--- title: LangChain's three-layer agent stack: Deep Agents,... description: LangChain now maintains three distinct open source projects at different abstraction levels: LangGraph (runtime with deterministic graph-based execution),... canonical: https://daily.dev/posts/langchain-s-three-layer-agent-stack-deep-agents-langchain-and-langgraph-explained-czmxajawj twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: LangChain's three-layer agent stack: Deep Agents, LangChain, and LangGraph explained | daily.dev og:description: LangChain now maintains three distinct open source projects at different abstraction levels: LangGraph (runtime with deterministic graph-based execution),... og:url: https://daily.dev/posts/langchain-s-three-layer-agent-stack-deep-agents-langchain-and-langgraph-explained-czmxajawj og:image: https://api.daily.dev/og/posts/czMXAJawJ.png og:image:alt: LangChain's three-layer agent stack: Deep Agents, LangChain, and LangGraph explained og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # LangChain's three-layer agent stack: Deep Agents, LangChain, and LangGraph explained **[Collections](https://daily.dev/sources/collections)** · 2 min read · 5 upvotes · 0 comments ## Summary LangChain now maintains three distinct open source projects at different abstraction levels: LangGraph (runtime with deterministic graph-based execution), LangChain (minimal un-opinionated agent loop), and Deep Agents (high-level harness launched July 2025 with built-in context management, filesystem access, subagents, and memory). Stripe built an enterprise AI assistant called Kai on Deep Agents in about a week, making it the fastest-spreading internal AI tool at the company. The post also covers Microsoft Research's Orchard agent training framework, Cloudflare's durable agent workspace abstraction, Google's session-aware load balancing for long-lived voice agents, OpenAI's GPT-Live simultaneous listen/speak capability, MCP's growing role as an integration layer, and a cautionary note about LLM-generated SQLite CVEs that cited non-existent code and were later downgraded. ## Content LangChain has been busy lately, and most of it circles around one idea: Deep Agents. Here's what's actually going on across the open-source stack, the new managed offering, and the broader shift toward hosted agent infrastructure. ## Three layers, one philosophy LangChain's open-source stack now has three distinct pieces, and Harrison Chase has been clear about where each one fits: - **LangGraph** — the runtime. Graph-based workflows, maximum determinism, human-in-the-loop, fault tolerance. Use this when you need tight control. - **LangChain** — the framework. A minimal, un-opinionated agent loop with middleware hooks for customization. Less structure, more flexibility. - **Deep Agents** — the harness. Launched in July 2025 and inspired by Claude Code and Manus, it bundles context management best practices (filesystem access, subagents, skills, memory) out of the box. All three are composable and share the same underlying design philosophy. If you're starting fresh, the recommendation is to reach for `create_deep_agent` first and only drop down to LangChain or LangGraph when you need more control or determinism than the harness gives you. There's also a course now — LangChain Academy's Deep Agents intro covers dynamic and async subagents plus a course tutor skill, if you want to go deeper than the docs. ## Managed Deep Agents: taking the harness to production The bigger news is Managed Deep Agents, now in public beta. It's a hosted runtime on LangSmith that takes the open-source harness and handles the operational side that usually turns into a multi-week infrastructure project: durable execution, checkpointing, streaming, sandboxed tool access, cross-thread memory, human-in-the-loop workflows, and observability. You write agents in Python or TypeScript using the same harness structure (AGENTS.md, skills/, subagents/, tools.json), test locally, then deploy with one CLI command — `mda deploy`. There's also an API-first interface at `/v1/deepagents` for teams that want to integrate programmatically. A few specifics worth knowing: - **Context Hub** lets agents persist and update working context across runs, so they can actually improve from real usage instead of starting fresh every time. - **LangSmith Engine** can review traces, catch bugs, and update agent context automatically. - Slack and GitHub channel integrations, OIDC-based identity, and Harbor-powered evals come built in. - It's currently available on LangSmith Cloud in the US region, CLI-first. One line from the launch stuck with me: ## Questions this post answers ### What is the difference between LangGraph, LangChain, and Deep Agents? LangGraph is the runtime, modeling agent workflows as graphs for deterministic execution, human-in-the-loop checkpoints, and fault tolerance. LangChain is a minimal, un-opinionated agent loop with middleware hooks. Deep Agents, launched July 2025, is the highest-level harness inspired by Claude Code and Manus, bundling context management, filesystem access, subagents, skills, and memory via `create_deep_agent`. Drop down to LangChain or LangGraph only when you need more control. _Engineers choosing between these layers for a new agent project track real-world adoption stories like Stripe's on daily.dev._ ### What did Stripe build with LangChain Deep Agents and how long did it take? Stripe built Kai, an enterprise-wide AI assistant, on Deep Agents in about a week. The implementation used production middleware and a federated skills system. Kai became the fastest-spreading internal AI tool at Stripe, serving as a practical proof point for the productivity gains the Deep Agents harness abstraction provides over lower-level frameworks. _Teams evaluating Deep Agents for internal tooling find case studies and infrastructure updates like this on daily.dev._ --- Tags: [#llm](https://daily.dev/tags/llm), [#ai-agents](https://daily.dev/tags/ai-agents), [#mcp](https://daily.dev/tags/mcp), [#langchain](https://daily.dev/tags/langchain), [#langgraph](https://daily.dev/tags/langgraph) [View this post on daily.dev](https://daily.dev/posts/langchain-s-three-layer-agent-stack-deep-agents-langchain-and-langgraph-explained-czmxajawj) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"LangChain's three-layer agent stack: Deep Agents, LangChain, and LangGraph explained","url":"https://daily.dev/posts/langchain-s-three-layer-agent-stack-deep-agents-langchain-and-langgraph-explained-czmxajawj","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/langchain-s-three-layer-agent-stack-deep-agents-langchain-and-langgraph-explained-czmxajawj"},"datePublished":"2026-08-06T18:40:47.707Z","dateModified":"2026-08-13T18:17:31.636Z","description":"LangChain now maintains three distinct open source projects at different abstraction levels: LangGraph (runtime with deterministic graph-based execution),...","isAccessibleForFree":true,"articleSection":"Collections","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Collections","logo":"https://media.daily.dev/image/upload/s--fk_6ycEi--/f_auto,q_auto/v1780996001/logos/collections?_a=BAMAMiWQ0","url":"https://daily.dev/sources/collections"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/langchain-s-three-layer-agent-stack-deep-agents-langchain-and-langgraph-explained-czmxajawj","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":5},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"llm,ai-agents,mcp,langchain,langgraph","timeRequired":"PT2M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Collections","item":"https://daily.dev/sources/collections"},{"@type":"ListItem","position":3,"name":"LangChain's three-layer agent stack: Deep Agents, LangChain, and LangGraph explained"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/langchain-s-three-layer-agent-stack-deep-agents-langchain-and-langgraph-explained-czmxajawj#faq","mainEntity":[{"@type":"Question","name":"What is the difference between LangGraph, LangChain, and Deep Agents?","acceptedAnswer":{"@type":"Answer","text":"LangGraph is the runtime, modeling agent workflows as graphs for deterministic execution, human-in-the-loop checkpoints, and fault tolerance. LangChain is a minimal, un-opinionated agent loop with middleware hooks. Deep Agents, launched July 2025, is the highest-level harness inspired by Claude Code and Manus, bundling context management, filesystem access, subagents, skills, and memory via `create_deep_agent`. Drop down to LangChain or LangGraph only when you need more control. Engineers choosing between these layers for a new agent project track real-world adoption stories like Stripe's on daily.dev."}},{"@type":"Question","name":"What did Stripe build with LangChain Deep Agents and how long did it take?","acceptedAnswer":{"@type":"Answer","text":"Stripe built Kai, an enterprise-wide AI assistant, on Deep Agents in about a week. The implementation used production middleware and a federated skills system. Kai became the fastest-spreading internal AI tool at Stripe, serving as a practical proof point for the productivity gains the Deep Agents harness abstraction provides over lower-level frameworks. Teams evaluating Deep Agents for internal tooling find case studies and infrastructure updates like this on daily.dev."}}]} ```