WEBVTT 00:00.960 --> 00:02.960 Hello everyone and welcome. 00:03.560 --> 00:07.680 In today's session we will learn about agents and Amazon bedrock. 00:08.080 --> 00:09.280 What are agents? 00:09.560 --> 00:16.280 An agent helps your end user complete actions based on organizational data and user input. 00:16.920 --> 00:24.880 An agent orchestrates interaction between foundation models, data sources, software applications, 00:25.040 --> 00:27.120 and user conversations. 00:27.680 --> 00:34.880 Agents automatically call APIs to take actions and invoke knowledge bases to supplement information 00:34.880 --> 00:36.200 to these actions. 00:36.640 --> 00:42.840 Developers can save weeks of development effort by interacting agents to accelerate the delivery of 00:42.840 --> 00:45.080 generative AI applications. 00:45.520 --> 00:48.680 What are the different tasks that agent can perform? 00:48.920 --> 00:55.880 They can extend foundation models to understand user requests and break down the tasks that the agent 00:55.880 --> 00:56.880 must perform. 00:56.880 --> 01:04.170 In smaller steps, they collect additional information from a user through natural conversation. 01:04.610 --> 01:10.330 They take actions to fulfill a customer's request by making API calls. 01:10.810 --> 01:16.770 Agents can also augment performance and accuracy by querying the data sources. 01:17.290 --> 01:21.730 What are the prerequisite steps for agents to use an agent? 01:21.770 --> 01:23.810 You perform the following steps. 01:24.210 --> 01:28.690 Configure an agent and add at least one of the following components. 01:28.970 --> 01:32.290 You can add action groups that agent can perform. 01:32.650 --> 01:37.650 You can associate a knowledge base with agents to augment the agent's performance. 01:38.050 --> 01:46.890 Optionally, you can customize the agent's behavior by modifying the prompt templates for the pre-processing, 01:46.930 --> 01:55.890 orchestration, knowledge based response generation, and post-processing steps that the agent performs. 01:56.330 --> 01:59.650 You can then test your agent in the bedrock console. 02:00.130 --> 02:03.390 So what are the different components within the agents. 02:04.030 --> 02:06.870 The core component of agents are as follows. 02:07.310 --> 02:15.110 Foundation model agent invokes foundation model to interpret user input and subsequent prompt in its 02:15.150 --> 02:16.990 orchestration process. 02:17.470 --> 02:18.630 Instructions. 02:19.030 --> 02:23.190 You can write instructions that describe what the agent is designed to do. 02:23.630 --> 02:27.270 Then you can also have one of the following action groups. 02:27.430 --> 02:33.670 You can define actions that agents should perform for users by providing the following resource. 02:33.830 --> 02:41.470 You can either invoke OpenAPI schema to define the API operations that agent can invoke to perform its 02:41.470 --> 02:47.230 tasks, or you can write a Lambda function with input and output. 02:47.830 --> 02:53.310 The other important aspect of the components that agent has is knowledge bases. 02:53.590 --> 02:56.910 You can associate knowledge base with an agent. 02:57.270 --> 03:00.070 You can also provide prompt templates. 03:00.390 --> 03:07.360 Prompt templates are the basis for creating prompts that need to be provided to the functional model 03:07.360 --> 03:09.600 or the foundation model. 03:10.160 --> 03:12.920 So how does it look like doing the build time? 03:13.200 --> 03:16.200 Here's the build time diagram of the agent. 03:16.640 --> 03:22.680 If you notice here, center is the agent that interacts with action groups on the left hand side. 03:23.120 --> 03:26.480 It has open API schema or Lambda. 03:27.000 --> 03:33.160 It can also interact with the knowledge bases to get the context of the relevant use case. 03:33.680 --> 03:40.960 It pulls the prompt from the prompt store and there's a default prompt templates available that it uses 03:40.960 --> 03:41.680 as well. 03:42.200 --> 03:48.280 So this is the build time phase of the agent that has different components and how it interacts with 03:48.280 --> 03:48.680 them. 03:49.120 --> 03:55.560 Let's understand in our next video how we can use these components and create a runtime environment 03:55.560 --> 03:58.280 for agentic behaviors to execute. 03:58.720 --> 04:02.280 Thank you and I'll see you in the next video.