Test Drive for AI Agents — Patronus Secures $50M

AI agents are taking on jobs like code management, financial analysis, and customer service. Yet, passing a benchmark test doesn’t guarantee they’ll perform well once deployed. That’s why Patronus AI — which was founded in 2023 by former Meta AI researchers Anand Kannappan and Rebecca Qian — has raised $50 million in Series B funding to address the problem. The startup’s revenues have grown 15x over the last year.

Patronus creates “digital worlds” that are synthetic versions of real-life systems (like webpages, databases, internal applications) for testing AI agents. Agents are taught to navigate these worlds using rewards and punishments and learn from their mistakes in safe spaces.

1The test-drive before the real road

It’s like Waymo teaching self-driving cars to navigate roads by simulating millions of miles in dangerous scenarios before putting them on actual streets. Patronus wants to do the same thing for software agents. For now, the company is focusing on software development and finance — sectors where agents can make bold but potentially disastrous decisions.

Patronus just closed a $50 million Series B round led by Greenfield Partners with participation from Notable Capital, Lightspeed, Datadog, and Samsung. Overall, the company has raised $70 million. “We have seen nearly unlimited demand,” said Notable Capital’s Glenn Solomon. “They are really good at spotting the hacks and making sure they are holding the models accountable.” Patronus already works with almost every frontier AI lab.

If Patronus’s revenue increase of 15x over the past year is any indication, this isn’t a small problem. With more companies deploying autonomous agents, figuring out how to ensure they’ll function properly becomes crucial. Patronus is constructing the framework for answering that question, and apparently, lots of teams are already thinking about it.