Build Daily
Tinley Park · August 23, 2026Building Governed Agents: A Framework for Cost, Control and Compliance
This video introduces LangSmith's framework for building governed AI agents, focusing on managing cost, reliability, and compliance through the agent development lifecycle. It outlines key governance areas such as LLM calls, tool usage, MCP interactions, and agent-to-agent communication to ensure secure and scalable deployment.
The outline
Tap a timestamp to jump there in the video.
Introduction to LangSmith
The video introduces LangSmith as an agnostic platform for the agent development lifecycle, covering build, test, deploy, and monitor stages.
Company Context & Trust
LangChain highlights its widespread adoption by startups and enterprises across regulated industries to establish trust in their agent engineering tools.
Why Governance Matters
Production agents require centralized runtime controls to manage the unique risks of autonomy, unpredictable costs, reliability needs, and compliance requirements.
Five Areas of Governance
Governance is structured into five areas: governing actions/identity, model selection/failover, boundary enforcement, observation/measurement, and assurance/evidence.
Integrated vs. Bolted-on Governance
The speaker emphasizes that governance must be integrated into the core agent infrastructure rather than bolted on to avoid missing critical pathways for fixes and deployment.
External Interaction Risks
As agents interact with external LLMs, tools, MCP servers, and other agents, robust authorization is required to prevent unauthorized actions at scale.
LLM Call Governance
Managing LLM calls involves controlling costs through spend limits and routing, ensuring availability via fallbacks, and protecting private data with redaction.
Tool & MCP Call Controls
Governance for tool and MCP calls focuses on strict permissioning, access control, and audit trails to prevent unintended or unauthorized actions in production systems.
Agent-to-Agent Identity
Managing agent-to-agent communication requires defining clear identities and permissions to prevent compounding errors and unauthorized access across agent chains.
Enforcement & Scaling Challenges
While basic proxies are easy to build, maintaining robust governance controls and integrations becomes significantly harder as organizations scale beyond a few dozen people.
The agent development lifecycle matters for anyone shipping AI products, not just large organizations. Solo builders need the same controls around cost, reliability, and compliance to ship agents that hold up in production.