Judgment Labs Opens General Access to Its Trace Monitoring and Continuous Improvement Platform for Long Running AI Agents
Today, @JudgmentLabs opens for general access. We believe long running agent systems will generate more exhaust data than the world can currently process. Hidden within agent traces are rich records of how to make these systems better: what users ask for, when agents stumble, and how agents interact with the external world. Judgment is the most performant monitoring and improvement layer for long-horizon agents. We’re excited to see what you build!
Context
As enterprise AI shifts from single turn chatbots to multi hour autonomous workflows, traditional input output evaluations fail to catch errors buried deep inside thousand step trajectories, creating demand for automated server side trace judges.
San Francisco based Judgment Labs emerged from stealth in May 2026 with 32 million dollars in combined Seed and Series A funding led by Lightspeed Venture Partners and Greenoaks, with participation from Nova Global, SV Angel, Valor Equity Partners, and Dynamic. Prior to opening self serve general availability on October 27, the company spent five months deploying its observability pipeline with design partners running production multi step agents.
Deep dive
On October 27, Judgment Labs CEO Alex Shan announced that the company's continuous monitoring and improvement platform for long running AI agents is now open for self serve general access. Shan noted that autonomous agent systems generate massive volumes of execution exhaust data, and that the root causes of agent failures are typically hidden deep inside multi step trajectories rather than obvious in final responses.
Built on OpenTelemetry and paired with the Apache 2.0 licensed Judgeval SDK for Python and TypeScript, Judgment captures complete agent trajectories including reasoning chains, model calls, tool invocations, handoffs, and token usage across frameworks such as Claude Agent SDK, OpenAI Agents SDK, LangGraph, Vercel AI SDK, and Google ADK. Teams can also connect their coding agents directly to Judgment over a Model Context Protocol server to automate instrumentation setup and query production traces via a read only SQL interface.
Instead of relying on manual trace review, engineers define custom Agent Judges that evaluate live production traffic server side with zero latency impact. These judges inspect long traces to flag specific behavioral failures such as ignoring newly retrieved evidence or prematurely escalating tasks, attach structured behavior labels with cited span evidence, and trigger real time Slack alerts. Teams can then replay production judges against historical trace datasets in continuous integration pipelines to verify whether prompt, tool, or model updates fix regressions before deployment.
What to expect next
Having doubled its engineering headcount over the past three months and opened self serve onboarding alongside full compatibility with existing observability platforms like LangSmith, Logfire, Braintrust, and Langfuse, Judgment Labs is positioning its Agent Judges as the default continuous integration regression gate for production agent teams. Watch for deeper automated prompt, tool, and harness optimization loops built directly on top of labeled production trace datasets.