Modern enterprise AI demands more than brittle, monolithic prompt chains. This​ workshop delivers an engineering blueprint for designing, orchestrating, and scaling decentralized multi-agent systems using the Google Agent Development Kit.
Moving beyond basic single-agent prototypes, you will explore the full operational lifecycle of distributed intelligence. You will master standardizing tool integration via the Model Context Protocol, establishing seamless cross-framework communication through the Agent-to-Agent (A2A) standard, and coordinating complex multi-turn workflows with human-in-the-loop safeguards. The curriculum emphasizes vendor flexibility and production readiness, equipping you to package resilient agent networks with OpenTelemetry observability, strict security guardrails, and true hybrid, multi-cloud portability.

Objectives

• Master the architecture and runtime engine of Google Agent Development Kit (ADK 2.0).
• Architect decentralized, cross-framework multi-agent ecosystems using the Agent-to-Agent (A2A) protocol and Model Context Protocol (MCP).
• Implement complex orchestration topologies, task delegation frameworks, and persistent state management sessions.
• Integrate enterprise-grade security, guardrails, and Human-in-the-Loop (HITL) validation checkpoints into distributed workflows.
• Deploy, containerize, and monitor production-ready agent services across hybrid and multi-cloud environments.

Target Audience

AI Engineers, Enterprise Solutions Architects, and Senior Software Engineers tasked with deploying secure, production-grade autonomous agent systems across hybrid enterprise environments.

Prior experience with modern programming languages (Python or TypeScript) and foundational LLM concepts is recommended.

Prerequisites & Experience

Prior experience with modern programming languages (Python or TypeScript) and foundational LLM concepts is recommended.

Agenda

The Modern Agentic Landscape & ADK 2.0 Foundations

• Evolution of Agent Architectures: From Monolithic LLM Chains to Modular Multi-Agent Systems
• Introducing ADK 2.0: Core architecture, runtime engine, and new developer paradigms
• Model-agnostic design with native Gemini 3.x reasoning and structured outputs
• Local development, debugging, and visualization using ADK Dev CLI & Web UI

Tooling & Context Architecture – MCP Integration

• Equipping Agents with Actions: Native tools, custom toolsets, and lifecycle callbacks
• Model Context Protocol (MCP) in ADK: Standardizing agent-to-data and agent-to-tool connections
• Long-term memory, session state management, and persistent artifacts

Agent-to-Agent (A2A) Protocol Deep Dive

• A2A Standard: The open protocol for decentralized, cross-framework agent collaboration
• MCP vs. A2A: Dissecting Agent-to-Tool vs. Agent-to-Agent boundaries
• Agent Discovery & Capabilities: JSON Agent Cards 
• Exposing and consuming A2A endpoints: Converting local sub-agents into remote A2A services
• Message formatting, task delegation, structured negotiation, and asynchronous handoffs

Multi-Agent Orchestration & Enterprise Topology Patterns

• Multi-Agent Architectural Patterns
• Task API: Mixed delegation patterns, multi-turn task modes, and state reconciliation
• Cross-language & cross-cloud agent collaboration using RemoteA2aAgent
• Implementing Human-in-the-Loop (HITL) checkpoints in distributed agent workflows

“Deploy Anywhere” – Enterprise Production, Security & Governance

• Packaging and containerizing ADK agents for hybrid and multi-cloud environments
• Agent Governance: Guardrails, prompt injection mitigation
• Observability, distributed OpenTelemetry tracing, and A2A transaction logging