Crux Labs raises over $6M for human-judgment agent improvement

Crux Labs launched with more than $6 million in funding. Benchstrength and Misneach led the round, and Analog Ventures and Pareto Holdings also took part. The company aims to close the gap between how agents behave and what subject matter experts consider good work.

According to the company, observability platforms help teams trace and inspect agent activity but make it hard to improve behavior at high volume. Teams often review only the top 1% or fewer of traces, chosen by a metric, threshold or at random, so most failures go unnoticed. Crux uses its own agents as evaluators and an improvement team to capture employee judgment and apply it across millions of interactions.

Crux says it has seen strong traction among high-growth startups and Fortune 500 businesses in healthcare, cybersecurity, financial services and other complex industries. Co-founder and CEO Julie Kheyfets and co-founder and CTO Krishna Kulkarni described the divide between engineering teams and domain experts as the core problem.

Teams running agents at high volume may find that sampling a small share of traces for manual review leaves most failures unexamined. Crux's approach of automated evaluation against expert criteria is one option to evaluate. The performance claims are the company's own and have not been independently verified.