Judgment

Judgment Labs is a provider of a continuous-improvement stack for AI agents that monitors and improves agent behavior at scale in production environments. The platform captures traces of agent interactions and provides behavioral signals from production data, giving teams continuous visibility into how agents act and where they fail. Key capabilities include agent search that queries across trajectories at a behavioral level beyond simple keyword filters, agent judge technology that creates cheaper and more accurate trajectory-level evaluators using harnesses, behavior discovery that surfaces failure modes and usage patterns from unlabeled production trajectories, and auto-rubrics that automatically construct and refine evaluation rubrics from verifiable signals. The platform integrates directly with Slack for real-time investigation, allowing teams to ask questions about agent failures, triage issues by deploying agent swarms to find similar failure cases, test proposed fixes against production cases before shipping, and automatically track behaviors to protect against model drift and regressions. Judgment also offers MCP integration with Claude, Codex, and Cursor for workflow-embedded monitoring. Founded in 2025 and headquartered in San Francisco, United States, Judgment has raised 2 million at the Series A stage with investors including Lightspeed Venture Partners, Nova Global, SV Angel, Valor Equity Partners, and Dynamic Growth.

AI Infrastructure > Enabling Technologies > Software > Testing & Optimization

College Wise > University of California (Los Angeles), University of California (Berkeley), Stanford University