Judgment Labs raises a Series A to help agent-native startups improve long-horizon agents from production data

Judgment Labs is coming out of stealth with a Series A co-led by Lightspeed Venture Partners and Green Oaks, adding to earlier backing from Nova at preseed and Lightspeed at seed. No round size was disclosed.

The company's product sits between the logs agent-native startups already generate and the improvements those teams actually want to make. Long-horizon autonomous agents produce vast amounts of runtime data — reasoning tokens, tool calls, retries, memory traces — but most teams end up manually combing through tables trying to find failure patterns. Judgment Labs uses agents to parse that data automatically, surfacing which failure modes occur most frequently, which customer segments are affected, and which task types are most exposed. The output is actionable enough to point a team directly at the broken part of an agent framework rather than leaving them to guess.

Alex Shan frames the commercial logic around a flywheel: companies with large user bases generate more production data, and Judgment Labs converts that distribution advantage into a durable product improvement cycle. Monaco, the AI sales agent company, is a named customer.

“Judgment Labs is the platform for improving long horizon agents from production data. These autonomous long horizon agents are gonna consume the vast, vast majority of the economic value that AI is set to create across the next decades. Nova backed us at the preseed all the way to Lightspeed backing us at the seed and Lightspeed doubling down to co-lead the Series A with Green Oaks.”

Target customer

The sweet spot is Series A to Series B agent-native startups producing enough production data to make the analysis meaningful. Judgment Labs works across vertical agents — coding, site reliability, ticket resolution, finance, legal, and sales — though Shan argues domains with verifiable outcomes will get solved first. Coding agents are the clearest example: the feedback loop is tight and the pass/fail signal is unambiguous. Tax and accounting follow similar logic. Domains like drug discovery, where outcomes take years to confirm, are structurally harder and further out.

The core bet is that as long-horizon agents proliferate, the companies that learn fastest from production will win, and the gap between raw logs and actionable insight is wide enough to support a dedicated platform to close it.