LangChain: LangSmith Fleet On-Call Copilot, Computer Use, and Deep Agents RubricMiddleware

LangChain’s June newsletter announced several significant upgrades to the LangSmith platform and the agent ecosystem, with measurable business results among users.

What is an alert triage agent and why does it matter?

An alert triage agent is an autonomous system that, when an alert fires — instead of an on-call engineer — automatically searches relevant traces, source code, and runbooks to identify the root cause and propose a fix. LangSmith Fleet On-Call Copilot is exactly that: a ready-made, prebuilt agent that connects to existing monitoring and reduces the number of alerts requiring human intervention.

Computer Use: agents get their own VMs

The new Computer Use in Fleet capability gives agents isolated virtual machines in which they can run code, read and write files, and call external APIs — all without risk of unwanted side effects on the production environment. Additionally, a new Voice Traces UI displays audio traces with an inline player and span-level visibility, making it easier to debug voice agent flows.

Deep Agents RubricMiddleware: self-evaluation in an open loop

RubricMiddleware is a new open-source component that adds self-evaluation to agents — the ability to assess their own output against defined criteria and iteratively improve it before delivery. Instead of a single pass, the agent goes through a loop until the output satisfies the specified rubrics.

Real-world user results speak for themselves:

• Box used Deep Agents RubricMiddleware to speed up agent improvement 3× compared to its previous approach.
• Harmonic used LangSmith Deployment to increase user retention 4×.

The platform grows toward production maturity

These upgrades together show the direction in which LangChain is developing LangSmith: from a debugging tool toward a complete platform for production agent management — with automated triage, secure execution environments, and measurable business impact.