accelerates on computer-use benchmarks like OSWorld and Terminal-Bench, the need for robust training infrastructure becomes more acute.
The timing may prove fortuitous. Fortune cited a ResearchAndMarkets projection estimating the reinforcement learning tools and environments market at roughly $11.6 billion in 2025, ballooning to over $90 billion by 2034. Those are the kinds of numbers that get venture firms excited, though they also attract intense competition.
Whether Deeptune can carve out defensible market share remains an open question. The company will need to prove that its simulation approach delivers measurable value at scale—and that it can maintain its lead as larger players inevitably enter the space. The $43 million gives them runway to make that case.
For now, Deeptune is focused on expanding its engineering and operations teams while accelerating development of its simulation platform. The hiring continues in New York, where the team seems committed to building in person, together. In an industry increasingly shaped by distributed workforces and asynchronous collaboration, there's something almost defiant about that choice.
Then again, executing a pivot this dramatic probably requires being in the same room.
