AI agents
1. AI agents
SPEAKER: Let's proceed to the most recent AI application, AI agents. In previous lessons, you learned about foundation models like Gemini and development tools like Vertex AI Studio for building Gen AI applications. You might now be wondering how to further leverage these technologies to enable AI to not only chat and research but also take action. For instance, can AI be used to automate your workflows and make decisions on your behalf? This is where AI agents and Agentic AI takes center stage. Join us in this lesson to explore the evolution of Gen AI and then dive into AI agents, discovering how they take action for you. Think about the applications you've explored so far. Many are conversational in nature. You pose a question with a prompt, and the AI delivers an answer. Think back to the example with Bea and Ann. They asked Vertex AI Studio to generate an insurance analysis report from a set of instructions. And voila! The AI did a fantastic job in mere seconds, saving Ann tremendous time. Naturally, Bea and Ann wished the AI could automate the subsequent steps, like verification, decision-making, and policy structuring. However, these tasks require accessing internal documentation and interacting with various existing applications. This falls outside the scope of foundation models, which base their knowledge on pre-trained data that's often not field-specific and cannot connect to different applications. An AI agent solves this problem and adds real value to foundation models. AI agents connect to information and applications outside of foundation models, take action, and observe feedback from the environment to improve over time. Bea and Ann are amazed by the potential of AI agents. What if they desire a unified agent system that coordinates multiple agents for insurance, underwriting, and claims? This is termed agentic AI, which can be imagined as a more autonomous, complex reasoning and coordinating system for multi-step tasks involving multiple agents, surpassing the capabilities of a single AI agent. Gen AI is embarking on a fascinating journey, evolving from simple chatbots to sophisticated AI agents and, ultimately, to proactive Agentic AI. This progression promises to make AI increasingly practical, empowering us to tackle more complex real-world challenges. Given the evolution of Gen AI, let's zoom in on an AI agent and discover how it works. First and foremost, what exactly is an AI agent? In generative AI, an AI agent is an application that combines AI models for reasoning, tools for external interaction, and sophisticated coordination to achieve a desired goal. The agent's operation is driven by its logical architecture, which has a few key features. It is goal-oriented, uses an AI model as its brain, employs tools for action, and has a potential reasoning and decision-making capability that allows it to operate autonomously. To achieve these attributes, an AI agent coordinates three essential components-- model, tools, and orchestration. First, the model-- the brain. The model, which can be one or multiple AI foundation models, is the agent's reasoning center. Like a brain, it acts as the central decision-maker, thinking, planning, and figuring out the steps needed to achieve a goal. Models are typically general purpose, but can be refined with specific examples to showcase capabilities. Next the tools-- hands, feet, and senses. Tools are the connectors that allow the agent to interact with the outside world, often taking the form of APIs, using comments such as GET, POST, PATCH, DELETE. Like hands and feet, they perform actions such as sending an email. Like senses, they gather new, external information, such as fetching weather data. And lastly, the orchestration layer-- the nervous system. The orchestration layer is the central, cyclical process that governs the agent's operation. Like a nervous system, it acts as the communication network. It takes the brain's decisions, uses tools to take action, and then carries feedback from that action back to the brain to inform the next step. To learn more about AI agents, please refer to Google's whitepaper, Agents, accessible via the QR code on the screen and URL in the reading list. Quiz time-- Bea and Ann learned a lot about an AI agent and how it works with three components. They want to design an AI agent to handle insurance claims. This agent needs to first access the customer's claim history from an external database. Then the agent needs to validate the claim by checking it against the internal policy documentation. And finally, send a confirmation email to the customer. Let's match the components from the following list to the questions. Which AI Agent component is responsible for managing the sequence of actions and decisions which connects to internal and external resources and services? And which comprehends communication and logic-- the model, the tools, or the orchestration layer? Yes, A, the model, acting as the brain, comprehends communication and logic. B, the tools, connect to resources and services. While C, the orchestration layer, manages the sequence of actions and decisions. Did you get all of them right? In summary, AI agents advance the AI function from conversational, chatbots, to actionable. But how can you create an AI agent on Google Cloud? You'll find out soon.2. Let's practice!
Create Your Free Account
or
By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.