Pinterest's Observability team is rethinking how they handle data monitoring and analysis, aiming to enhance both the integration of observability practices and the responsiveness of their production systems. They employ a dual strategy: "shift-left" focuses on embedding better logging and instrumentation in the development phase, while "shift-right" ensures robust monitoring in production. The team is integrating AI into their observability framework through the Model Context Protocol (MCP) and Agent2Agent (A2A) systems, creating a unified stream of data for intelligent agents to analyze. This approach aims to streamline root-cause analysis and facilitate actionable insights for engineers. Historically, Pinterest's observability infrastructure has been fragmented, with separate tools for logs, metrics, and traces. This siloed setup complicates the troubleshooting process, requiring engineers to navigate multiple interfaces, which can lead to inefficiencies and delays. The introduction of OpenTelemetry (OTel) and similar tools has set a new standard in the field, but the existing infrastructure makes it challenging for Pinterest to fully leverage these advancements. The Observability team is committed to bridging these gaps, using AI to automate insights and streamline problem-solving. A key component of this initiative is context engineering, which enhances the usefulness of AI agents by providing them with relevant data. The MCP, introduced by Anthropic, has become essential for this process, allowing agents to interact with various data types like metrics, logs, and alerts. This capability enables the agents to draw connections across disparate data sources, enhancing their ability to identify trends and offer solutions. By maintaining control over the data accessed by these agents, Pinterest ensures that the system remains secure and effective.