Crux Labs Raised $6 Million to Check AI Agents' Work. Any Business Using Them Faces the Same Question

NEW YORK — Companies are handing more customer conversations, claims and back-office tasks to AI agents. Crux Labs, a New York startup that emerged from stealth this week, wants to make sure those agents are doing the work the way a company’s best people would.

Crux announced on Oct. 29 that it had raised more than $6 million in an oversubscribed round co-led by Benchstrength and Misneach, with participation from Analog Ventures and Pareto Holdings, according to its press release. The company says it will use the money to expand research and engineering and speed up deployments with enterprise customers.

The Problem: Checking 1% of the Work

The company’s argument starts with a gap in how AI agents are reviewed. According to Crux, many teams sample just 1% of agent interactions, known as traces, for manual review, which means many failures go unseen. Existing observability tools make traces easier for people to inspect, Crux says, but do not monitor and improve performance across millions of interactions.

SiliconANGLE described the cost of that gap: when the definition of “good” lives only in experts’ heads, fixes arrive late, sometimes weeks after a small glitch first shows up.

How Crux Says It Works

Crux’s own agents connect to a customer’s tools and capture how the company’s subject-matter experts judge good and bad work. Crux then builds custom evaluations calibrated to that judgment, runs them on 100% of production traffic, flags problems and feeds improvements back into the live agents. Tech Funding News put it this way: in health care and financial services, that means applying clinicians’ or compliance officers’ standards to every interaction.

“Agents are incredible at attaining goals, but they lack access to the years of heuristics that we as humans have collected in our heads,” co-founder and CEO Julie Kheyfets said in the release.

Co-founder and CTO Krishna Kulkarni described the gap as organizational. “Engineering teams know how to build AI agents, and subject matter experts know what success looks like, but enterprises lack the bridge between them,” he said.

Crux says it has customers among Fortune 500 companies and fast-growing startups in health care, cybersecurity, financial services and other industries. It has not disclosed customer names, revenue or results, Tech Funding News reported.

The Founders

Kheyfets and Kulkarni met at Princeton. Kheyfets previously led North America for Tractable, which uses AI to assess accident damage, and later became CEO of Block Renovation, the Brooklyn-based renovation company. Kulkarni led brand engineering at Meta, built AI products at Checkr and helped scale Healthify, according to the release.

Kheyfets told Tech Funding News that in both of her earlier roles, software made decisions that would normally fall to an expert. She sees the same gap with AI agents.

A Crowded Field

Crux is entering a busy market. Tech Funding News listed recent activity among agent evaluation and observability companies: Braintrust raised $80 million at an $800 million valuation in February, Cisco announced plans to acquire Galileo, ClickHouse acquired Langfuse in January, Patronus AI raised a $50 million Series B and Judgment Labs closed a $32 million round. Crux says its difference is calibrating scores to each customer’s own experts. The open question, as Tech Funding News noted, is whether a young company can own that process or whether it becomes a feature inside bigger platforms.

What Small and Midsize Businesses Can Take From This

Most New York small businesses won’t buy an enterprise evaluation platform. But many already use AI chatbots, voice agents or automated email tools for customer service, scheduling or sales. The problem Crux describes applies to them too. A few basic habits, which don’t require special software, can help:

• Write down what “good” looks like. List the answers, tone and escalation rules your best employee would follow, and use that list to judge the AI’s work.
• Review more than a handful of conversations. Read a regular sample, and pay extra attention to complaints, refunds and anything involving money, health or legal issues.
• Track the edge cases. Unusual requests are where agents often fail. Keep a running list and test new versions against it.
• Keep a human escape hatch. Make sure customers can reach a person, and that the agent knows when to hand off.
• Ask vendors how they measure quality. Before signing up, ask an AI vendor how it tests its agents, what it monitors after launch and how you’ll see the results.

As AI agents take on more customer-facing work, checking their output is becoming part of running a business, not just a job for engineers.

Entrepreneurs Of New York is an independent publication and is not affiliated with Crux Labs or its investors.