Tech news in 3 minutes
Patronus AI lands $50M to build ‘digital worlds’ that stress-test AI agents
AI agents are evolving from answering simple questions to autonomously executing multi-step complex tasks like booking trips or financial analysis. However, ensuring their reliability across diverse real-world scenarios remains a challenge. Patronus AI, founded in 2023 by former Meta AI researchers Anand Kannappan and Rebecca Qian, addresses this by building simulated digital environments to evaluate agent performance. The San Francisco-based startup has attracted virtually every frontier AI lab and many emerging startups as customers, according to Glenn Solomon of Notable Capital, who describes demand as nearly insatiable. Patronus’ revenue has grown 15-fold in the past year, and on Thursday it announced a $50 million Series B round led by Greenfield Partners, with participation from Notable Capital, Lightspeed, Datadog, and Samsung, bringing total funding to $70 million. Patronus uses "digital world models" to create replicas of websites and internal systems. In these environments, agents are stress-tested after training using reinforcement learning, which rewards successful task completion and penalizes errors. This approach mirrors how Waymo trained autonomous cars in synthetic worlds to handle rare hazards like severe weather. For AI agents, the challenge is their tendency to take shortcuts and fail tasks correctly. "Patronus is really good at spotting the hacks and making sure they are holding the models accountable," Solomon said. Currently focused on software engineering and finance, Patronus plans to expand into areas that are harder to verify. "We want to be able to actually create the environment in which you can operate an agent that can run for 10 hours or 10 days or 10 weeks," Kannappan said. The startup primarily competes against internal evaluation teams at AI labs rather than human-data firms like Mercor or Surge, as Patronus evaluates agent behavior without human involvement.