AI startup Deeptune has raised $43 million in Series A funding, SiliconANGLE reported on March 19 local time.

The funding was led by Andreessen Horowitz. Participants included 776, Abstract Ventures and Inspired Capital, as well as OpenAI Group PBC researcher Noam Brown (노암 브라운) and Mercor.io CEO Brendan Foody (브렌던 푸디).

Deeptune, headquartered in New York, is building a "training system" for AI agents. Deeptune CEO Tim Lupo (팀 루포) said in an interview with Fortune that it is simulating complex digital work environments for professionals such as developers, accountants and customer support staff through high-quality reinforcement learning.

The strategy is to train AI to automate complex, multi-step tasks using software such as Salesforce and Slack.

Lupo said, "AI cannot learn like a human pilot simply by reading books," and added, "It needs practice in realistic scenarios."

The funding comes as the AI industry faces a looming data shortage. Most usable online data has already been collected, but it is still not enough for AI training. Deeptune is seeking to overcome these limits by approaching data as an engineering and computing problem rather than simply collecting it.

AI models run "rollouts" in Deeptune's simulation environment to practice tasks, and receive virtual rewards when they succeed, guiding them to find optimal learning paths.

Deeptune is currently working with major AI developers worldwide to build hundreds of virtual training systems.