May 20
Loading video player...
In this lecture, we explore how to instrument your ADK (Agent Development Kit) multi-agent systems with observability — the ability to infer what's happening inside your agents by monitoring external signals like logs and traces. We cover the out-of-the-box observability features ADK provides, how to configure them, and how to forward trace data to Google Cloud Trace for production-grade monitoring. ━━━━━━━━━━━━━━━━━━━━━━━━━ 🎓 FULL COURSE – Building Multi-Agent Systems: Google ADK Masterclass ━━━━━━━━━━━━━━━━━━━━━━━━━ This video is part of the complete Google ADK Masterclass on Udemy, where you'll learn everything from building your first agent to deploying production-ready multi-agent systems on Google Cloud. 👉 Enroll now (limited-time discount — first 30 days): https://www.udemy.com/course/building-multi-agent-systems-google-adk-masterclass/?couponCode=ADK_GENESIS What's covered in the full course: • Building single and multi-agent systems with Google ADK • Agent orchestration, delegation & subagent design patterns • Tool creation, callbacks & state management • Deploying agents to Google Cloud Run & Agent Engine • Observability: structured logging, tracing, AgentOps & Cloud Trace • Building production-ready UIs for our multiagent system • CI/CD pipelines for agent deployments • And much more! ━━━━━━━━━━━━━━━━━━━━━━━━━ ⏱ TIMESTAMPS ━━━━━━━━━━━━━━━━━━━━━━━━━ 0:03 – Introduction & what observability means 0:21 – Observability analogy: the driver's dashboard 1:09 – ADK agent observability: LLM calls, tool calls & callbacks 1:34 – ADK default observability features (structured logs & distributed tracing) 3:44 – Demo: Instrumenting the WealthPilot multi-agent app 4:10 – Viewing default INFO level logs with ADK Web 5:42 – Enabling DEBUG mode for verbose logging 8:05 – ADK Tracing: viewing the full agent call stack in ADK Web UI 10:07 – Deploying to Google Cloud Run with Cloud Trace enabled 12:03 – Verifying the Cloud Run deployment with trace env vars 13:10 – Running WealthPilot on production & checking traces 13:50 – Exploring trace data in Google Cloud Trace 16:15 – Configuring logging for custom runners (Python logging library) ━━━━━━━━━━━━━━━━━━━━━━━━━ 🔑 WHAT YOU'LL LEARN ━━━━━━━━━━━━━━━━━━━━━━━━━ ✅ What observability is and why it matters for AI agents ✅ ADK's default INFO-level structured logging ✅ Enabling DEBUG mode to see full LLM prompts, tool args & state transitions ✅ Using the ADK Web UI to visualize traces locally ✅ Deploying to Google Cloud Run with the --trace-to-cloud flag ✅ Viewing distributed traces in Google Cloud Trace ✅ Configuring the Python logging library for custom ADK runners ✅ Security implications of debug logging & ADK Web in production ━━━━━━━━━━━━━━━━━━━━━━━━━ 🔗 RESOURCES & LINKS ━━━━━━━━━━━━━━━━━━━━━━━━━ 📖 ADK Tools & Integrations: https://google.github.io/adk-docs/ ☁️ Google Cloud Trace: https://cloud.google.com/trace 🤖 AgentOps (covered in next lecture): https://www.agentops.ai/ ━━━━━━━━━━━━━━━━━━━━━━━━━ 📌 SERIES: Section 4 – ADK Observability ━━━━━━━━━━━━━━━━━━━━━━━━━ This is Lecture 1 of the ADK Observability section. In the next lecture, we'll integrate AgentOps for session replay, full trace visibility, and token cost tracking. ━━━━━━━━━━━━━━━━━━━━━━━━━ 🌐 CONNECT WITH ME ━━━━━━━━━━━━━━━━━━━━━━━━━ 🐦 X (Twitter): https://x.com/_thinking_reed 💻 Website: https://alonge.dev ADK #AgentDevelopmentKit #MultiAgentSystems #AIObservability #GoogleCloud #CloudTrace #LLM #AIAgents #Python #GoogleADK #UdemyCourse #AIEngineering