AI agents with LangChain and LangGraph

Abstract

The Marie Vaucher’s workshop on building AI agents with LangChain and LangGraph. A great opportunity to share several months of hands-on experience developing our internal agents — and to highlight three realities we often overlook: - Agents don’t actually “understand” what they do, they execute linguistic patterns. This lack of real understanding makes them vulnerable. - Pre-configured ReAct agents are perfect for rapid prototyping, but their limited traceability quickly becomes an issue when moving toward production-grade systems.- An agent’s reliability depends far more on the architecture around the model than on the model itself, state management, task orchestration, output control, observability… the LLM is only one component in a much larger system.