Patronus AI just raised a $50 million Series B. The round was led by Greenfield Partners with participation from Notable Capital, Lightspeed, Datadog, and Samsung. Total funding now sits at $70 million.
The San Francisco startup, founded in 2023 by former Meta AI researchers, builds simulated digital environments where AI agents get stress-tested before deployment. Think crash-test dummies for autonomous software.
Here’s the problem Patronus is solving. AI agents are moving from answering questions to executing multi-step tasks on their own — booking trips, running financial analysis, writing code. But benchmarks that test model performance don’t actually prove an agent can handle real, messy scenarios. Agents take shortcuts. They find hacks. Patronus spots them.
The company builds what it calls “digital world models” — replicas of websites and internal systems. Agents run through these simulated environments after training with reinforcement learning. The approach mirrors how Waymo tests self-driving cars in synthetic worlds before putting them on real roads.
Demand is intense. Glenn Solomon from Notable Capital calls it “nearly insatiable.” Revenue has grown 15x over the past year. Virtually every frontier AI lab and many emerging startups are customers.
Right now Patronus focuses on software engineering and finance — domains where you can verify whether the agent completed the task correctly. But co-founder Anand Kannappan says harder, non-verifiable domains are next. The goal: environments where agents can run for hours, days, even weeks.
Competition comes mainly from internal evaluation teams at AI labs, plus human-data firms like Mercor and Surge. Patronus differentiates by evaluating agent behavior without human involvement — fully automated testing in simulated worlds.
