Crux Labs Raises $6M+ to Audit AI Agents Against Human Expertise

Crux Labs, a New York-based startup developing software to evaluate and improve enterprise AI agents using human expertise, has raised more than $6 million in an oversubscribed funding round. The round was led by Benchstrength and Misneach, with participation from Analog Ventures and Pareto Holdings.

Founded by Julie Kheyfets and Krishna Kulkarni, Crux Labs is building a platform that helps companies evaluate AI agent activity against the judgement and standards of their own subject matter experts. The company is emerging from stealth as enterprises increasingly deploy AI agents across customer-facing and operational workflows.

The funding will support research and development, accelerate enterprise deployments, and help the company expand its engineering and research teams.

Auditing AI agents beyond limited manual reviews

Many companies still manually review only a small portion of their AI agents' activity. According to Crux Labs, that leaves a significant amount of production activity without detailed evaluation.

Crux wants to address this gap by automating the testing and monitoring process.

Its platform connects to a company's existing tools and learns from the knowledge of its internal experts. It then uses that expertise to generate customised tests for the company's AI agents.

These tests are run across production traffic, allowing the platform to identify potential issues, generate alerts and support faster improvements to live agents.

The approach is designed to help businesses evaluate AI systems against their own operational requirements rather than relying entirely on generic evaluation benchmarks.

For industries such as healthcare and financial services, this could mean assessing AI interactions against the standards expected by clinicians, compliance officers and other specialists.

Turning human expertise into continuous AI evaluation

Crux's central proposition is that AI agents can execute tasks but may lack the experience and judgement that human specialists develop over years of working in a particular field.

The company aims to translate that expertise into a continuous evaluation system.

Rather than asking subject matter experts to manually inspect every interaction, Crux attempts to capture their standards and use them to generate tests that can be applied across an agent's production activity.

This creates a feedback process in which businesses can identify where AI agents are not meeting expectations and make improvements based on domain-specific requirements.

The company says its scoring system is customised to each customer's own experts, differentiating its approach from evaluation products that rely on more generalised criteria.

Founders bring experience from Tractable, Block Renovation, Meta and Checkr

Crux Labs was founded by Julie Kheyfets and Krishna Kulkarni, who met at Princeton University.

Julie Kheyfets, co-founder and CEO, previously led Tractable's North American business. Tractable uses AI to assess vehicle accident damage from photographs and raised $65 million in its Series E round in 2023.

Kheyfets later joined Block Renovation, a Brooklyn-based renovation marketplace, where she became CEO in January 2025.

Her experience working with software that makes decisions traditionally handled by human experts helped shape the problem Crux is addressing.

Krishna Kulkarni, co-founder and CTO, previously led brand engineering at Meta and built AI products at Checkr. He also helped take Healthify from zero to 20 million patients.

Kulkarni believes enterprises need a way to connect the technical teams building AI agents with the subject matter experts who understand what successful outcomes should look like.

Crux is designed to provide that connection by turning domain expertise into a continuous process for evaluating and improving AI agents.

Competing in a growing AI observability market

Crux is entering an increasingly competitive market for AI observability, evaluation and agent monitoring.

Several companies have attracted significant funding or strategic interest in this category. Braintrust raised $80 million at an $800 million valuation, while Patronus AI secured $50 million in a Series B. Judgment Labs also raised $32 million in a round led by Lightspeed.

The market has also seen acquisition activity, including Cisco's announced intention to acquire Galileo and ClickHouse's acquisition of Langfuse.

Crux's differentiation is its focus on aligning evaluation criteria with the judgement of each customer's own experts.

The company says it is already working with several Fortune 500 companies and fast-growing startups across healthcare, cybersecurity and financial services. However, it has not disclosed customer names, revenue figures or specific performance results.

Funding to accelerate research and enterprise deployments

The funding round was led by Benchstrength and Misneach, with Analog Ventures and Pareto Holdings also participating.

Crux plans to invest the capital in research and development while accelerating deployments across enterprise customers. It is also hiring researchers and engineers to strengthen its technical capabilities.

Benchstrength General Partner John Monagle said the company's approach addresses the need for AI agents that reflect the judgement of experienced people within an organisation.

As more businesses hand operational responsibilities to AI agents, evaluating whether those systems consistently meet company-specific standards is becoming an important challenge.

Crux Labs is betting that turning human expertise into automated tests can help enterprises improve agent reliability and maintain oversight across production workflows.

The next challenge will be scaling that approach across different industries and proving that its expert-driven evaluation model can deliver measurable value as competition in AI observability intensifies