Real-Time Telemetry in Development: Using MCP and Datadog for AI-Assisted Coding

Whitney demonstrates an innovative approach to AI-assisted development by integrating real-time telemetry data through the Datadog MCP server with Claude Code. The session showcases Commit Story, a demo application that automatically generates journal entries from Git commits and Claude Code conversations. The key innovation is using OpenTelemetry instrumentation throughout development (not just production) to give AI agents accurate, real-world data about system behavior. Whitney shares compelling debugging scenarios where telemetry data helped identify root causes when Claude Code made incorrect assumptions - like comparing data flow from three weeks ago to current behavior to discover a bypassed filter function. The demo includes viewing a heavily instrumented helper function where instrumentation code significantly outweighs business logic, highlighting the need for auto-instrumentation tooling. Whitney is building an auto-instrumentation agent that uses OpenTelemetry Weaver for semantic conventions, automatically adds instrumentation to code changes, ensures proper correlation between logs/metrics/traces, and validates instrumentation by checking data appears in Datadog. The conversation explores how telemetry helps maintain clean, efficient codebases by preventing AI agents from duplicating existing functionality, and how having complete visibility into system behavior enables more informed development decisions about data handling, filtering strategies, and resource allocation.