LangChain raises $300M at $3B: keep your agent traces portable

LangChain raised a $300M Series C at a $3B valuation. The company announced the round on November 13. CapitalG, an existing investor, led it. About a year ago LangChain was worth $1.25B. The money did not go to a model lab. It went to the layer around the model: the runtime, the tests and the controls. That layer now holds the most important record of how your agents work, and you must make sure you can take that record with you.

What actually happened

Harrison Chase published the Series C post on Friday, and TechCrunch reported the round the same day:

• The round. $300M at a $3B valuation. CapitalG led. GIC and DFJ Growth are new investors. IVP and Sapphire Ventures came back.
• The step up. TechCrunch reports the previous round was $125M at $1.25B, led by IVP, about a year ago.
• The customers. LangChain says more than 7,000 organizations use LangSmith. It names Harvey, Rippling, NVIDIA, Toyota and Salesforce. It says Clay monitors hundreds of millions of agent runs each month.
• The product. LangChain started in 2022 as an open-source library. It now sells LangSmith as a platform with three parts: an agent runtime, agent improvement (observability, evaluation, fine-tuning) and agent governance (an LLM gateway and a tool gateway).
• What is missing. Neither source gives revenue or pricing. LangChain says it will show new LangSmith features at an event on November 20.

LangChain also says Workday cut the time to deploy a new agent from five months to less than one day. That is a vendor claim from a customer story. Read it as such.

Why the LangChain Series C matters for your business

The round tells you where the lock-in moves. A model is easy to replace: you change one endpoint. The agent platform is not. It holds your traces, your evaluation datasets, your prompts and, with a gateway, each call to each model and tool. That is months of operating knowledge.

LangChain calls LangSmith "framework-agnostic," and the open-source libraries stay free. But the more parts of the stack you buy from one vendor, the more a move costs. We saw the same pattern when LangChain shipped managed deep agents. A vendor with $300M of new money will add more managed parts, not fewer.

What to do before you sign with any agent platform:

1. Ask for bulk export. Can you export all traces and evaluation datasets through an API, in a format you can read without the vendor?
2. Keep prompts in your repository. The platform can read them. Git must be the source.
3. Own the evaluation set. Your test cases are the asset. Keep a copy that runs without the platform.
4. Know the gateway cost. If all model calls go through a vendor gateway, find out what a failure or a price change does to you.

Sources: LangChain: Series C announcement, TechCrunch.