LangChain

Confidence 0.90 · 16 sources · last confirmed 2026-10-11

A US-based AI company building agent-engineering infrastructure. Founded in 2022 by Harrison Chase as the open-source langchain Python framework; expanded over 2023–26 into a layered product stack covering most of the Agent Development Lifecycle. The company entered the wiki as a named-but-unsubstantiated organization across agent-harness, ai-agents, and generative-ai before Chase 2026 (9 May) supplied the first first-party LangChain source — at which point the cumulative-mention threshold for entity promotion was met.

Why LangChain matters in this wiki

Bumped to source_count: 9 on 12 August 2026 with a ~2-minute Toyota customer-testimonial clip — the wiki’s first LangChain source from a customer rather than the company or its conferences, and its first automotive-manufacturing deployment. Ravi Chandu Ummadisetti, Director of Agentic AI and Product Research at Toyota North America, describes R&D GPT: an internal agent searching Toyota’s research, paint-corrosion and manufacturing data in plain language. Two details are worth the page even at that length — his description of Deep Agents as “one single command… having this entire ecosystem of harness” pulling tools in “from different wings,” and Toyota’s practice of authoring per-function curated skills (branding, research, manufacturing, supply chain) from a central Enterprise AI group to “feed in the institutional knowledge that we have at Toyota.” On LangSmith he singles out Polly and Insights for production observability — specifically usage composition and outliers. It is promotional content with no numbers of any kind; see that source’s scope warning.

LangChain’s product stack is the clearest worked example of the four-layer Build vocabulary Chase 2026 articulates — frameworks / runtimes / harnesses / no-code — because LangChain ships at every layer:

Layer (Chase 2026)	LangChain product	What it is
Frameworks	LangChain	Open-source abstractions for model calls, tools, prompts, retrieval, structured outputs, agent loops
Runtimes	LangGraph	Stateful agent runtime; control flow / branching / looping / persistence / pause-resume
Harnesses	Deep Agents	Prompts, skills, MCP servers, hooks, middleware, memory, virtual filesystem
No-code	LangSmith Fleet	Domain-expert-facing UI to create and configure agents without code
Test / Deploy / Monitor (lifecycle phases)	LangSmith Platform (Observability, Evaluation, Deployment, Sandboxes)	Tracing, evals, durable runtime, sandboxes

This makes LangChain a vendor whose product taxonomy is the wiki’s vocabulary for the ADLC — both because Chase coined the four-layer Build split, and because LangChain ships at all four layers.

Products referenced in this wiki

• LangChain (the framework) — the original 2022 release; agent abstractions over LLM providers.
• LangGraph — agent runtime for stateful, durable, human-in-the-loop graph-shaped workflows.
  • Evaluated on a competitor’s stack in [[2026-09-30-zamora-feroz-why-your-ai-agent-fails-in-production|Google Cloud Tech’s AI Agent Clinic (September 2026)]]: DocsHound, a LangGraph agent whose traces LangSmith displays, is scored by Google’s ADK-built eval toolkit once its spans carry OpenInference labels. A deterministic check reads the LangGraph span sequence directly.
• Deep Agents — open-source pattern (langchain-deep-agents) demonstrating the harness layer with virtual-filesystem-as-working-memory.
• LangSmith — eval/observability/deployment platform.
  • LangSmith Platform — the umbrella.
  • LangSmith Observability — traces, signals, feedback, dashboards.
  • LangSmith Evaluation — datasets, metrics, experiments.
  • LangSmith Deployment — durable agent runtime hosting.
  • LangSmith Sandboxes — isolated execution environments.
  • LangSmith Fleet — no-code agent configuration. Operational metrics from LangChain’s internal go-to-market agent (per Interrupt 26 demo): 84% of go-to-market team uses weekly; lead-to-qualified conversion up 240%; 40 hours saved per rep per month. Originally built in code; rebuilt in Fleet so the GTM team could own it end-to-end without code. 200+ built-in tools; Arcade partnership for 7,500+ additional tools; MCP support; native Slack / Gmail / Outlook channel integration; cost tracking + usage controls; first-class human-in-the-loop; built on top of deep agents; downloadable agent files for code modification.
  • LangChain Academy — educational resources.
• LangChain Labs — research group inside LangChain “aimed in particular at continual learning” — announced at Interrupt 26 (Day 2 keynote, 21 May 2026). LangSmith’s trace + feedback data substrate named as the foundation for the Labs continual-learning agenda.
• LangSmith Engine — an agent over a tracing project that clusters issues, proposes code fixes and adds evals and dataset examples. First announced at Interrupt 26. Engine v2 was announced at Interrupt NYC (Sep 2026): it tests fixes on LangSmith Deployment preview branches, red-teams a deployed agent from its traces and GitHub repo, and comes to self-hosted with bring-your-own-key in v17. Vendor figures: 70M traces scanned, >21K issues, 2x better finding / 25% better fixes / 40% lower cost (no method).
• Launched or spotlighted at Interrupt NYC (24 Sep 2026):
  • Managed Deep Agents (0.8): Deep Agents harness plus managed infrastructure, “this company harness”. New: auth primitives (connections, user- or service-identity, channels), user-level memory in Context Hub, built-in web search via Parallel.
  • LangSmith LLM Gateway (beta): OpenAI/Anthropic-format normalisation, pass-through, coding-agent integrations, spend limits, rate limiting and fallbacks; it hosts TypeSafe’s Jev and the open-source decision model SemIf.
  • LangSmith Trajectories: one message-list format parsed from many SDKs’ traces.
  • LangSmith Fine-Tuning with the smithtune CLI (training on Fireworks AI or Baseten).
  • LangSmith Custom Apps, built through LangSmith Chat.
  • SmithDB: an in-house store for agent traces, launched at Interrupt 26.
• Deep Agents + OpenShell blueprint — a joint announcement with NVIDIA (Chase interview, 8 Jul 2026): LangChain Deep Agents running Nemotron 3 Ultra inside OpenShell, NVIDIA’s secure open runtime. Packages model + harness + runtime + acceleration stack as a reusable enterprise starting point for building domain-specific “super agents” — the wiki’s first joint-vendor blueprint spanning a harness company and a substrate/silicon company.

Conferences

• Interrupt 2026 — LangChain’s first major industry conference. Day 2 keynote (Chase + Sproul + di Vittorio) anchored the wiki’s articulation of the two-types-of-agents typology (long-horizon vs customer-experience) and the three-layer continual-learning model (model / harness / context).
• Interrupt NYC (24 Sep 2026; London followed on 13 Oct) — the first New York edition. Chase’s opening keynote framed the company’s mission as “enable every company to own their intelligence” and walked through LangSmith as runtime, observability and evals, and an intelligence layer.

Concepts LangChain co-shapes in this wiki

• agent-development-lifecycle — Chase’s 4-phase + governance ring is the wiki’s second formalization of the ADLC and the source most influential on the concept page’s current shape.
• agent-harness — the frameworks vs. runtimes vs. harnesses vs. no-code sub-layering refines the wiki’s harness vocabulary.
• ai-agents — LangChain has been named throughout the wiki as a top-of-mind framework when the topic is agent abstractions.
• generative-ai — LangChain is named in the deployed-tools landscape.

People

• Harrison Chase — co-founder/CEO of LangChain. Promoted from Dangling to entity page on 21 May 2026 after the second substantive source (Interrupt 26 keynote) followed the ADLC essay (9 May) 12 days earlier. The canonical vendor-CEO voice on agent-engineering infrastructure; coined both the frameworks / runtimes / harnesses / no-code four-layer Build vocabulary and the model / harness / context three-layer continual-learning model. See the entity page for the full framings catalogue.
• Sydney Runkle — product manager, LangChain open-source team. Lead author and presenter of the Jev integration material (September 2026); promoted to an entity page on 22 September 2026.
• Brace Sproul — LangChain (product walkthrough at Interrupt 26 ~17:23–17:46; “LangSmith Fleet is built on top of deep agents”). Currently Dangling; promote on second-source mention.
• Caroline di Vittorio — LangChain (live Fleet demo at Interrupt 26 ~17:46–19:30; quantified the internal GTM-agent metrics). Currently Dangling; promote on second-source mention.

Partners

• NVIDIA — joint Deep Agents + OpenShell blueprint (8 Jul 2026), pairing Nemotron 3 Ultra with the Deep Agents harness inside NVIDIA’s OpenShell runtime.

Credited as the origin of harness, and told it isn’t new (2026-09-03)

The GitHub Podcast, S02E02 is the wiki’s first source in which a competitor’s developer-advocacy team names LangChain, unprompted and from memory, as where the harness vocabulary came from: “the first time I saw harness and harness engineering defined, I’m pretty sure was Harrison from LangChain… and I think Viv from their team was one of the first people as well to start using that term.” That is Harrison Chase and Vivek Trivedy (The Anatomy of an Agent Harness) — corroboration of the attribution agent-harness already holds, arriving from outside the LangChain orbit.

The same host then denies the construct is new, on LangChain’s own evidence: “historically with LangChain you could build an agent and basically you were already building a harness, because you were using this infrastructure where it was developing a small loop and you could add context to that, you could add MCP servers… to me it’s very similar to what we already had with just an agent.” Both halves are worth the page. LangChain is credited with naming the layer; it is simultaneously used as the proof that the layer predates the name.

Distributing a third-party model class: Jev (2026-09-22)

In one week of September 2026 LangChain published three pieces on Jev, the non-generative “System One” decision model from TypeSafe AI, and shipped the langchain-typesafe integration. This is the first time the wiki records LangChain as the distribution channel for another company’s model, not as the builder of the layer around it:

• Building a Harness with Jev (blog): two harness middleware components, ModelRouterMiddleware and AutoModeMiddleware. The second is pitched as open-sourcing the tool-risk classifier that closed coding harnesses keep internal.
• Jev-as-a-Judge for Agent Evals (blog): a five-case LangSmith experiment comparing Jev with GPT-5.6 Luna, GPT-5.6 Terra and Claude Sonnet 4.6 as eval judges. It is the only measurement of Jev in the wiki.
• Building a Harness with Jev (video, presented by Sydney Runkle).

All three ended by promoting a joint livestream with TypeSafe on 22 September. Read them as partner marketing that includes one small, candidly scoped experiment.

LangChain Labs: three questions for agent engineers (2026-10-11)

Jake Broekhuizen (LangChain Labs, October 2026) turns the own your intelligence message of the Interrupt NYC keynote into an engineering method: a fine-tuned judge that mines production traces for perceived errors, world and task specs that reproduce failures in an environment (illustrated with LangChain’s internal GTM agent over Gong, Salesforce and calendars), and fixes at the harness or model layer validated by re-running the same task. Vendor talk; no figures.

Open questions

• Customer evidence. The wiki has not yet ingested a customer case study deploying LangChain at production scale. Chase’s piece references “LangSmith customers” abstractly; substantive case studies would substantiate concept pages further.
• Open-source community vs. commercial product. LangChain straddles open-source (the langchain framework, LangGraph, Deep Agents) and commercial (LangSmith Platform). The boundary is operationally important for procurement and lock-in concerns; the wiki has not yet substantiated it.
• Comparative positioning. CrewAI, Claude Agents SDK, AWS AgentCore, Daytona, E2B, Temporal, n8n are all named alongside LangChain in Chase 2026. None has been substantiated independently. A market-landscape source (analyst report, comparison study) would resolve the relative positioning question.