A.I. Agent Architect Harrison Chase Wants Companies to Own Their Intelligence
Observer ·
Harrison Chase wants enterprises to control their A.I. agents. LangChain's latest software releases aim to make that easier.
As OpenAI and Anthropic expand their tools for building A.I. agents, LangChain cofounder and CEO Harrison Chase has a message for developers: own your intelligence.
That means controlling the models, context and systems surrounding an A.I. agent rather than depending entirely on a single model provider.
Chase made that argument at LangChain’s Interrupt conference in London on Oct. 13, nearly a year after the company raised $125 million at a $1.25 billion valuation . The funding helped establish LangChain as a major player in developer infrastructure. Now, Chase must show that its software remains useful even as the largest A.I. companies develop competing products.
“Our mission at LangChain is to enable every company to own their intelligence,” Chase said during his keynote presentation .
For enterprises, that means being able to build agents around their own information and expertise while retaining the flexibility to change underlying models.
Chase’s argument centers on what developers call the agent harness: the system that governs how an A.I. model uses tools, accesses information and carries out tasks.
An enterprise might use a powerful model supplied by OpenAI or Anthropic, but the company’s own data, business processes and specialized tools determine how useful that model becomes in practice.
Chase argues that organizations should maintain control over those components rather than surrender them to a model provider.
He is particularly interested in model optionality , which allows developers to choose between models based on performance, cost or suitability for a particular task.
That does not necessarily require using open-weight models. Chase also advocates keeping the harness open and controllable, explaining that open-source software can give developers greater flexibility.
The distinction has commercial implications for LangChain. If enterprises want the ability to switch model providers without rebuilding their applications, an independent software platform could help them do it.
Chase’s keynote was followed by an Oct. 15 roundup of LangChain’s latest releases , covering products introduced during the preceding weeks.
Among them was LangSmith AI Search, which allows developers to search records of agent activity using natural-language questions. Instead of relying solely on traditional filters, they can look for conversations in which a customer became frustrated or an agent exposed sensitive information.
LangSmith Engine v2 adds automated testing and tools designed to detect problems before agents are deployed. The update, first announced in September, also identifies inefficient behavior and tests proposed fixes.
Managed Deep Agents v0.9, released Oct. 7, introduced scheduled tasks, per-run configuration and Slack reactions. LangSmith Fine-Tuning, meanwhile, allows developers to use records of agent activity to train and evaluate models for specialized applications.
Together, these offerings extend LangChain’s business beyond the original open-source framework that helped popularize the company. They address problems that become more important once companies begin using agents in daily operations: monitoring performance, identifying mistakes and improving reliability.
They also expand the services LangChain can offer paying enterprise customers.
LangChain began as an open-source project in October 2022 before Chase and cofounder Ankush Gola established the company the following year.
Its products now include LangChain and LangGraph, which help developers build and coordinate agents, and LangSmith, a commercial platform for evaluating, deploying and monitoring them.
The company says it works with 35 percent of Fortune 500 companies and has surpassed 1 billion cumulative open-source downloads. Those figures are company-reported and do not establish how much revenue the individual customers generate.
Chase’s vision also reflects a broader concern among technology executives about who controls enterprise A.I. systems.
Microsoft CEO Satya Nadella has emphasized the importance of organizations owning the learning processes that improve their A.I. applications. Palantir CEO Alex Karp has similarly argued that enterprise customers want control over their computing infrastructure, models and data.
At Interrupt London, attendees could experiment with agent behavior at the LangChain Agent Arena, where they trained agents to play Tetris and competed on a leaderboard.
Outside the arena, the commercial stakes are considerable. OpenAI, Anthropic and other model developers are expanding the tools they provide for building agents, potentially reducing the need for outside software.
Chase’s answer is to make LangChain useful regardless of which model a customer chooses. Its newest products suggest how he intends to do that: by helping companies manage the systems they build around those models.