AI agents have become the latest craze in the world of generative AI. These advanced agents go beyond simple question-and-answer capabilities and can now take actions and interact with back-end systems in an automated manner.
During Google Cloud Next on Tuesday, the company unveiled a new tool designed to assist businesses in building their own AI agents.
“Vertex AI Agent Builder makes it incredibly simple and fast for individuals to create conversational agents,” stated Google Cloud CEO Thomas Kurian. “Users are able to construct and deploy production-ready agents powered by generative AI, and instruct and guide them just as they would a human, enhancing the accuracy and quality of responses from the models.”
This no-code product builds upon Google’s previously released Vertex AI Search and Conversation and utilizes their latest Gemini large language models. It makes use of both RAG APIs and vector search, two widely-used methods in the industry for reducing hallucinations – a phenomenon where models provide incorrect responses when unable to find the correct answer.
One of the techniques implemented by Google to improve the accuracy of responses is known as “grounding”. This involves tying the answers to reliable sources, and in this case, Google Search is the source (although its accuracy may vary).
“By implementing grounding with Google Search, we are bringing the vast knowledge of the world that Google provides directly to models,” explained Kurian. “Additionally, we also support the ability to ground against enterprise data sources, which may be more suitable for our enterprise clients.”
In a demo, Google showcased how this capability can be used to create an agent that analyzes a company’s previous marketing campaigns and understands its brand style. This knowledge is then applied to generate new ideas that align with the company’s established style. The demo utilized over 3,000 brand images, descriptions, videos, and documents from a fictional company’s products stored on Google Drive.
While the potential to build any type of agent is present, this particular example places Google in direct competition with Adobe, a company that released its own creative generative AI tools, Firefly and GenStudio, in the past year. These tools assist in building content that stays true to a company’s established style. The flexibility to create anything is available, but the question remains whether it makes more sense to purchase something off-the-shelf if it meets the required criteria.
According to Google, these new capabilities are already available. The product supports multiple languages and offers country-specific API endpoints in the United States and European Union.
