Blog

Congratulations to Harrison, Ankush, Julia and everyone at LangChain on today’s $125M Series B funding announcement. It’s been a delight working with this team since leading the Series A. 

Beyond the funding, today marks the release of langchain 1.0, a major rite of passage for any open source company. Releasing 1.0 reflects the team’s conviction that LangChain, as currently built and packaged, is the right foundational architecture to become the scaffolding for agent engineering.

LangChain is one of the most popular open-source projects in AI with 90M monthly downloads, and is used by 35% of the Fortune 500. The company has quickly grown from an open source package to a full-fledged agent engineering platform, including a low-level agent orchestration and runtime called LangGraph that’s used by companies such as Uber, Klarna, LinkedIn and J.P. Morgan; and an observability and agent deployment platform called LangSmith used by Clay, Cloudflare, Replit, Vanta, Rippling, Mercor and more.

LangChain is one of the most used, debated and sometimes misunderstood projects and companies in AI. Below are some reflections on the evolution of LangChain to date and how I view the opportunity ahead.

Early Days

Harrison Chase’s superpower is simple: he sees the future of agents. He saw it before just about anybody, and has consistently pushed the frontier of agent engineering forward.

Harrison started LangChain as a nights-and-weekends side project back in 2022, right as ChatGPT was taking off. The original LangChain was a proof-of-concept of what language models were capable of becoming: connect them with tools, chain them together into a sequence of calls and business logic, and voila, model becomes agent. 

Harrison’s vision struck a chord with developers, and the LangChain project quickly became one of the fastest adopted open source projects. By the time we led the Series A in early 2023, it had more than 2,000 open-source contributors and was being used by more than 50,000 LLM applications. After two years of exponential growth, it is now surpassed only by OpenAI in download volume.

Crucible Moment: LangGraph

As the ecosystem matures, LangChain has matured alongside it, constantly growing the core open source package while also adding in new capabilities.

Because the LLM ecosystem is evolving at such a rapid clip, maintaining the scaffolding to support a developer ecosystem of this size hasn’t always been easy. But Harrison has been resilient and unflappable, constantly listening hard to his challengers and evolving the package towards the future.

Sometime last year, it became increasingly clear that the abstraction level on which the OG LangChain was built was a convenient launching point for developers, but wasn’t sufficiently low-level enough to support all the granular control that developers were craving as the space matured.

Harrison made the hard decision to develop another framework in parallel: LangGraph allows developers to control every step of their custom agent with low-level orchestration, memory, and human-in-the-loop support, and manage long-running tasks with durable execution. 

This was a tough but remarkably wise call. Today, LangGraph is the most popular low-level agent orchestration framework and runtime, used by companies including Uber, Klarna, LinkedIn and J.P. Morgan and downloaded 12M times a month.

Getting one open-source project to broad product-market fit could be a function of luck. Getting two open-source projects to PMF is a testament to how clearly Harrison and the broader LangChain team see the future of agent development.

LangChain 1.0

Today marks a new chapter, merging LangChain’s past and future: the team has completely rewritten langchain in its 1.0 release to be opinionated, focused and powered by LangGraph’s runtime. 

This means that langchain 1.0 has all the “0-to-1 booster fuel” the community has come to expect—pre-built architectures for common agent patterns, improved model and tool integrations, and more—but gives developers access to a deep level of customization and control. 

In addition, helping developers on their agent engineering journey means expanding beyond frameworks into developer tools. Working with LLMs requires a mindset shift, from deterministic software to embracing stochasticity, constantly observing and testing data. To that end, the team has invested heavily in LangSmith, which enables agent engineering across the full lifecycle: observability, evaluation and deployment.

LangSmith has proven successful as a standalone developer platform for LangChain and non-LangChain developers alike, and is currently used by companies such as Clay, Cloudflare, Replit, Vanta, Rippling and Mercor, with traffic growing 12x year over year.

What’s Ahead

LangChain’s ambitions are large. This team will not stop until they become the de facto platform for all of agent engineering. They will get there by deepening their core offerings, and continuing to push the frontier.

Deepening the core: The team has pushed hard on deployment, making it possible to ship your agent in one click on LangSmith, using scalable infrastructure built for long-running tasks. They’ve also launched their first 1P agent, an Insights Agent in LangSmith Observability that automatically categorizes agent behavior patterns, as well as a no code text-to-agent builder experience for business users.

Pushing the frontier: As the model space progresses, the agentic building blocks also evolve. LangChain has been hard at work on many of the core primitives we believe will drive the future of agentic engineering, including memory.

It’s hard to believe LangChain is just three years old. We are proud to partner with a team that has adapted so resiliently, and more important, driven so much innovation in the agent engineering ecosystem. Today’s milestone is just the beginning.