LangChain, a Side Project Turned $3 Billion Startup in Four Years, Raises $300 Million
Harrison Chase posted the project as a hobby in October 2022. On November 13, LangChain closed a $300 million Series C led by CapitalG, valuing the company at $3 billion. A valuation is what investors think the company is worth, not its revenue. The company makes its money from LangSmith, a "control tower" for AI agents that shows what they are doing and lets teams restrict their access to models and data. For teams running agents in production, this is a signal that controlling agents is becoming a separate line item in the budget.
AI-processed from X: @pukerrainbrow; edited by Hamidun NewsIn October 2022, engineer Harrison Chase released the LangChain library as an open-source side project. Four years later, investors value the company at $3 billion. In this context, "valuation" means the price of the whole company based on the latest funding round, not its revenue. On November 13, LangChain announced a Series C: $300 million in new money led by the CapitalG fund. Customers pay not for the library but for LangSmith, a platform for running AI agents, watching what they do, and limiting what they can access. More than 7,000 organizations already use it. If you are building agents, this story shows the stage at which the market is ready to pay.
How the Hobby Project Grew into a Company
Chase graduated from Harvard in 2017, then worked as a machine learning engineer at Kensho Technologies and Robust Intelligence. There he met Ankush Gola, who earned an electrical engineering degree from Princeton in 2015 and spent four years as a senior engineer at Facebook. A month after LangChain's release, ChatGPT appeared, and the first wave of developers who wanted to build applications around language models came to the project, among others. As interest grew, Chase and Gola incorporated the project as a company in 2023.
Funding moved fast. In April 2023, Benchmark led a $10 million seed round, and then Sequoia led a $25 million Series A. In October 2025, IVP led a $125 million Series B at a $1.25 billion valuation, and now the valuation has risen to $3 billion. Participants in the Series C included GIC, DFJ Growth, IVP, Sapphire Ventures, Salesforce Ventures, Insight Venture Partners, QuantumLight, and MongoDB Ventures. Sequoia, Amplify, and Benchmark invested again. In total, the company has raised $460 million.
LangChain's open-source projects have been downloaded more than 1 billion times, and they now see more than 350 million downloads a month. That developer base works as a free sales channel for the paid product.
What Customers Pay For: LangSmith
You can think of LangSmith as a control tower for agents. It works with any framework and any model, not only with LangChain itself. Agents can run in LangChain's cloud or in the customer's own AWS, Azure, or Google Cloud account. The platform tracks and evaluates agent behavior, and it also controls which models, tools, and data agents can use.
A separate new feature is LangSmith Engine. It is an agent that improves other agents: it finds errors in production and writes a fix, whether code or a prompt. In effect, it is an on-call engineer who reads logs and proposes a change.
The customer figures are as follows: more than 7,000 organizations, including Harvey, Rippling, NVIDIA, Toyota, and Salesforce. Clay tracks hundreds of millions of agent runs per month on the platform. Workday cut the time to release a new agent from five months to under a day. The company itself provides these data, and no independent verification exists yet. But they show that the bottleneck has shifted: building an agent has become easier, while making sure it behaves predictably is still hard.
What Developers and Product Managers Should Take Away
Chase told The Wall Street Journal: "Agent behavior is the hardest part of control." An agent may perform well in a demo and behave strangely on real data. Noticing this without tracing, quality evaluation, and access limits is nearly impossible.
Practical takeaways:
- build in observability from day one: logs of agent steps, outcome evaluation, and tracking of request costs;
- restrict agents' permissions before launch, not after the first incident: which tools and data they can access;
- avoid tying yourself to a single framework: LangSmith's value is that it is independent of your choice of model and library;
- compare solutions on your own stack, not on vendor marketing case studies.
There are caveats too. The $3 billion valuation assumes continued growth in the agent market, which is still mostly enterprise and has not yet settled. Competitors, from cloud providers to other observability tools, are also building similar dashboards. And a self-improving agent like Engine needs at least as much testing as any other: an AI-written fix still requires human review.
LangChain currently has 435 employees, and by the end of 2027 the company plans to reach 850, nearly doubling its headcount. This is a bet that companies will deploy agents at scale and pay for the infrastructure that keeps them under control.
- Nov 22, 2026Android work profile closes real-IP leak that Happ and v2rayNG don't catch
- Nov 22, 2026Anthropic Goes Public on Nasdaq: Shares Rise from $83 to $111, but Can Safety Hold Up?
- Nov 22, 2026Databricks Is in Talks for a Round at a $225 Billion Valuation, Nearly 20% Above August
- Nov 22, 2026Nvidia Stops Selling H200 in China and Rejects Musk's Push to Bring Back Huang
Key Questions Answered
How the Hobby Project Grew into a Company?
Chase graduated from Harvard in 2017, then worked as a machine learning engineer at Kensho Technologies and Robust Intelligence. There he met Ankush Gola, who earned an electrical engineering degree from Princeton in 2015 and spent four years as a senior engineer at Facebook. A month after LangChain's release, ChatGPT appeared, and the first wave of developers who wanted to build applications around language models came to the project, among others.
What Customers Pay For: LangSmith?
You can think of LangSmith as a control tower for agents. It works with any framework and any model, not only with LangChain itself. Agents can run in LangChain's cloud or in the customer's own AWS, Azure, or Google Cloud account. The platform tracks and evaluates agent behavior, and it also controls which models, tools, and data agents can use. A separate new feature is LangSmith Engine. It is an agent that improves other agents: it finds errors in production and writes a fix, whether code or a prompt.
What Developers and Product Managers Should Take Away?
Chase told The Wall Street Journal: "Agent behavior is the hardest part of control." An agent may perform well in a demo and behave strangely on real data. Noticing this without tracing, quality evaluation, and access limits is nearly impossible.
Ship AI Faster with Founders, Engineers & Weekly Architecture Reviews
Inside Vibe Club (hamidun.com/club) — weekly live masterminds with Zhemal Khamidun, Tech Radar, full LMS access, and 370+ production Claude Code skills.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.