In the first part of a two-part series on AI agent observability, the guide demonstrates how to build a financial research assistant AI agent using tools like Node.js, OpenAI, Alpha Vantage, and Helicone. The financial assistant is designed to fetch real-time stock information and news while using Retrieval-Augmented Generation (RAG) to answer company-related queries, showcasing the complexities involved in such systems, especially the challenges of the black box problem in production environments. Key components of the AI agent include function-calling tools, basic Helicone monitoring for cost and latency tracking, and an agent loop for processing queries. Despite successful initial tests, the guide identifies potential issues like hallucinations and retrieval failures, emphasizing the need for proper observability tools like Helicone to address these challenges in production. The article highlights the importance of metrics such as latency, token usage, and error rates for optimizing AI agent performance, and previews how the second part will focus on comprehensive monitoring to resolve these issues effectively.