Crux Labs raises $6M to turn expert judgment into AI agent evaluation

The core challenge is that AI agents often fail to meet business-specific criteria, even when tasks are completed. Crux argues that manual review cannot scale with large deployments, as teams typically sample just 1% of agent interactions, leaving most unchecked. The startup’s solution involves capturing expert insights to calibrate custom evaluations, enabling real-time monitoring and alerts across 100% of production traffic. Julie Kheyfets, co-founder and CEO, emphasizes that agents lack the years of human heuristics needed to align with business goals. The company claims its approach drives iterative improvements in agent behavior without relying on scattered feedback from support tickets or individual reviewers.

Crux’s founding team includes Kheyfets and Krishna Kulkarni, both with enterprise AI experience. The startup reports early traction with Fortune 500 enterprises and high-growth businesses in sectors like healthcare, cybersecurity, and financial services, though no named customers or specific metrics are disclosed.