Check in, grab a coffee, and connect with fellow attendees.

- Discover the core mission and driving force behind Google Developer Groups.
- Learn how to plug into local tech communities, mentorship programs, and future events.
- Hear from local GDG chapters

Explore the latest perspectives on agentic AI and how organizations can move from AI strategy to practical implementation.

Let’s be real: basic prompt loops are dead. If your agent falls apart from context rot the second a task takes longer than four minutes, you're building toys. Real production work takes hours: human approvals take time, subagents run in parallel, and feedback arrives tomorrow. The golden rule of modern agent design is simple: an agent is defined by where its state lives, because its process has to be allowed to die.
In this hands-on lab, you’ll build and ship an autonomous channel-running agent built with Google’s ADK, Gemini on Vertex AI, BigQuery Property Graphs (GQL), and Vertex AI Memory Bank. You'll swap messy while loops for clean graph engineering, fan out parallel research nodes without thread pools, and park your runs at zero compute cost using durable SQLite sessions. Along the way, you’ll handle asynchronous human-in-the-loop approvals, run hybrid GraphRAG over real audience data, and give your agent a cross-run Memory Bank that consolidates what it learns.

Putting an LLM in the middle of a live, millisecond-scale auction loop is an architectural nightmare. Hand-crafted bidding rules fall apart the moment traffic patterns shift, while raw LLMs are too slow, expensive, and non-deterministic to call on every impression. Modern agentic data engineering flips the script, where the agent analyzes real-time telemetry, generates deterministic bidding logic, and continuously patches its own execution harness.
In this hands-on lab, you’ll build and deploy an autonomous AI Data Engineer using Google ADK, Gemini on Vertex AI, BigQuery streaming ingestion. You'll ditch manual SQL tuning and brittle scripts to equip your agent with declarative tools to inspect partitioned live auction telemetry and generate optimized Python bidding policies. Then, you’ll close the self-evolution loop, deploy in-run failure monitors that catch spend velocity anomalies, trigger automated harness self-patching, and validate every strategy update against long-horizon eval benches before safely merging code to production.

Continue the conversation with peers, Google Cloud experts, and partners over coffee and snacks.