Unlock the power of AI Agents with Tools, Actions and Enterprise Data
The AI revolution is disrupting industries and creating new business opportunities and user experiences. This presents significant potential across sectors. In finance, AI enables personalized recommendations and capital market research. Retail benefits from conversational commerce and customer centricity. Healthcare can achieve improved efficiency, personalized and proactive care, and faster drug development.
AI Agent Evolution
Large language models (LLMs) and AI agents are central to the current wave of technological innovation, offering significant potential across various industries. Initially, users interacted with LLMs by providing prompts and receiving responses based on the vast datasets the models were trained on. Enterprises soon recognized the crucial role of their internal data in these interactions. This led to the first generation of agents, which grounded their responses in enterprise data through Retrieval Augmented Generation (RAG).
The next stage in agent development involved the ability to take actions on a user’s behalf using tools. The evolution continues towards more complex tasks that demand agents with a chain of thought and a reasoning loop. These advanced agents understand goals, break down tasks, plan execution strategies, and monitor their environment and user interactions to evaluate the outcomes of their actions and overall task completion.Today, we observe increasingly intricate deterministic flows involving multiple agent systems to fulfill end-user requests.
Figure 1 - AI agents evolution
Agent2Agent protocol (A2A)
The increasing complexity of tasks demands that AI agents collaborate within multi-agent systems. It is challenging to communicate between remote agents created in different frameworks across network boundaries and SaaS vendors. Consider the example of booking a trip: a central travel agent must seamlessly interact with specialized agents for flight reservations, hotel bookings, and travel policy. Recognizing this need for streamlined collaboration, Google introduced the Agent2Agent (A2A) protocol. This protocol facilitates the discovery of agents, ensures interoperability between them, and enables secure communication. Through A2A, AI agents can effectively exchange information and coordinate their actions across a diverse range of enterprise platforms and applications, unlocking new possibilities for complex problem-solving and task automation.
AI Agents and Tools
To accomplish user tasks, AI agents require interaction with a diverse range of both internal and external systems. This includes systems like ERP and CRM platforms, Google maps, cloud storage solutions, data warehouses such as BigQuery, and legacy systems like AS400. To facilitate these interactions and execute actions on behalf of users, AI agents utilize tools. These tools can manifest as local Python functions or API calls that expose underlying data and services.
With the rapid proliferation of AI agents and tools, the ability to share and reuse the tools across the enterprise is now more critical than ever. Google Cloud’s Apigee API management platform addresses this need with its API Hub, which serves as a centralized enterprise catalog for all API tools, regardless of their type (REST, gRPC, SSE, etc.). The API Hub provides developer code modules, enabling the exposure of APIs as LLM-friendly tools with a few lines of code.
The OpenAPI specification(OAS) for APIs hosted on Apigee are automatically published to the API Hub. Similarly, integration flows built with Google’s Application Integration product that enable access to data in the enterprise systems are published as APIs within the API Hub. You can also publish APIs running on third-party platforms to API Hub. It is recommended to expose data in data warehouses and reusable functions as serverless deployments and APIs to promote seamless sharing and reuse throughout the enterprise agent landscape.
Figure 2 - API Hub*:* API and Integration Tool Registry
Agent Development Kit (ADK) and Agent Actions with Tools
Google announced Agent Development Kit (ADK) at Google cloud NEXT 2025. ADK is a new open-source framework designed to simplify the full stack end-to-end development of agents and multi-agent systems. ADK is model agnostic, allowing the use of Gemini or any model in the Vertex AI model garden. It provides capabilities to easily create multi-agent systems, bi-directional streaming with audio and video, and implement workflows using workflow agents (Sequential, Parallel, Loop).
ADK enables the execution of agent actions with a diverse set of tools: pre-built tools (Search, Code Exec, API Hub Toolset, Application Integration Toolset), Model Context Protocol (MCP) tools, third-party libraries (LangChain, LlamaIndex), and even other agents as tools (LangGraph, CrewAI, etc.).
ADK features native integration with API Hub through the APIHubToolset module. Agents can invoke custom APIs deployed on Apigee, integration flows on Application Integration, Cloud Run and data services exposed as APIs, and third-party platform APIs using the APIHubToolset module. It also allows for the specification of the security scheme for tool invocation.
Example Usage:
custom_tool = APIHubToolset(
name="product_recommendations",
apihub_client=APIHubClient(sa_credential),
apihub_resource_name="projects/test-project/locations/region/apis/uniqueID"
)
Model Context Protocol (MCP)
The Model Context Protocol (MCP) allows Large Language Models (LLMs) to invoke tools with relevant context. It employs a client-server architecture where the AI agent acts as the MCP client, communicating with the MCP server. The MCP server then connects to various local and remote services, including APIs. Model Context Protocol (MCP) with the Agent Development Kit (ADK) enables the invocation of custom APIs deployed on Apigee and integration flows on Application Integration, facilitating access to data within enterprise systems like ERP, CRM, cloud storage, BigQuery data warehouses, and legacy systems such as AS400.
Figure 3 - ADK, MCP, Apigee, and Application Integration for Agent actions with enterprise systems
We are in the middle of an AI era with technology and industries evolving at an unprecedented pace. Now is the time to leverage the power of LLMs and AI agents, implementing agent actions through seamless interactions with enterprise systems via AI tools. The next horizon in AI agent evolution anticipates agents creating agents and tackling unforeseen challenges. Are you ready for this transformative next phase?
Call to Action:
- Explore Apigee API Hub: reusable AI tool registry for APIs and integration flows.
- Start building AI agents with the Google Agent Development Kit (ADK) and APIHubToolset module.
- Explore creating AI agents with MCP client and Server using the Agent Development Kit (ADK)
- Simplify multi-agent system creation and ensure enterprise security with the Agent2Agent (A2A) protocol.
- Discover and utilize AI agents across your organization with Google AgentSpace.
Talk to a Google Cloud Sales Specialist
Ready to unlock the full power of your AI agents?
You’ve read about the tools, actions, and enterprise data. Now, meet Google Cloud experts in person to see how to securely integrate, develop, and deploy AI agents at our free roadshow.
-
New York City: September 16
-
Chicago: September 18
-
Sunnyvale: October 8
Welcome to the Google Cloud Community!