LangChain updates LangSmith: an on-call copilot for alerts, voice trace debugging and Deep Agents Rubrics

LangChain has published its June digest, covering the key updates to the LangSmith platform, a new agent evaluation system called Deep Agents Rubrics, an educational course on deployment, and an announcement of in-person events for the second half of the year.

An on-call copilot for alerts

The central LangSmith update in June is the Fleet on-call copilot, a tool for fast alert triage in agentic systems. It is meant to help the on-call engineer get to the heart of an incident quickly: it gathers context from traces, analyzes error patterns and proposes the most likely cause of the failure, so nobody has to scroll through hundreds of log lines by hand.

Any team that runs agents in production knows the problem: the more complex the agentic system, the harder it is to tell where exactly something went wrong. The error could originate in the prompt, in a tool, in the long-term memory mechanism, or in the orchestration logic of several agents running at once. The Fleet on-call copilot narrows the search space and shortens the time from the first alert to understanding its cause.

New features for developers

Beyond the on-call copilot, LangSmith gets four practical additions:

• Computer use for agents — agents can now interact with a computer interface directly: click buttons, fill in form fields and read screen content in real time
• Voice trace debugging — developers can replay voice interactions right in the LangSmith interface and see exactly where the agent misinterpreted speech
• Experiment status tracking — a new dashboard shows the progress of long test runs without checking each run manually
• Programmatic subagents — the ability to launch nested agent chains directly from code, without manually configuring an orchestration layer

All four updates address specific engineering needs and came from requests by teams that already operate agentic systems in real production, not just prototype them.

Deep Agents Rubrics and training

LangChain announced Deep Agents Rubrics, a system of structured criteria for evaluating complex multistep agents. It answers a long-standing industry problem: how do you objectively measure the quality of an agent that solves a nonlinear task in several stages? Simple "pass/fail" metrics are not enough here.

Rubrics break a complex scenario into subtasks that can be assessed and assign scores to each. This is especially useful when comparing different versions of an agent or choosing between architectural approaches, for example one large orchestrator agent versus several specialized executor agents. Structured evaluation helps teams make engineering decisions on solid ground rather than relying on a subjective feeling that it "seems to work better."

Alongside it, LangChain released a new course on deploying LangSmith for those who want to move from a first prototype to a full production deployment. In-person events are planned for the second half of 2026 in Chicago, Berlin, Washington and Las Vegas.

What this means

The June LangSmith updates mark an important shift: the industry's center of gravity is moving from building agents to running them reliably. The on-call copilot, voice tracing and evaluation rubrics are no longer prototyping tools but operational infrastructure for teams whose agents run in real production.

AI agents are moving from the experimental stage to the stage of engineering discipline.