> Markdown version of [/videos/1918-building-and-deploying-multi-agent-systems-with-adk-and-vertex-ai](https://www.wearedevelopers.com/videos/1918-building-and-deploying-multi-agent-systems-with-adk-and-vertex-ai). Every page supports `.md` or `Accept: text/markdown`. Links point to the HTML versions so they work for humans too. Agent guide: [/agents.md](https://www.wearedevelopers.com/agents.md).

---

# Building and Deploying Multi-Agent Systems with ADK and Vertex AI

Monolithic prompts lead to fragile systems and context bloat. Discover how to build and deploy deterministic, debuggable multi-agent workflows using Google ADK and Vertex AI.

- **Speakers:** [Saoussen Chaabnia](https://www.wearedevelopers.com/@saoussen-chaabnia)
- **Event:** World Congress 2026 Europe - Virtual Stage
- **Published:** June 30, 2026
- **Duration:** 1:08:35
- **URL:** https://www.wearedevelopers.com/videos/1918-building-and-deploying-multi-agent-systems-with-adk-and-vertex-ai

## Summary

**From Monolithic Prompts to Specialized Agents** Transitioning from simple AI prompting to sophisticated autonomic workflows is reshaping enterprise operations, including automated customer support routing and internal HR content creation. However, monolithic prompts often result in context window bloat and fragile systems. This session outlines how to solve these bottlenecks using Google Agent Development Kit (ADK) to build highly maintainable multi-agent systems. By designing specialized agents targeting discrete functions, organizations can divide complex workloads into deterministic, easily debuggable pathways that mimic real-world team hierarchies. **Architecting Deterministic Workflows** The core architecture integrates dynamic reasoning with predictable structures. Using ADK, developers construct execution trees where large language models provide dynamic decision making, while dedicated workflow agents handle sequential or parallel orchestration. A crucial architectural decision lies in sub-agent design: internal team-like sub-agents utilize shared session states, whereas self-contained agent tools act as external contractors for standalone tasks. Furthermore, agent callback systems ensure that operational concerns like safety guardrails and authentication checks remain decoupled from core behavioral logic. **Securing Production Deployments** Translating local setups into production exposes immediate gaps in containerization, security compliance, and network telemetry. Deploying to the managed Gemini Enterprise Agent Runtime automates these infrastructure hurdles entirely. By combining this managed runtime with a FastAPI proxy layer, frontend client applications maintain a secure distance from direct LLM configuration calls. This backend proxy handles authentication, multi-turn session continuity, and streaming rate limits in a centralized location, illustrating how specialized AI orchestration seamlessly integrates into existing full-stack architectures.

**Keywords:** multi-agent system architecture, google adk deployment, gemini enterprise agent runtime, vertex AI engine infrastructure, deterministic workflow orchestration, fastAPI proxy layer integration, specialized llm sub-agents, ai session memory management, monolithic prompt mitigation, agent callback hook system, containerized ai backend

## Chapters

1. **Workshop session outcomes and companion resource repository** (00:02) — The core outcomes of the session and where to find the companion workshop code.
1. **Defining AI agents and their autonomous capabilities** (03:57) — How AI agents perceive their environment, reason through decisions, and execute actions.
1. **Agent components, memory types, and execution loops** (07:57) — The fundamental building blocks of an agent including short-term context and long-term memory.
1. **Core capabilities of the Google ADK** (12:43) — How ADK enables deterministic control, agent communication, and model-agnostic execution in Python.
1. **Building specialized multi-agent hierarchies instead of monolithic prompts** (17:29) — The advantages of dividing complex workflows into organized trees of specialized agents.
1. **Workflow agents versus LLM reasoning agents in ADK** (23:17) — Choosing between dynamic reasoning capabilities and deterministic coordination patterns like sequential execution.
1. **Implementing logic hooks with ADK execution callbacks** (28:17) — How to inject operational logic like logging and authentication without modifying core behavior.
1. **Design patterns for child sub-agents and agent tools** (31:25) — When to pass full session scope to a child agent versus invoking it as a discrete functional tool.
1. **Bridging the gap from local development to production deployments** (34:09) — Resolving containerization, security, networking, and observability gaps before deploying agents to the cloud.
1. **Deploying applications directly to Gemini Enterprise Agent Runtime** (38:25) — How the managed runtime automatically handles scaling, context management, and tracing for workflows.
1. **Packaging ADK workflows with Python deployment scripts** (44:59) — Writing initialization logic to push code artifacts into registries and managed cloud endpoints.
1. **Connecting frontends via a FastAPI proxy backend layer** (46:51) — Using remote runners and server-sent events to safely proxy UI requests into deployed agent endpoints.
1. **Architecting a multi-agent content creation pipeline application** (50:25) — Mapping sequential, looping, and parallel agent responsibilities into a master orchestrator application.
1. **Live demonstration of the deployed multi-agent content application** (55:07) — Executing the content generation flow and reviewing workflow traces within the runtime console.
1. **Examining ADK Python configurations and sample class setups** (65:11) — Final walkthrough of the repository structure and sample Python class setups for custom tools.

## Related Moments

- [Introduction to distributed multi-agent systems](https://www.wearedevelopers.com/videos/1976-designing-and-deploying-distributed-multimodal-multi-agent-systems-with-google-s-ai-stac) (from "Designing and Deploying Distributed Multimodal Multi-Agent Systems with Google's AI Stac")
- [Utilizing the ADK framework for agent logic](https://www.wearedevelopers.com/videos/1976-designing-and-deploying-distributed-multimodal-multi-agent-systems-with-google-s-ai-stac) (from "Designing and Deploying Distributed Multimodal Multi-Agent Systems with Google's AI Stac")
- [Deploying a web application through an AI agent workflow](https://www.wearedevelopers.com/videos/1539-agentic-devops-how-ai-powered-automation-transforms-software-delivery-on-github-and-azure) (from "Agentic DevOps: How AI-Powered Automation Transforms Software Delivery on GitHub and Azure")
- [Testing the completed multi-agent database workflow](https://www.wearedevelopers.com/videos/1727-build-a-multi-agent-role-playing-game-master-with-strands-agents) (from "Build a Multi-Agent Role-Playing Game Master with Strands Agents")
- [Architectural setup for the agentic AI live deployment demo](https://www.wearedevelopers.com/videos/100071-from-static-rules-to-reasoning-platforms-scaling-intelligent-canary-delivery-in-2026) (from "From Static Rules to Reasoning Platforms: Scaling Intelligent Canary Delivery in 2026")
- [Introduction to building real-world AI agent solutions](https://www.wearedevelopers.com/videos/1538-composable-intelligence-how-henkel-and-microsoft-are-shaping-the-agent-ecosystem) (from "Composable Intelligence: How Henkel and Microsoft Are Shaping the Agent Ecosystem")

## Related Articles

- [From Prototype to Production: Build AI Agents with This Free 4-Course Learning Path](https://www.wearedevelopers.com/magazine/655-from-prototype-to-production-build-ai-agents-with-this-free-4-course-learning-path)
- [Graph and AI Trends 2026: Why Is AI Running but Not Yet Delivering?](https://www.wearedevelopers.com/magazine/680-graph-and-ai-trends-2026-why-is-ai-running-but-not-yet-delivering)
- [What is Agentic Programming and Why Should Developers Care?](https://www.wearedevelopers.com/magazine/625-what-is-agentic-programming-and-why-should-developers-care)
- [Exploring AI: Opportunities and Risks for Developers](https://www.wearedevelopers.com/magazine/522-exploring-ai-opportunities-and-risks-for-developers)

## Related Jobs

- [Senior AI Agent Software Engineer (Go, Python) (m/f/x)](https://www.wearedevelopers.com/jobs/48277-senior-ai-agent-software-engineer-go-python-m-f-x) at **Dynatrace**
- [AI Software Engineer (Germany)](https://www.wearedevelopers.com/jobs/48317-ai-software-engineer-germany) at **Sunhat**
- [Principal Product Manager, Agent Platform](https://www.wearedevelopers.com/jobs/ext/277541-principal-product-manager-agent-platform) at **GitHub**
- [Principal Field Architect - AI Agents](https://www.wearedevelopers.com/jobs/ext/1442858-principal-field-architect-ai-agents) at **Twilio**
- [AI Full Stack Engineer](https://www.wearedevelopers.com/jobs/ext/1354435-ai-full-stack-engineer) at **Almedia**
- [Staff Software Engineer, Copilot Experiences](https://www.wearedevelopers.com/jobs/ext/164361-staff-software-engineer-copilot-experiences) at **GitHub**