Why Own Your Intelligence? Harrison Chase Keynote | Interrupt London
Summary
The core thesis is that for agents to succeed and become transformative, they must move beyond being simple LLM calls and instead be built within a robust, governed system—the 'harness.' LangChain emphasizes that owning intelligence requires making the agent apparatus (model, harness, context) open, compounding, and governed. The platform stack, LangSmith, provides comprehensive tools for building, deploying, and monitoring these complex, multi-tenant agents in the cloud, focusing heavily on agent trace observability and automated issue detection.
Key takeaways
-
The Importance of Systemic Ownership
2:00
True agent capability comes not just from the LLM, but from the entire system around it (the harness, skills, tools, context, and governance). To 'own your intelligence,' companies must specialize the system around their proprietary data and processes.
-
The Three Pillars of Agent Intelligence
5:40
Agents must be Open (model optionality, controllable harness, diverse context sources), Compounding (improving over time via observability/memory), and Governed (defining who the agent acts on behalf of and managing costs).
-
Automated Agent Improvement Cycle
29:40
The process of building agents is iterative (the flywheel). LangSmith Engine acts as a proactive agent engineer, monitoring traces, clustering issues, proposing fixes via a Pull Request (PR), and even running red teams to stress-test the agent.
Technical details
-
Agent Runtime and Harnessing
650s
The Deep Agents framework provides a solid base harness, including primitives for execution environments (file system, sandbox, code interpreter), context management (summarization, context offloading), and advanced features like planning and subagents. The future requires agents to run in the cloud, not locally, necessitating robust infrastructure.
-
Managed Deployment and Governance
950s
Managed Deep Agents bundles the harness layer and the managed infrastructure layer into one, simplifying the path to production. Key components include: LangSmith Deployments, a Tool Server (using MCP protocol), and an LLM Gateway (for reliability and cost control). Governance is critical, requiring tracking who the agent acts on behalf of and managing spending.
-
Agent Trace Observability (LangSmith)
1300s
Agent traces are considered a new, deeply nested data type. LangSmith uses SmithDB to store and analyze this data. Key features include: * **Ingest:** Using Decision Models (Jev) for cheap and efficient online evaluations. * **Store:** Supporting Bring Your Own Cloud (BYOC) and implementing granular Role-Based Access Control (RBAC) on inputs/outputs to protect sensitive data. * **Format:** LangSmith Trajectories provide a standardized, user-friendly format for viewing and annotating traces. * **Query:** AI Search/AI Query allows natural language querying over the semantics of the traces, finding issues that traditional logs miss.
-
Advanced Agent Capabilities
1850s
The platform supports advanced operational features: * **Self-Reporting:** Agents can call a self-report tool/endpoint while running to flag issues they encounter, providing immediate context for debugging. * **Red Teaming:** Proactively stress-testing the agent by sending new probes to find failure points. * **Customization:** LangSmith Custom Apps allow users to build governed, custom UIs (vibe-coded) for specific domain data views.
Mentioned resources
Channel & topics
Watch on YouTube · Back to latest
This independent, AI-assisted summary is provided for commentary and informational purposes. It may contain errors or omit important context. Please watch the original video for the creator's complete presentation. Video, thumbnail, and related copyrights belong to their respective owners.