Google Agent Development Kit (ADK)¶

What it is¶

The Google Agent Development Kit (ADK) is an open-source framework designed for building, debugging, and deploying enterprise-grade AI agents at scale. As of June 2026, the ADK is in General Availability (GA), serving as the unified orchestration layer for Google's agentic ecosystem, including Gemini 3.5 and Vertex AI.

What problem it solves¶

It addresses the "Prototype-to-Production" gap. While many frameworks excel at simple chat loops, the ADK provides the rigorous state-machine orchestration, standardized "Skills" (tool-calling), and robust evaluation frameworks required for mission-critical, multi-agent business workflows.

Where it fits in the stack¶

Category: Frameworks / Enterprise Agent Framework. It sits between the frontier models (Gemini, Claude) and the application layer, providing a structured runtime for agentic logic.

Typical use cases¶

Multi-Agent Enterprise Orchestration: Coordinating complex, multi-step tasks across specialized agents (e.g., an "Accountant Agent" and a "Legal Agent").
Mission-Critical Workflows: Building agents where execution paths must be predictable, auditable, and reliable.
Skill-Based Capability Extension: Rapidly adding new tools to agents using standardized, cross-language Skill definitions.
Vertex AI Deployment: Scaling agents to production using Google Cloud's native infrastructure (Cloud Run, GKE).

Strengths¶

Production-Grade Reliability: Based on the same infrastructure Google uses for internal AI services.
Multi-Language Native: Full, first-class support for Python, TypeScript, Go, and Java.
General Availability (GA): (June 2026) Fully supported with enterprise SLAs and comprehensive documentation.
Native MCP 3.0 Integration: Seamlessly connects to any Model Context Protocol server for tool and data access.
Standardized "Skills": A robust pattern for defining and discovering agent capabilities across different projects.

Limitations¶

Architectural Complexity: The emphasis on state-machines and explicit orchestration may feel overly complex for simple, linear agents.
Cloud-Centric Optimization: While open-source, it is heavily optimized for the Google Cloud/Vertex AI ecosystem.

When to use it¶

When building large-scale, multi-agent systems that require high reliability and clear state management.
For enterprise projects where cross-language compatibility (e.g., a Go orchestrator with Python skills) is a requirement.
When you are already utilizing the Vertex AI ecosystem and want native integration with its evaluation and monitoring tools.

When not to use it¶

For quick, experimental prototypes where a lighter framework like smolagents or instructor would be faster.
If you prefer a purely "vibe-based" or non-deterministic agent loop over structured state-machines.

Getting started¶

The ADK is available via standard package managers.

Python Installation¶

pip install google-adk

TypeScript Installation¶

npm install @google/adk

CLI examples¶

1. Initialize a New Project¶

adk init my-agentic-service --language python

2. Run Local Debugger¶

adk debug agent.yaml

3. Deploy to Vertex AI¶

adk deploy --project my-gcp-project --region us-central1

API examples¶

Defining a Standardized Skill (Python)¶

The ADK uses decorators to transform functions into discoverable agent skills.

from google_adk import Skill

@Skill.define(
    name="get_stock_price",
    description="Fetches real-time stock pricing from the internal financial API."
)
def fetch_stock(ticker: str) -> float:
    # Logic to query internal API
    return 150.25

# Registering the skill with an agent
from google_adk import Agent
agent = Agent(name="FinanceAgent", model="gemini-1.5-pro", skills=[fetch_stock])

Orchestrating a State-Machine Agent (TypeScript)¶

import { Agent, StateMachine } from '@google/adk';

const sm = new StateMachine();
sm.addState('INITIAL', async (context) => {
  return context.input.includes('help') ? 'SUPPORT' : 'ROUTING';
});

const agent = new Agent({
  name: 'Orchestrator',
  workflow: sm
});

Gemini — The primary model backbone for ADK agents.
LangGraph — Competitive framework for stateful multi-agent systems.
CrewAI — Popular alternative for multi-agent role-playing.
MCP — Integrated standard for tool connectivity.
Vertex AI — Google Cloud's AI platform for deployment.
Jules — Advanced agentic assistant developed by Google.
Firebase Genkit — Google's framework for AI-integrated web apps.
Agentic Workflows — The fundamental pattern the ADK implements.

Sources / references¶

Contribution Metadata¶

Last reviewed: 2026-06-23
Confidence: high