Last year, the focus was on moving LLMs into production. This year, the conversation has shifted to agents — autonomous, multi-step systems that make decisions, call tools, and act on their own. But with autonomy comes a new set of questions: how teams can determine whether an agent is behaving correctly, how complex reasoning chains can be debugged when something goes wrong, and how hallucinations, PII leaks, and prompt injections can be detected before they reach customers.

In this session, attendees will learn how the deep integration between Google's Agent Development Kit (ADK) and Datadog AI Observability provides end-to-end visibility into agents from prototype to production, with zero code changes.

The session covers building production-ready agents with Google’s Agent Development Kit (ADK), including the A2A protocol, multi-agent orchestration, and native tool use, along with automatic instrumentation using Datadog AI Observability to trace every agent decision, tool call, and LLM interaction without manual setup. It also includes built-in evaluations for hallucination detection, PII leak prevention, prompt injection defense, and LLM-as-a-judge custom evaluators, as well as the full lifecycle of production agents on Google Cloud from offline experiments to online monitoring.