Orchestrating Multiple Agents in n8n with Google ADK: A Practical Walkthrough

In our recent Google ADK Node inside n8n video, we showed how it can power real-world agentic workflows.

This post goes one step deeper.

Instead of focusing on a single agent performing a single task, we explore how multiple specialized agents can collaborate, reason together, and execute a complete workflow end to end with minimal human coordination.

The example we'll walk through is a weekly marketing content workflow, designed for teams that produce newsletters or recurring communications.

Why This Workflow Exists

Many teams face the same recurring challenge:

• Weekly content needs to be created
• Topics change, but the structure stays the same
• Research, writing, review, and distribution are handled manually
• Coordination overhead often outweighs the actual writing effort

Rather than relying on one large, monolithic agent, this workflow demonstrates a different approach:

agent orchestration, where each agent has a clearly defined responsibility.

Starting Point: A Scheduled Trigger

The workflow begins with a weekly scheduled trigger.

At runtime, the trigger checks an n8n data table to identify the topic assigned for the current week. This ensures the workflow always starts with the correct context and avoids duplication of previously used topics.

This step sets the foundation: no prompts, no manual kickoff, just structured, predictable execution.

Research Agent: Gathering Context

Once the topic is identified, a Research Agent takes over.

Using Google Search through the Google ADK Node, this agent gathers relevant context, references, and insights related to the topic. Its role is not to write, only to research and enrich the input.

The output from this agent is passed forward as structured context for downstream agents.

This separation keeps responsibilities clear and reduces reasoning overload.

Initial Writer with Tool Calling

Next, the Initial Writer Agent generates the first draft.

What makes this step interesting is tool calling.

The writer doesn't rely only on static prompts. Instead, it connects to a Google Sheet through a provider tool, retrieves live data (such as current offers or promotions), formats that information, and blends it into the drafted content alongside the research output.

At this stage, the workflow already demonstrates a key ADK strength: agents that can reason and interact with live systems.

Critic and Refine: Loop-Based Reasoning

Draft content rarely meets quality standards on the first attempt.

Instead of sending the draft to a human reviewer, the workflow introduces a Loop Agent composed of two sub-agents:

• A Critic Agent, which reviews the content for clarity, tone, and structure
• A Refine Agent, which applies improvements based on the critique

These agents run in a loop, passing the content back and forth until predefined quality conditions are met.

This is not a one-off review. It's iterative reasoning, executed autonomously.

Final Step: Email Organization

Once the content is approved, it moves to the Email Organizer.

This agent:

• Draft a professional email directly in Gmail
• Inserts the finalized content
• Marks the topic as "used" in the data table to prevent reuse

At this point, the workflow has completed the full lifecycle:

What This Workflow Demonstrates

This example highlights three important Google ADK capabilities inside n8n:

1. Sub-Agent Nodes
   Breaking responsibilities into focused agents instead of one general-purpose agent.
2. Tool Calling
   Allowing agents to fetch and work with live data from external systems.
3. Loop-Based Reasoning
   Enabling agents to evaluate and improve their own outputs iteratively.

Rather than scaling intelligence vertically, this approach scales it horizontally, through orchestration.

Why Orchestration Matters

The real shift here is not content generation.

It's architectural.

Instead of asking one agent to "do everything," we design systems where:

• Each agent has a clear role
• Reasoning is distributed
• Quality emerges through collaboration

This is where agentic systems become reliable, understandable, and scalable.

What's Next

If you're building agentic systems in n8n, orchestration is where things start to get interesting.

This workflow is only one example. In future deep dives, we'll explore:

• More complex agent collaboration patterns
• Conditional agent routing
• Governance and oversight for agent-driven workflows

Because building agents is easy. Designing agent systems is where the real work begins.

If your team is looking to implement production-ready agent orchestration using the Google Agent Development Kit inside n8n, explore our Google ADK development services to see how we help organizations design, deploy, and scale secure multi-agent systems with confidence.