Tom Hundley
🤖 Ghostwritten by Claude Opus 4.6 · Fact-checked & edited by GPT 5.4
Harrison Chase co-founded LangChain and released its first version as an 800-line Python package on October 24, 2022, weeks before ChatGPT launched. By mid-2023, LangChain had become the fastest-growing open-source project on GitHub. By October 2025, the company Chase built with co-founder Ankush Gola had raised a $125 million Series B at a $1.25 billion valuation, with $260 million in total funding, roughly 1,000 customers, and about $16 million in revenue.
What explains that rise is not just timing. Chase became one of the clearest advocates for a specific view of agent systems: reliable behavior depends less on model choice alone and more on the quality of the context, tools, memory, and orchestration around the model. He calls that idea context engineering, and it now anchors a three-product platform spanning LangChain, LangGraph, and LangSmith.
For leaders evaluating agent infrastructure, Chase's trajectory is a useful case study in how an open-source abstraction can evolve into a broader platform when the underlying developer need is real.
TL;DR: Chase's work at Kensho Technologies and Robust Intelligence gave him direct exposure to the production problems that LangChain would later address.
Chase graduated from Harvard University in 2017 with a BA in Statistics and Computer Science. He then joined Kensho Technologies, where he worked from 2017 to 2020 as an ML Engineer and Entity Linking Team Lead.
That background matters. Entity linking requires systems to resolve ambiguous references against structured knowledge, which makes context, retrieval, and inference quality central to performance. In retrospect, it foreshadowed the same design concerns that would later define agent infrastructure.
From 2020 to 2022, Chase served as ML Team Lead at Robust Intelligence. LangChain was initially built during that period. The sequence helps explain why LangChain resonated so quickly: it was not a framework invented in the abstract, but one shaped by real production bottlenecks in applied machine learning.
LangChain's first release arrived on October 24, 2022, before the broader surge of developer interest that followed ChatGPT's launch. When demand for LLM tooling accelerated, LangChain was already positioned as a practical way to connect models, prompts, tools, and external data.
That head start helped it capture mindshare early, but the larger advantage was conceptual. LangChain gave developers a vocabulary and set of abstractions for building LLM-powered applications before those patterns had fully standardized.
TL;DR: LangChain expanded from a single open-source framework into a broader stack for building, orchestrating, and evaluating AI agents.
By mid-2026, the company spans three core products:
| Product | Function | Milestone |
|---|---|---|
| LangChain | Open-source framework for building LLM-powered applications and AI agents | 1.0 release, October 2025 |
| LangGraph | Stateful agent-orchestration runtime | Platform GA, May 2025; 1.0 release, October 2025 |
| LangSmith | Observability and evaluation platform | GA with paid plans, July 2024 |
The product logic is straightforward.
LangChain established the open-source foundation. LangGraph addressed the harder runtime problem of stateful, multi-step agent execution. LangSmith added tracing, evaluation, and monitoring so teams could understand whether those systems were actually working in production.
In late 2025, Chase also introduced Deep Agents, described as a general-purpose harness with planning, subagents, filesystem, and memory. That addition signaled a shift from offering flexible primitives to promoting more opinionated patterns for long-running agent systems.
The company's funding history tracks the market's growing conviction that agent infrastructure is a category of its own:
Per the curated profile, the Series B came with about $260 million in total funding raised, roughly 1,000 customers, and approximately $16 million in revenue. Those figures suggest investors were backing category leadership and ecosystem position as much as near-term financial scale.
TL;DR: Chase's central argument is that agent reliability depends on the quality of context and infrastructure around the model, not on model upgrades alone.
Chase is closely associated with the idea of context engineering. In his framing, many agent failures are not failures of raw model intelligence. They are failures of system design: the wrong instructions, the wrong retrieved information, the wrong memory, or the wrong tool state reached the model at the wrong time.
He has summarized that view directly: "Better models alone won't get your AI agent to production — infrastructure and context engineering matter just as much." He has also put the point even more sharply: "Context engineering is the real moat: agent failures come from wrong context, agent success comes from right context."
That philosophy has several practical implications:
Chase has also popularized the related term harness engineering, which focuses on the scaffolding around the model rather than the model in isolation. Together, those ideas have influenced how many developers now think about agent architecture.
TL;DR: Chase argues that 2026 marks the move from short-lived task agents to systems that plan, remember, and adapt over longer periods.
Chase has declared that "2026 is the first year of long-horizon agents — autonomous systems that plan, remember, and adapt across time."
That framing points to a meaningful change in what teams expect from agent systems. Earlier production deployments often focused on bounded tasks such as answering questions, processing documents, or completing a narrow workflow. Long-horizon agents aim to persist across longer timeframes, maintain memory, revise plans, and coordinate multiple steps or subagents without collapsing under context drift.
Deep Agents fits squarely into that thesis. A harness that combines planning, subagents, filesystem access, and memory is designed for agents that need continuity, not just one-shot execution.
For technical leaders, the implication is clear: as agents become more persistent, infrastructure requirements become more demanding. State management, evaluation, governance, and cost control all become more important when an agent operates across extended time horizons.
Harrison Chase is the Co-Founder and CEO of LangChain. He released the first version of LangChain as an 800-line Python package on October 24, 2022, and later helped build the company with Ankush Gola into a $1.25 billion business spanning LangChain, LangGraph, and LangSmith.
Before LangChain, Chase earned a BA in Statistics and Computer Science from Harvard in 2017, worked at Kensho Technologies from 2017 to 2020 as an ML Engineer and Entity Linking Team Lead, and then served as ML Team Lead at Robust Intelligence from 2020 to 2022. LangChain was initially built during his time at Robust Intelligence.
Context engineering is Chase's term for the discipline of making sure an agent sees the right information, instructions, memory, and tool state at the right time. The idea is that many agent failures come from poor context design rather than from the model itself.
LangChain is the open-source framework for building LLM-powered applications and agents. LangGraph is the stateful orchestration runtime designed for more complex agent workflows that need persistence, branching, and multi-step execution. LangGraph Platform reached GA in May 2025, and both LangChain and LangGraph reached 1.0 in October 2025.
Long-horizon agents are systems that plan, remember, and adapt over time rather than completing a single short task. In Chase's framing, 2026 is the first year these systems become a serious design target, which raises the importance of memory, orchestration, evaluation, and governance.
Harrison Chase's significance is not just that he launched an early open-source framework at the right moment. It is that he helped define the infrastructure vocabulary for the agent era. LangChain gave developers a starting point, LangGraph addressed orchestration, and LangSmith addressed evaluation and observability.
The broader lesson is that agent systems do not become reliable through model choice alone. They become reliable through better context, better runtime design, and better feedback loops. Chase's influence comes from making that argument early, clearly, and at the scale of a platform the market adopted.
Discover more content:
