An introduction to Duet AI in Google Cloud

The video begins with Preston Holmes, a product manager at Google, welcoming the audience and introducing the focus of the presentation, which is Duet, a product under active development and innovation. He emphasizes the dynamic nature of AI products and highlights the distinction between Google's consumer-facing generative AI and enterprise/business solutions. The talk outlines the purpose of Vertex as a platform for value creation using generative AI and Duet as a service for consuming that value. Preston assures that what is presented is only the beginning and that the products are being developed on an ever-evolving foundation of innovation.

Preston discusses Duet AI's role in simplifying interactions with Google Cloud interfaces, such as the IDE for developers and specialized data interaction services. He explains that Duet AI is designed to make cloud services more accessible and intuitive, even for newcomers. The presentation also touches on the challenges faced by new employees in understanding and navigating the extensive range of Google Cloud services. Preston emphasizes the importance of the user experience in Duet AI and the aim to make it easy to use for non-AI experts.

The presentation delves into the specific features of Duet AI, including its ability to provide expert advice and guidance in real-time, its integration into platform interfaces, and its move towards a more agent-based system. Preston talks about the long-term goals for Duet AI, such as infusing the model with direct context awareness, enhancing productivity through code and data table interpretation, and supporting users throughout their entire journey with Google Cloud Platform (GCP) services.

Preston introduces the concept of assisted development with Duet AI, using the example of a developer named Sam. He explains how Duet AI can aid in code completion, suggestions, and understanding complex codebases through an interactive chat component in the IDE. The discussion includes the benefits of Duet AI in identifying and using GCP services more accurately due to specialized tuning and the release of new features such as writing unit tests. Preston also highlights the importance of trust and transparency in using generative AI for coding, with Duet AI providing visibility into the origins of its suggestions.

Vijay Krishnan from Turing shares his company's experience with Duet AI, discussing how Turing has assisted over a thousand companies, particularly large enterprises, in managed software projects and team building. He emphasizes the extensive developer pool and rigorous testing process at Turing, which allows for detailed developer profiles and effective machine learning algorithms. Vijay explains the impact of Duet AI on developer productivity, citing a 33% increase in pull request mergers and the potential for further gains through behavioral changes and training. He also discusses Turing's future plans to leverage Duet AI for code base indexing and debugging.

The presentation shifts to discussing how Duet AI enhances operations across GCP products. Preston explains that Duet AI can provide quick, in-context understanding of product functionalities and configurations. He outlines the benefits of Duet AI in summarizing documentation, offering curated experiences for recommendations and debugging, and assisting in understanding error texts and third-party products. The discussion also covers Duet AI's capabilities in data analysis, particularly with BigQuery, where it can provide SQL suggestions and translate natural language queries into SQL code.

Preston and Vijay address Duet AI's role in security and no-code experiences. They discuss how Duet AI can aid security professionals in being more productive and help newcomers contribute effectively. The presentation covers the use of Duet AI in summarizing threat intelligence reports, constructing search terms for security tools, and understanding vulnerabilities in plain English. The talk also touches on the potential of Duet AI in the no-code platform, AppSheet, where it can generate applications from a single statement and help refine existing applications.

The presentation concludes with a discussion on Duet AI's enterprise readiness, emphasizing its data segmentation and isolation principles. Preston outlines the AI principles followed in building Duet AI, including safety checks in responses. He also shares future plans for Duet AI, such as rich admin controls, integrated troubleshooting experiences, enhanced data exploration with machine learning, and improvements in natural language search creation. Preston encourages the audience to sign up for the Duet AI preview and provide feedback to aid in the ongoing development and improvement of the product.