News

Google Cloud Unveils Vertex AI Agent Builder for GenAI Agent Development

• By John K. Waters
• April 11, 2024

Google Cloud this week introduced Vertex AI Agent Builder, a new tool that consolidates Vertex AI Search and Conversation products, enhancing the developer's toolbox for generative AI agent development.

Vertex AI Agent Builder offers developers a suite of tools to construct and deploy AI applications ready for enterprise use, catering to a wide spectrum of expertise levels — from those preferring no-code options to others who might utilize open-source frameworks like LangChain.

The new tool was designed to simplify the process of incorporating generative AI output into enterprise data. It includes Vertex AI Search for immediate grounding and retrieval augmented generation (RAG) APIs for document processing. It provides vector search capabilities for building embedding-based agents. And it comes with integrated security and enterprise controls—all of which serves to enhance model response precision and applicability.

Generative AI models offer remarkable capabilities for content creation and analysis, but actual AI applications require packaging these capabilities securely and usefully, said Burak Gokturk,
VP and GM in Google's Cloud AI & Industry Solutions group, in a blog post. AI agents take this further by integrating model capabilities with external systems for in-depth personalization and action execution on behalf of users. Vertex AI Agent Builder is designed to streamline these integrations, Gokturk said.

Developers can utilize Vertex AI Agent Builder’s no-code console to build and deploy generative AI agents quickly using Google’s advanced Gemini models. The Agents section in Vertex AI guides developers in creating new agents by setting goals, providing instructions, and giving conversational examples for the agent to emulate.

For more intricate objectives, developers can interlink multiple agents, enabling sophisticated workflow management. These agents can execute functions, tap into enterprise data for more factual responses, or connect with applications to perform user tasks.

Vertex AI Agent Builder also includes advanced tools for building, orchestrating, and maintaining agents, such as the ability to enhance prototype agents, monitor their performance in real time, and refine responses to specific queries through natural language training.

When it comes to grounding AI outputs in factually relevant information, Vertex AI Agent Builder presents an extensive set of capabilities, such as out-of-the-box RAG systems from Vertex AI Search. With a few clicks, developers can set up systems for document layout, ranking, and retrieval checks. More complex systems can also be designed using powerful vector search capabilities.

Google provides a unique offering by allowing grounding of model outputs in Google Search, merging the foundation models with up-to-date, quality data to significantly refine response accuracy and completeness.

Vertex AI Agent Builder further enables actions on the user’s behalf with extensions, function calls, and data connectors. These tools include pre-built modules for connecting models to specific APIs or tools, function calls that allow Gemini models to intelligently select and invoke the right APIs, and data connectors for ingesting data from enterprise and third-party applications.

Transitioning from prototype to production can pose technical hurdles; Vertex AI Agent Builder mitigates these with support for compliance and security standards, such as HIPAA and ISO 27000-series. It integrates with Vertex AI Studio for model tuning and connects with other Vertex AI offerings for cohesive workflows. This comes with the standard access controls, governance tools, and data sovereignty options that Google Cloud’s clientele expects, allowing innovators to focus on creation while maintaining data integrity and service reliability.