LangChain raises $300 million at a $3 billion valuation for AI agents

CapitalG led the round. It is an existing investor, which means it backed LangChain before and chose to put in more money.

GIC and DFJ Growth joined as new investors, and IVP and Sapphire Ventures came back. LangChain also welcomed Salesforce Ventures, Insight Venture Partners, QuantumLight and MongoDB Ventures. It thanked Sequoia, Amplify and Benchmark for doubling down.

By the numbers: a year ago, in October 2025, LangChain raised $125 million at a $1.25 billion valuation. The new price is more than double that.

The company says the money will help it push toward one goal: letting every company "own its intelligence." We will unpack what that means below.

An AI agent is a program that uses an AI model to do jobs for you. A chatbot answers a question. An agent can go further: search the web, fill in a form, look something up in a database, then report back.

Companies are building agents to answer customer emails, check contracts or sort sales leads. LangChain's bet is that businesses will want many agents, each trained on their own data and built for one job.

In real life Think of an agent as a new intern who never sleeps. Someone still has to hand out tasks, check the work and make sure the intern does not wander into the wrong filing cabinet.

LangChain began in 2022 as an open-source project by Harrison Chase, then an engineer at a machine learning startup. Open source means the code is public and free to use.

Early AI models, called LLMs (large language models, the engines behind chatbots), were bad at some basic tasks. They could not search the web, call other software through an API (a doorway that lets one program talk to another) or read from a database.

LangChain's code solved those problems, and developers loved it. Chase turned it into a company with a $10 million seed round from Benchmark in April 2023. A week later came a $25 million Series A led by Sequoia, reportedly at a $200 million valuation.

Then the big AI labs began adding those features themselves. That left LangChain with a question: what do you sell when the model makers give away your first trick?

The answer is a platform called LangSmith. LangChain describes it as a framework-agnostic platform to run, improve and govern agents. "Framework-agnostic" means it does not care which tools you used to build the agent.

LangChain says more than 7,000 organizations already rely on it. The list includes startups such as Harvey, Rippling and Listen Labs, and large companies such as NVIDIA, Toyota and Salesforce.

The company shared two customer examples:

• Clay: uses LangSmith to monitor hundreds of millions of agent runs per month and improve its agents.
• Workday: uses LangSmith to run and govern agents across the company, cutting the time to launch a new agent from five months to under a day.

Both numbers come from LangChain's own announcement, so treat them as the company's account.

LangChain splits LangSmith into three parts. Each one answers a worry a company has about agents.

• Agent Runtime: the machinery that runs agents reliably. It includes Deep Agents, LangGraph and LangChain for developers, plus LangSmith Agent Server and Sandboxes for running agents securely, in the cloud or inside a company's own systems. A sandbox is a walled-off space where an agent can work without touching anything else.
• Agent Improvement: tools to test and fix agents. Observability (seeing what an agent did and why) and Evaluation (scoring how well it did) sit here. So does LangSmith Engine, an agent that finds errors, writes code and prompt fixes, and watches live traffic for new problems.
• Agent Governance: central rules for how agents use models, tools and data. LangChain's gateways help teams manage model budgets, set fallbacks if a model fails, protect sensitive data and control who can use which tool.

LangSmith also offers fine-tuning. That means taking records of what an agent did and using them as training material to teach a smaller, specialized model.

This is the heart of the pitch. LangChain says companies should be free to choose the best model for each job, control their own data and systems, and keep improving their agents.

The contrast is with the big cloud providers and AI labs, which tend to offer an agent tool that works best with their own models. LangChain says LangSmith is not trying to lock customers into any one model, cloud or ecosystem.

That promise comes from its open-source roots. It is also a crowded field: other startups make similar pitches, and some are doing well too.

The catch: a promise of freedom is easy to make. The test is whether companies stay with a neutral tool when the model maker next door offers a tidy package.

LangChain did not share revenue in this announcement. A report last year, citing unnamed sources, put LangSmith's annual recurring revenue between $12 million and $16 million. LangChain did not confirm that figure.

LangSmith had a free start and a $39 a month plan for small teams at that time, plus custom plans for large organizations.

The company's open-source projects remain popular. At the time of last year's round, LangChain had about 118,000 stars on GitHub, the site where developers share code. A star works like a bookmark with a thumbs up.

That reach helps. A developer who learned LangChain for free on a weekend may later push a company to pay for LangSmith.

LangChain says it will share several new LangSmith capabilities on Friday, November 20, at a special release event. The announcement did not say what they are.

Last year's round came with big updates, including version 1.0 of LangChain and LangGraph, so a launch right after new funding is the company's pattern.

You can see the company's earlier plans in its Series B post, where it first called its field "agent engineering."

• If you build software: expect more tools that watch, test and control agents, not just build them. LangSmith is one option among several.
• If you work at a company: the AI helper you meet at work may soon run on a platform like this, with rules about what data it can touch.
• If you follow the industry: investors still pay a lot for "picks and shovels" tools, the products that help others build with AI.
• If you worry about lock-in: ask whether your agents can switch to another AI model later. That is the freedom LangChain is selling.