AI agents are the new craze in generative AI. Unlike previous generations of chatbots, these agents do more than just answer questions. You can also perform actions based on conversations or interact with backend transaction systems to perform actions in an automated manner.
At Google Cloud Next on Tuesday, the company introduced new tools to help enterprises build AI agents.
“Vertex AI Agent Builder makes it incredibly easy and fast to build conversational agents,” said Google Cloud CEO Thomas Kuran. “By building and deploying production-ready conversational agents powered by generative AI that can be directed and guided like a human, we can improve the quality and accuracy of answers from our models. Masu.”
This no-code product builds on Google's previously released Vertex AI Search and Conversation product. It is also built on the company's latest Gemini large-scale language model and relies on both the RAG API and vector search. These are two common methods used across the industry to reduce hallucinations, allowing models to compensate for incorrect answers when the correct response cannot be found.
One of the ways the company improves the quality of its answers is through a process called “grounding,” where answers are associated with what are considered reliable sources. In this case, you're relying on a Google search (which may or may not actually be accurate).
“We now offer grounding for Google Search, bringing the power of global knowledge that Google Search provides to models through our grounding service. We also support grounding capabilities for enterprise data sources. ” said Kurian. The latter may be more suitable for enterprise customers.
Image credit: Frederic Lardinois/TechCrunch
The company used this feature in a demo to create an agent that analyzes previous marketing campaigns to understand a company's branding style and applies that knowledge to generate new ideas that match that style. The demo analyzed his 3,000+ brand images, descriptions, videos, and documents related to this fictitious company's products stored in Google Drive. It then helped generate photos, captions, and other content based on an understanding of the fictional company's style.
You can build any kind of agent, but in this particular example, Google would be in direct competition with Adobe. Last year, Adobe released Firefly, a creative generation AI tool that helps build content that stays within the company's style, and last month released GenStudio. You have the flexibility to build anything, but the question is, if one exists, would you buy one off the shelf instead?
Google says the new feature is already available. Supports multiple languages and provides country-based API endpoints in the US and EU.