--- type: "post" id: "1-langsmith-engine-v2-2-managed-deep-agents-de58e2" slug: "1-langsmith-engine-v2-2-managed-deep-agents-de58e2" title: "LangChain announced a coordinated LangSmith release spanning Engine v2, Managed Deep Agents v0.8" description: "LangSmith’s announcement is best read as an attempt to turn agent operations into a closed improvement loop rather than as five isolated features." retrieval_nugget: "LangSmith’s announcement is best read as an attempt to turn agent operations into a closed improvement loop rather than as five isolated features." published_at: "2026-09-25" updated_at: "2026-09-25" record_date: "2026-09-25" date_kind: "published_at" topics: [] source_urls: ["https://www.langchain.com/blog/langsmith-engine-agents-fine-tuning-trajectories"] schema_version: "newruntime-agent-readable-v0.2" stable_id: "post:1-langsmith-engine-v2-2-managed-deep-agents-de58e2" status: "published" visuals: [] editorial_provenance: {"schema_version":"newruntime-editorial-copy-v1","content_status":"source_grounded_final","final_copy_sha256":"sha256:a9e7608f7069e8ce2b1384be39c0c331fc8aab1ffc8f21a2276bcb6572bf6f5b","reviewed_at":"2026-09-25T20:05:00.485Z","source_evidence_count":1,"verified_claim_count":4,"site_analysis_schema_version":"newruntime-site-analysis-v1","site_object_kind":"field_note","observed_fact_count":4,"implication_count":2,"watch_condition_count":2,"related_record_count":0} analysis: {"schema_version":"newruntime-site-analysis-v1","object_kind":"field_note","thesis":"LangSmith’s announcement is best read as an attempt to turn agent operations into a closed improvement loop rather than as five isolated features.","observed_facts":[{"text":"LangChain says Engine has analyzed more than 60 million traces since its May launch.","source_urls":["https://www.langchain.com/blog/langsmith-engine-agents-fine-tuning-trajectories"]},{"text":"Engine v2 is described as detecting errors, unmet requests, latency, cost, repetitive tool calls, and overly long trajectories.","source_urls":["https://www.langchain.com/blog/langsmith-engine-agents-fine-tuning-trajectories"]},{"text":"Managed Deep Agents v0.8 adds user-level memory that is separate from agent-level memory.","source_urls":["https://www.langchain.com/blog/langsmith-engine-agents-fine-tuning-trajectories"]},{"text":"Trajectories aggregates messages from humans, AI, tools, agents, and subagents in chronological order.","source_urls":["https://www.langchain.com/blog/langsmith-engine-agents-fine-tuning-trajectories"]}],"mechanism":"The mechanism is production traces becoming a common substrate for diagnosis, review, evaluation, candidate repair, and post-training data preparation.","why_now":"This matters now because long-running, tool-using agents create operational evidence that is too nested for many reviewers to inspect directly.","implications":["The implication is a more integrated path from observed failure to measured improvement, but it also concentrates workflow, data, and governance decisions in one platform.","Teams may be able to use readable trajectory views to involve subject-matter reviewers who do not work directly with execution metadata."],"evidence_boundary":"The evidence boundary is that these are vendor claims in a product announcement; the source does not provide independent performance comparisons, pricing, or evidence that automated fixes improve outcomes across customers.","watch_conditions":["Watch whether teams report lower time-to-resolution and durable quality gains after human review, rather than simply more automated activity.","Watch for disclosed controls governing data isolation, credential use, and human approval in cross-tenant or regulated deployments."],"related_records":[],"new_branch_reason":"No prior New Runtime record was supplied; this announcement creates a distinct record about a unified production-to-improvement agent lifecycle."} editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]} routes: {"html":"https://newruntime.com/posts/1-langsmith-engine-v2-2-managed-deep-agents-de58e2/","markdown":"https://newruntime.com/posts/1-langsmith-engine-v2-2-managed-deep-agents-de58e2.md","json":"https://newruntime.com/posts/1-langsmith-engine-v2-2-managed-deep-agents-de58e2.json"} --- # LangChain announced a coordinated LangSmith release spanning Engine v2, Managed Deep Agents v0.8 ## Retrieval answer LangSmith’s announcement is best read as an attempt to turn agent operations into a closed improvement loop rather than as five isolated features. LangSmith’s announcement is best read as an attempt to turn agent operations into a closed improvement loop rather than as five isolated features. LangChain says Engine has analyzed more than 60 million traces since its May launch and that Engine v2 can detect errors, unmet requests, latency, cost, repetitive tool calls, and overly long trajectories. It also says that, for agents on LangSmith Deployment, the system can test candidate fixes on offending inputs and a broader evaluation set before presenting a result for human review. Managed Deep Agents v0.8 adds user-level memory separated from agent-level memory, alongside user- or agent-owned external credentials. Trajectories provides a chronological conversational view across agents and subagents, while the fine-tuning offering converts trajectories into supervised-training material. The mechanism is production traces becoming a common substrate for diagnosis, review, evaluation, candidate repair, and post-training data preparation. This matters now because long-running, tool-using agents create operational evidence that is too nested for many reviewers to inspect directly. The implication is a more integrated path from observed failure to measured improvement, but it also concentrates workflow, data, and governance decisions in one platform. The evidence boundary is that these are vendor claims in a product announcement; the source does not provide independent performance comparisons, pricing, or evidence that automated fixes improve outcomes across customers. Watch whether teams report lower time-to-resolution and durable quality gains after human review, rather than simply more automated activity. LangSmith’s announcement is best read as an attempt to turn agent operations into a closed improvement loop rather than as five isolated features. The mechanism is production traces becoming a common substrate for diagnosis, review, evaluation, candidate repair, and post-training data preparation. This matters now because long-running, tool-using agents create operational evidence that is too nested for many reviewers to inspect directly. The implication is a more integrated path from observed failure to measured improvement, but it also concentrates workflow, data, and governance decisions in one platform. The evidence boundary is that these are vendor claims in a product announcement; the source does not provide independent performance comparisons, pricing, or evidence that automated fixes improve outcomes across customers. Watch whether teams report lower time-to-resolution and durable quality gains after human review, rather than simply more automated activity. Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json ## Continue in the graph Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`. - collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis. - collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph. - whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)