Judgment Labs

About Judgment Labs

Judgment Labs provides infrastructure to enhance the reliability of AI agents by monitoring their behavior in production environments. The company develops tools that allow developers to track, evaluate, and improve multi-step AI agent workflows in real-time. It addresses the challenge of deploying autonomous agents in critical applications by offering a platform to turn production data—such as reasoning traces, tool usage, and memory—into actionable insights and continuous improvement loops.

Founded in 2024 by childhood friends Alex Shan (CEO), Andrew Li (Chief Scientist), and Joseph Camyre (CTO), the San Francisco-based startup was built on their combined expertise in AI and large-scale systems. Shan researched Natural Language Processing (NLP) at Stanford, Li was an early research hire at TogetherAI, and Camyre built infrastructure at Datadog. Their collaboration stems from a shared history, with Li introducing Shan to NLP and Camyre teaching Shan to code in their youth. The founders identified a gap in the market, recognizing that traditional tools were insufficient for understanding the complex, multi-step behavior of modern AI agents.

On May 12, 2026, Judgment Labs announced it had raised a combined $32 million in seed and Series A funding, with both rounds led by Lightspeed Venture Partners. Other participants include Nova Global, SV Angel, Valor Equity Partners, and Dynamic. This capital is intended to expand the company's team and development efforts.

The core of Judgment Labs' offering is an agent behavior monitoring (ABM) platform and an open-source framework named Judgeval. This toolkit provides deep observability by tracing the entire trajectory of an agent's actions, rather than just evaluating the final output. Key features include comprehensive logging of LLM calls and tool usage, automated behavior classification, real-time alerts for identified issues, and dashboards for tracking performance metrics like latency and cost over time. The system enables clients to define what correct behavior looks like through rules, custom logic, or human scoring, thereby identifying and fixing failure patterns at their source. The platform is designed for sectors where AI reliability is paramount, such as legal, finance, and enterprise support.