Building Agents on Google¶

Part of: Build Workflows > Agents

Google offers three approaches to building agents on the Gemini platform, ranging from natural-language automation to full programmatic control.

Workspace Studio¶

Workspace Studio is Google's natural-language automation platform built into Workspace. It lets anyone create automated workflows across Gmail, Drive, Sheets, and other Workspace apps by describing what they want in plain language — no coding or visual builder required.

Official docs: Workspace Studio

What you can build¶

• Automated workflows for everyday tasks — email triage, meeting summaries, cross-app data transfer, report generation
• Multi-step automations that connect Workspace apps with third-party services

How it works¶

1. Describe — tell Gemini what you want to automate in natural language
2. Generate — Gemini creates a workflow from your description
3. Refine — adjust the flow using pre-configured steps and connectors
4. Manage — run, monitor, and edit workflows directly within Workspace apps

Key capabilities¶

• Create automations using natural language — no code, no visual builder
• Pre-configured steps for common Workspace actions (send email, create doc, update sheet)
• Manage workflows in-app from Gmail, Chat, and Drive
• Third-party connectors for tools like Asana, Jira, Salesforce, and Mailchimp
• Included with Google Workspace Business and Enterprise plans

Mapping your Design blueprint¶

Example: Implementing your Design blueprint¶

Here's how a research analyst agent from the Design phase maps to Workspace Studio:

1. Name → Enter the workflow name (e.g., "Company Research Brief") when creating a new automation
2. Description → Describe what the automation does: "Researches a company and produces a structured brief in Google Docs"
3. Instructions → In the prompt field, describe the full workflow in natural language: "When I provide a company name, search for the company, find recent news and key products, then create a Google Doc with a structured brief containing: company overview, key products/services, recent news, and relevant insights."
4. Model → Gemini is built-in and not configurable — Workspace Studio uses the default Gemini model
5. Tools → Workspace Studio auto-connects Workspace apps. Add third-party connectors (Jira, Salesforce) from the connectors menu if your workflow needs them

Workspace Studio is best suited for the Prompt execution pattern — describe what you want, Gemini generates the automation, and you refine it conversationally.

Gemini Enterprise Agent Designer¶

Gemini Enterprise Agent Designer is a no-code, low-code platform within Gemini Enterprise for building AI agents using natural language and a visual workflow editor.

Official docs: Gemini Enterprise — Agents overview

What you can build¶

• Single-step agents — straightforward tasks with one action
• Multi-step agents — a primary agent coordinates with subagents on complex workflows

How it works¶

The interface has two main sections:

• Chat pane — describe what you want the agent to do in natural language, and refine it conversationally
• Designer pane — visually edit agent workflows on an interactive canvas, plus schedule and test agents

Key capabilities¶

• Create and preview agents using natural language prompts
• Visually edit agent workflows on an interactive canvas
• Connect to Google and third-party tools (Gmail, Drive, Jira)
• Schedule agent executions for recurring tasks
• No programming required for basic agent creation, with low-code options for more control

Mapping your Design blueprint¶

Example: Implementing your Design blueprint¶

Here's how a research analyst agent from the Design phase maps to Gemini Enterprise Agent Designer:

1. Name → Enter research-analyst as the agent name when creating a new agent
2. Description → Add the agent's purpose in the description field: "Researches companies and produces structured briefs"
3. Instructions → In the Chat pane, describe your agent: "You are a research analyst who investigates companies. Search for the company, visit key pages, and summarize findings with sections for company overview, key products/services, recent news, and relevant insights." Refine conversationally until the agent behavior matches your blueprint.
4. Model → Select the Gemini model version in the Designer pane settings (e.g., Gemini 2.0 Flash for speed, Gemini 2.5 Pro for complex reasoning)
5. Tools → In the Designer pane, connect tools — add Google Search for web research, Gmail for email actions, or Jira for issue tracking

For multi-agent workflows, create a primary agent and add subagents in the Designer pane. Each subagent gets its own name, instructions, and tools. The primary agent coordinates by delegating tasks to subagents based on your orchestration pattern.

Agent Development Kit (ADK)¶

The Agent Development Kit is an open-source framework for developing and deploying AI agents. It's designed to make agent development feel like software development — model-agnostic, deployment-agnostic, and compatible with other frameworks.

Official docs: Agent Development Kit overview | Gemini Enterprise Agents

Key features¶

• Open-source — flexible and extensible framework
• Model-agnostic — optimized for Gemini but works with other models
• Deployment-agnostic — deploy anywhere, with managed hosting on Vertex AI Agent Engine
• Framework-compatible — built for interoperability with other agent frameworks

What you can build¶

• Custom agents with programmatic control over behavior and orchestration
• Multi-agent architectures with specialized agents coordinating on complex tasks
• Production agents deployed to Vertex AI Agent Engine (fully managed Google Cloud runtime)

Tools and integrations¶

ADK agents can use:

• Built-in tools — core capabilities provided by the framework
• Google Cloud tools — access to Google Cloud services
• MCP tools — connect to external services via the Model Context Protocol
• Ecosystem tools — third-party integrations

Getting started paths¶

Mapping your Design blueprint¶

Example: Implementing your Design blueprint¶

Here's how a research analyst agent from the Design phase looks in the ADK:

from google.adk.agents import Agent
from google.adk.tools import google_search

research_agent = Agent(
    name="research_analyst",
    model="gemini-2.0-flash",
    instruction=(  # Note: ADK uses "instruction" (singular), not "instructions"
        "You are a research analyst who investigates companies "
        "and produces structured briefs. Search for the company, "
        "visit key pages, and summarize findings with sections for: "
        "company overview, key products/services, recent news, "
        "and relevant insights."
    ),
    tools=[google_search],
)

For multi-agent workflows, use the sub_agents parameter:

from google.adk.agents import Agent

writer_agent = Agent(
    name="brief_writer",
    model="gemini-2.0-flash",
    instruction="You write structured company briefs from research notes...",
)

research_agent = Agent(
    name="research_analyst",
    model="gemini-2.0-flash",
    instruction="You research companies. After gathering findings, delegate to the writer...",
    tools=[google_search],
    sub_agents=[writer_agent],
)

Official reference: ADK quickstart

What's Next¶

• Agents overview — the platform-agnostic agent decision framework
• Design Your AI Workflow — produce the agent blueprint that feeds into these implementations