LangChain Raises $300M at $3B as 7,000 Companies Run Agents on LangSmith

Workday used to need five months to put a new AI agent into production. With LangSmith it now takes under a day, and on November 13 investors valued the company behind that tool at $3 billion. LangChain raised $300 million in a Series C led by CapitalG, Alphabet's independent growth fund, at 2.4 times the $1.25 billion valuation it carried about a year ago. For the more than 7,000 organizations already running agents on its platform, the round is a bet that companies will want to build their own agents rather than rent someone else's.

From an Open Source Library to $3 Billion

LangChain began in 2022 as an open source project: a library that let developers connect a large language model to documents, search and other tools. Harrison Chase, its co-founder and chief executive, built a company around it, and the code became one of the most common starting points for anyone trying to make a model do more than chat.

The company's business today is LangSmith, a platform for running, watching, testing and governing AI agents, whichever framework or model they were built with. That neutrality is what investors paid for on November 13.

CapitalG, an existing backer, led the $300 million round. GIC, Singapore's sovereign wealth fund, DFJ Growth, Salesforce Ventures, Insight Venture Partners, QuantumLight and MongoDB Ventures came in as new investors, and IVP, Sapphire Ventures, Sequoia, Benchmark and Amplify put in more. LangChain says it has now raised $460 million in total.

The more telling number sits elsewhere: how many companies already depend on the product, and what they do with it.

Five Months to Under a Day at Workday

An agent is software that takes several steps on its own: it reads, decides, calls a tool and tries again. Building one that works in a demo takes an afternoon. Keeping it reliable once thousands of people use it is the hard part, because nobody can see what it did, or why, without the right instruments.

LangSmith is organized around that problem in three layers. The runtime hosts agents built with LangChain's open source libraries, LangGraph and Deep Agents, along with sandboxes and managed agents. The improvement layer records every step an agent takes, scores its answers and can fine-tune models on the results. The governance layer adds gateways that control which models and tools each agent is allowed to call.

More than 7,000 organizations now use it, among them Nvidia, Toyota and Salesforce, and startups such as Harvey, Rippling and Listen Labs. Clay monitors hundreds of millions of agent runs a month on the platform. Workday cut the time it takes to put a new agent into production from five months to under a day.

That last figure explains the round better than any valuation. So why would a company want all of this from a platform that does not sell its own model?

Why 'Own Your Intelligence' Is the Pitch

LangChain's answer is a slogan Chase has been repeating on stage: own your intelligence. The traces, test sets and fixes a company builds while teaching an agent its job are valuable, and LangSmith keeps them with the customer even when the model underneath changes from one provider to another.

“We're building LangSmith to give companies the platform to own their intelligence. — Harrison Chase, co-founder and CEO of LangChain”

For a bank, a retailer or a hospital that wants agents handling real work, that means switching to a cheaper or better model next year without starting over. Chase says the move toward a world where every company builds and operates its own agents is still early. LangChain will show what the new money builds first at a LangSmith release event on November 20.

Four years after it started as a library on GitHub, LangChain is selling the tools that let any company teach an agent its trade and keep the lesson.