待翻譯:LangSmith Engine Improves Agent Issue Detection by 2x
AI 服務暫時不可用,以下為來源摘要,待恢復後補全翻譯:LangSmith Engine now detects agent issues over 2x better, proposes stronger fixes, supports Slack and Linear workflows, and is available for self-hosted deployments.
AI 服務暫時不可用,以下為來源正文,待恢復後補全翻譯。
LangSmith New in LangSmith Engine: >2x better issue detection August 25, 2026 5 min Go back to blog Create agents Key Takeaways Engine now offers >2x performance: Engine now performs over twice as well on internal benchmarks for identifying agent issues and 25% better on industry-standard benchmarks for fixing issues. Engine integrates into your flow of work: Get Slack alerts for new issues, open tickets in Linear from within Engine, and use Engine in self-hosted LangSmith deployments. Engine provides greater cost flexibility: Reduced Analysis mode gives users greater control over cost while still using Engine to identify agent issues. Improving agents is hard work. Production agents can generate millions of traces, and scanning each of these for errors, identifying their root causes, and writing fixes for each issue is impractical at that scale. That’s why we built LangSmith Engine – our in-platform agent that analyzes production traces, finds issues, proposes fixes, and monitors for regressions autonomously. With our latest release, Engine is now over twice as performant on internal benchmarks at identifying impactful issues in agent traces and 25% better on industry-standard benchmarks at writing fixes to prompts and code. What LangSmith Engine does Engine automates manual processes across the agent development lifecycle (ADLC). It’s an in-platform Deep Agent that analyzes traces in LangSmith to help engineers move quickly from agent failures to fixes. It identifies issues in your agent traces, grouping related occurrences together and providing the key artifacts needed to resolve them: A diagnosis with evidence: agent runs that show the problem, an incident timeline (including recurrences), and a root cause A proposed fix: a prompt or code change and a ready-to-deploy PR for your review Ground truth examples: the failing runs, formatted as dataset examples, so the fix can be verified offline before it goes out Ongoing monitoring: ongoing, issue-specific monitoring to help you catch and address regressions In practice, this means you can move through the ADLC using LangSmith, with Engine helping to accelerate each step. How we improve Engine We launched Engine in May. Since then, Engine has scanned over 60 million traces to find more than 20,000 issues across our customers’ agents, saving tens of thousands of hours of engineering work. We’re continually improving Engine’s performance, which we measure against several internal benchmarks that assess different components of Engine’s workflow. These include IssueBench, which evaluates how good Engine is at identifying and grouping issues in traces, and another that evaluates the effectiveness of Engine’s proposed fixes. Since its launch in May, we’ve increased Engine’s performance significantly. Engine scores over twice as well on IssueBench, and its fixes score 25% better on public benchmarks like Terminal-Bench. This means that Engine now: Surfaces higher-signal issues Detects more issues across agent runs And proposes more effective fixes Together, these improvements help our users stay focused on their most impactful issues and resolve them even more quickly than before. What else is new in Engine We’re also releasing new features that help customers use Engine within their flow of work and more effectively triage Engine-identified issues. Availability in Self-Hosted Deployments: Engine is now available for self-hosted customers, accelerating agent improvement while meeting enterprise security and compliance requirements. Engine’s orchestration runs inside the customer’s VPC and makes requests to LangSmith Intelligence — our managed, zero data retention service — for managed inference powered by purpose-trained models. Ecosystem Integrations: Engine now better fits into the tools and workflows teams already use to manage issues in their production agents. Slack: Notify agent engineers about Engine-detected issues in Slack so they can start investigating quickly. Linear: Automatically open Linear issues for Engine-detected problems, making it easier to prioritize and track remediation with the rest of the engineering backlog. Reduced Analysis: Reduced Analysis mode lets teams scan fewer traces with Engine. This helps cost-sensitive users reduce cost while still using Engine to find issues. Issue Hygiene: Engine automatically closes stale issues that have stopped appearing in agent traces, focusing engineers’ troubleshooting efforts on active, impactful issues. Stronger Integration with Tracing: Engine points to the specific places in LangSmith traces where a given issue occurred, helping users validate Engine’s findings before proceeding with remediation. What this looks like in production Customers are already using Engine to monitor production agents, surface recurring failures, and move faster from detection to triage. “Engine brings observability and triage into one place. I was amazed by the quality of its root-cause diagnostics and its drafted proposed fixes. It has dramatically reduced our time-to-detection and time-to-triage, meaning less time spent hunting through traces and a much shorter path from broken behavior to a deployed fix.” — Nancy Jin, AI Engineer @ Campfire Where Engine is going next We built Engine to find and diagnose issues first, because that's the part of the lifecycle that is hardest to do manually at scale. Our next focus is verification: Automated Verification: Engine will run its proposed fixes against your datasets and report the results alongside their respective issues, so you see whether a change works before you decide to ship it. Engine-Generated Datasets: Engine will create evaluation datasets from your agent’s traces and prompts, helping you test agent behavior quickly without the burden of building your own datasets. It will also support dataset creation for agents that don’t have existing traces. Getting started Engine is available in SaaS and self-hosted deployments for all LangSmith Plus and Enterprise plans. Pricing details can be found here. If you’re on a Plus plan, you can enable Engine with just a few clicks in LangSmith. If you’re on an Enterprise plan, you can reach out to your account team to get started. New to LangSmith? Get started today by signing up or reaching out to our sales team. See what your agent is really doing LangSmith, our agent engineering platform, helps developers debug every agent decision, eval changes, and deploy in one click. Try LangSmith Get a demo