AI agents are defined by their ability to perceive their environment, reason through decisions, and act autonomously to achieve goals.
Agents use short-term, long-term, and external knowledge memory to maintain context and coherence across interactions.
The agent loop is a continuous cycle of perceiving input, recalling memory, reasoning, acting, and storing new information.
The Agent Development Kit (ADK) is an open-source framework for building agents with deterministic control, built-in communication, and managed deployment.
ADK promotes using a tree hierarchy of specialized agents to simplify design, improve reliability, and increase modularity.
Use LLM agents for dynamic reasoning and workflow agents like sequential, loop, or parallel for deterministic, predictable execution paths.
ADK callbacks allow you to inject logic for logging, authentication, or state management at specific points in the agent execution cycle.
Choose sub-agents for tightly coupled workflows that share state, and agent tools for self-contained, function-like tasks with discrete inputs and outputs.
Moving an agent to production requires addressing five key gaps: containerization, security, networking, authentication, and monitoring.
Gemini Enterprise Agent Runtime is a managed service that automates containerization, scaling, security, and session management for ADK agents.
A FastAPI backend acts as a secure proxy between a React frontend and the deployed agent, managing authentication and streaming events.
The demo system uses an LLM orchestrator to route requests to sequential, loop, and parallel workflow agents for content generation.
The live demo showcases the full-stack application generating various content formats and analyzing text, with execution traces visible in the cloud console.
The complete source code, deployment scripts, and a step-by-step codelab are available to build and deploy the application yourself.
Jobs that call for the skills explored in this talk.