Designing & Deploying Distributed Multimodal Multi-Agent Systems with Google's AI Stac

A technical talk and live demonstration on designing and deploying distributed multi-agent systems using Google’s AI stack. Demonstrated a live system where one orchestrator on Gemini Enterprise Agent Platform Runtime (formerly Vertex AI Agent Engine) coordinates 5 specialist agents on Cloud Run to generate complete social media campaigns from a single prompt, showing how agents collaborate, delegate tasks, and scale beyond monolithic LLM applications.

This work was recently accepted and published as an official Codelab 𝗚𝗼𝗼𝗴𝗹𝗲’𝘀 𝗼𝗳𝗳𝗶𝗰𝗶𝗮𝗹 𝗱𝗲𝘃𝗲𝗹𝗼𝗽𝗲𝗿 𝗰𝗼𝗱𝗲𝗹𝗮𝗯𝘀 𝗽𝗹𝗮𝘁𝗳𝗼𝗿𝗺 ( my own work ).

• Practical introduction to Google ADK
• Package reusable agent knowledge with ADK Skills (SkillToolset) to keep prompts lean
• Bridge a text agent to an image generation model via a custom async tool
• Agent-to-Agent (A2A) Protocol for distributed agent communication
• Model Context Protocol (MCP) for integrating external tools like Notion
• Deployment strategies for Gemini Enterprise Agent Platform Runtime (formerly Vertex AI Agent Engine) and Cloud Run
• Architecture patterns, tooling, and production considerations on Google Cloud
• Intuition to avoid the most common pitfalls in distributed agent system design: context passing, output contracts, retry handling, and context compaction