LangChain raised $300 million at a $3 billion valuation to sell companies a model-neutral platform for running and governing their AI agents. The bet, CapitalG wrote, is that enterprises want an agent layer that is neutral and under their own control rather than one supplied by the model labs and hyperscalers.
CapitalG, Alphabet’s growth fund and an existing investor, led the Series C, which LangChain announced Friday. The valuation is more than double the $1.25 billion set when LangChain raised $125 million about a year ago, TechCrunch reported. A year ago most enterprises still treated agents as “promising science projects,” CapitalG partners Mo Jomaa and Alice Yang wrote, and now agents at some of the largest companies are handling customer support, back-office work and software engineering.
The pitch rests on a phrase that runs through LangChain’s announcement: every company should “own its intelligence.” In a blog post, co-founder and CEO Harrison Chase argued that the agents worth building will be domain-specific, set apart by their knowledge, data and user experience. “Companies should be free to choose the best models for their needs, control their data and infrastructure, and continuously improve their agents,” he wrote. LangChain’s commercial product, LangSmith, is built around that idea. It works with any framework and any model, and it can run in LangChain’s cloud, on a customer’s own servers, or inside the customer’s AWS, Azure or Google Cloud account, CapitalG said.
GIC and DFJ Growth joined as new investors, alongside Salesforce Ventures, Insight Venture Partners, QuantumLight and MongoDB Ventures. IVP, Sapphire Ventures, Sequoia, Amplify and Benchmark put in more, the company said in a thread on X.
More than 7,000 organizations use LangSmith, according to Chase, including Harvey, Rippling, Nvidia, Toyota and Salesforce. Chase wrote that Clay uses it to monitor hundreds of millions of agent runs a month and that Workday runs and governs its agents on it, cutting the time to deploy a new one from five months to under a day.
The free, open-source side of the business has grown too, though the two figures CapitalG gave are not like for like. LangChain and LangGraph, its two core packages, drew about 90 million downloads a month combined at the time of the Series B. Its open-source projects, which now include Deep Agents, draw more than 350 million. Many of the largest enterprise deployments started with a developer who already knew the frameworks, CapitalG’s partners wrote.
The competition is crowded and well funded. CapitalG acknowledged that the model labs and hyperscalers “are all shipping their own agent tooling, and much of it is very good.” Security vendors want in, too: CrowdStrike agreed this month to buy Fiddler and Dreadnode to track and test companies’ agents. And on Thursday, Mirendil, founded by former Anthropic researchers, raised $1 billion on a related bet about agents doing research work.
LangChain has been adding products to the platform. LangSmith Engine, which the company calls an in-platform agent, finds errors in production, writes code or prompt fixes and then watches for regressions. A fine-tuning service converts agent traces into training data for specialized models. Salesforce Ventures, in its note on the deal, wrote: “The longer a company runs on LangSmith, the more it has invested there and the harder the platform is to replace.”
LangChain plans to show new LangSmith features at a release event on Friday, Nov. 20, and is listed at AWS re:Invent in Las Vegas on Nov. 30. Amazon’s cloud unit, which runs that conference, is one of the hyperscalers building agent tools of its own.
WebProNews is an iEntry Publication