Hello everyone. Welcome to my session. My name is Samuel bfi and I am a principle solutions architect with AWS. On today's presentation, I'm gonna be talking about Amazon Bedrock Agents, and the goal of this presentation is to show you how really we have enter an agent ecosystem and we have entered the Gentech era. So actually, when we talked about agents it can be confusing because that terminology can be quite. Overuse and for different companies and can mean different things. So let's just set the level of understanding of what we're gonna be. Ma you know what we mean here on this presentation, when we talk about the Gentech AI at the very high level, eight AI agents are autonomous systems that can reason, plan and complete multi-step tasks on behalf of humans. So how do they do that? Agents will break down high level objectives into executable tasks That's also known as planning. Once they have those tasks, they'll try to accomplish those tasks using tools and informations. Information can be knowledge, for example, from knowledge base that from your specific proprietary enterprise data sets. And those tools and knowledges are gonna be made available to them, and that's what we call actions. More importantly, these systems are actually capable of observing the results and that updating the plan in real time until the defined goal set by the human is accomplished. And that's what we mean by reasoning. So I'll give you an example instead, if you have an an example of trying to check availability of a meeting room, rather than just checking the availability of the meeting room, an AI agent can actually access your calendar, check everyone's availability. That should be on that specific meeting. It can send, invite to that specific meeting. It can then monitor for responses on that specific meeting. And if some of the critical persons people actually respond with a no, it can follow up asking for potentially times that would work for them and that do that work. So be a multi-step and driven autonomous solution. So why is now the era for AI agents right? Multiple research organizations in the world last year and this year. These two numbers I'm gonna show here are actually from Garner from, trends that they expect to see in the next couple of years is starting this year. So the first one is 33% of enterprise software applications will include some sort of a ent AI by 2028, which is upfront less than 1% in 2024. And then 15% of the day-to-day work decisions will be made autonomously through Agent AI by 2028. So you can see, about three years, a very significant chunk of the work is expected to be handled by AI agents. So it's very important that companies start to realizing the prospect and the opportunity that lie ahead, but. There is a evolution of agents that have happened in the industry in the last year or so. As large language models became more and more intelligent and more and more powerful, it has unlocked different use cases. So in the beginning, you'd have maybe a simple assistant, would you be able to ask basic queries and you get a response and maybe it will be like. Using a single tool use, maybe you were gonna connect with your rag system with your knowledge base and you use that, right? It would be a simple, a single step test. Then, companies. Using the power of these large language models created deterministic agents, right? So it's a very street strict procedure within AWS we call the, or we call those more like flows. So be a flow of steps that are predetermined and you just have a router asking After that, a lot of companies. Like Amazon and AWS have enabled organizations to create autonomous agents. So you can actually have the planning aspect that I meant before where it can go and understanding the objective and monitor for the outputs that the model is putting into a multi-step tasks. But there'll be a single agent in this case, which is, it might be expanding the two sets to multiple tools. Maybe you have a calculator, maybe you have a browser access, maybe you have. Web search access, maybe have some sort of proprietary API that you're connecting and enabling that agent to use, but it's just a single agent. Right now, what has become more and more popular and what we believe is the future is this agent, virtual workers. Rather than just having a single agent that has the capability of running the planning and multi-step tasks, now you can have a combination of powerful experience and expert agents on a specific areas that will work together to achieve a much higher level task. So you can think about long running systems that you actually mimic human things rather than just mimic human actions. And it can be cross organizational boundaries. And the more you see on this slide that you see on the right, the more autonomy in business impact we will actually provide. So with that said, how do we at AWS offer services and capabilities for our customers to innovate and really jump into this agent tech award? We have what we call the AWS Generative AI stack, which is divided by three different layers. The first layer, which is the bottom one, is the infrastructure that AWS provides to build and train models. So we have Amazon SageMaker, ai, which is this huge ecosystem for, cleaning data, training your model, influencing your model, and you actually be the whole managed infrastructure. But on top of that, we also have our own hardware or partners for GPU and. AI acceleration like Nvidia, and we have our own hardware called AWS Training and NWS information that provides the best price performance into the AWS Cloud. The second level of our AWS generative. AI stack is the ability for you to use models and tools so you can think about building blocks to build your generative AI applications. And that's our very popular service, Amazon Bedrock. And we're gonna be talking today in the next couple of slides. And the demo that I want to show you is gonna be building on top of Bedrock, but our. Highest layer is more on the application to boost productivity. We have a suite of services called Amazon Queue, and Amazon Queue has different flavors. In this slide, this is presenting two of them, which is the Amazon Queue business, which is this assistant that can actually provide a chat interface, but also automate a lot of tasks and connecting to your very common applications like your. Office 365, your Google Drive, your Salesforce, your Zoom, and it can put all that into a single pane of class with a lot of the guardrails and security that enterprise requires. And on the other spectrum is Amazon Queue developer that provides developers the capability of using ai, powerful AI models to boost their productivity while developing applications. So that is the AWS generative AI stack that we offer at AWS at the very high level. What we are gonna talk today is about Amazon Bedrock, very specifically Amazon Bedrock agents, because this presentation is for agents itself, but Bedrock is a suite. Of building blocks that allows you to build very quickly and very with a lot of the scale in mind and also security. Your generative a application. So you have a lot of tools here. Tools could be your knowledge base, which is the manage, retrieve, augmented generation system. You can have, data automation that you can upload PDFs and you can extract all those PDFs with the location. You can fine tune your model. You can do model distillation. We are gonna talk about agents. So agents will be a very important part of this capability that is native. But you also have developer experience. You have prompt management. You can create prompts. You can create variables within those prompts. You can store them on AWS, you have a whole suite of ID capabilities for you to really experiment and do POCs and test different ideas. Through the id you have prompt optimization. That is pretty cool. Functionality on Bedrock. The core functionality on Bedrock is really making models, AI models, foundational models from Amazon and other third party like Tropic AI 21 Cohere, mistro Yama, and many more. Manage for you in a fully serverless capability, right? So you don't need to, you don't need to spin any instance. You don't need to pay for any GPU. You just pay for the consumption that you have, in input tokens. Output tokens and you know that is what you're gonna be paying. You can also explore different models. You can maybe train a model from another outside bedrock, and you can import the model to be hosted on bedrock. You can use the bedrock marketplaces that actually expand the capability of models. Much further than just what is available via serverless on demand capability, there is a whole suite of evaluation metrics, right? Applications are very powerful. AI applications are very powerful, but they're really important for you to run evaluations. So a Amazon bedrock also provides you with a suite of tooling and capabilities to do programmatic evaluation. A rag evaluation, human in the loop evaluation. And just recently we announced the large language model as a judge. So you can have another judge you can have another large language model judging the response from your AI application. And then, you can collect those metrics and have dashboards and, improve as you see the those results. And then of course, you can have inference at scale. You can have different. Flexibility of options on how you consume those models. You can have optimized inference. By prompt caching, you can have intelligent prompt routing, so depending on the question that you are sending to bedrock, bedrock can automatically redirect to the right model. That will be the best suited to answer and execute that specific task or question. And it's global reach. Amazon Bedrock's, global reach. There are many regions on AWS cloud that have bedrock enable, and you can also have cross region inference for a specific model. So you can actually utilize the capacity across different regions with a single endpoint, and that makes your life much easier, right? At the very high level and very quickly. That is what Bedrock is and bedrock just becoming bigger and providing you more flexibility and building blocks to build your generative a applications with the whole experience that you are, if you are already using AWS, you are familiar with. It's very important to mention it's secure and responsible. So you can use VPC private link, you can use encryption. No customer data to any model provider including AWS or third party are collected and store. You can use guardrails, so you can set different restrictions from what your application might be able to respond and the things you don't want the application to respond. There is a lot of compliance regulation that. Bedrock has passed and has the stamp. But also there is a lot of open source integration you can integrate to Lang Chan Graph Lamb Index through ai and many more. Open source integration frameworks can tie into the vast service, which is Amazon Bank. But because we are talking about agents AWS offers a broad choice for deploying agents. And here are just three examples, but these. Could be way bigger. If you want to run agents that are specialized, you can use the Q agents on the Q4 business or even Q developer and you can automate your enterprise productivity and workflow. We are not gonna talk about those agents, and then on the right side, you can do it yourself. You can, for expert developers potentially using. Open source frameworks you can see, build that on AWS, use our compute, use bedrock models, use our serverless capabilities like lambda step functions and automate and create your application. What we'll talk today from the remaining of the session today is using Amazon Bedrock agents, which are manage service that has all the scaling and capability. An orchestration that will be foundational. The foundational model will power those agents, right? So this is our fully managed solution for building and deploy agents. What sets actually Amazon Bedrock Agents apart is the building foundational models, powerful orchestration. So the orchestration and the flexibility on how you decide to build those agents are something that was taught from the ground up. This means you are not starting from scratch. You're leveraging sophisticated AI capabilities right out of the box that AWS has made it available for you. So if you don't have a large data science team, or you're small startup and you just want to start using agents that have the capability of orchestrating complex tasks at scale, Amazon Bedrock Agents is a great fit. So you can think about Amazon Bedrock agents as like having a team of AI experts working alongside your developers that can handle complex orchestration, while your team will actually focus on the unique aspects of your business logic. What this approach allows you to do is to dramatically reduces the time and expertise needed to get production ready agents up and running. And that is the important thing, is we are not just talking about POCs anymore. We want to go to production and actually provide the value both the economic and business value that our organizations are expecting into this new era. While the potential of agents are exciting, we keep hearing from customers that they face several challenges when they try to build production ready agents. Developer teams struggles with complex technology stacks and infrastructure setup. This is a very fast pacing ecosystem and new frameworks keeping merging. Maybe they're open source, maybe they are different partners, and that it becomes such a huge ask for the development team to keep adopting. The other area that is challenging is the operational complexity. Managing multiple agents and orchestrating complex workflows is challenging. How do you scale those? How do you actually provide the right sets of tools? How do you make sure entitlements and data are not actually being widely distributed when they should? Another one is, you know that I already mentioned a couple of aspects is agent ops requirements. What are the governance required for you to manage those agents? What are the security and compliance that your business will require you to have? And mostly important, which is one thing that when a lot of companies are doing POC with the agents, they potentially fail to realize this. What is the observability stack? How do you trace each step that this agent tech system is doing? How do you log those and how do you run evaluations to actually improve the capability and the performance of those agents? Right and rapidly evolving technology landscape makes it hard to stay current while maintain production seed. Systems. So the important thing here with all these challenges is Amazon Bedrock agents, because it's a managed service and AWS will take the heavy lifting away from you so you can focus on your business values and the outputs that you are expecting to do. You will make your life way easier. You still need to put a lot of work, but it'll be easier than potentially building on your own. So before we talked about what different types of agents does bedrock agents provide? How does an agent existing looks? So you might have a user input. That user input might have an, we will be talking to an agent application and you expect a response. But what happens behind the scenes is much more complex than just this flow that the user sees, right? The agent components that your agent application will potentially build, you're gonna talk about the foundational models. So different tasks might be using different foundational models from different partners. Maybe you're gonna use some of the tropic models, maybe you're gonna use some of the Amazon Nova models, right? But also now that you have those foundational models, those models, because they were not trained within your proprietary. Enterprise data needs access to your knowledge base using vector databases. So Amazon Bedrock also provides a managed re solution, which is known as knowledge base. On top of that, enterprises, we will make, wanna make sure that the responses that are provided to potentially our external users or internal users, depending on the type of application you're building, have some safeguards in terms of the response and the outputs that you're trying to. Respond that, and receive, and Guard Rails allows you to do that. You can create a lot of the tooling within Bedrock as well. You can connect with your own APIs by building action groups within the Bedrock agents. You can have memory built in, you don't need to ma manage the memory of those agents. You can have memory being managed store and maintained by bedrock, and you can just provide the ID and ask conversations for specific users. Keep growing. You still can maintain the memory, which is very important because it's context from the users. Requests, previous requests and responses. You can connect to other agents, but you can also have, deterministic flows which are, more like. Static steps that your agent can take, but all that, right? One of the things we see more and more in production is you wanna make sure that you have a continuous evaluation framework, right? So as your agents are potentially being consumed in the ward, once you go to production, you wanna log the valuation of those. Of those steps and you wanna make sure you have test cases that you keep evaluating. Once you make changes, you wanna make sure you have metrics and grading prompts. And you can use being used judges as foundational models that I mentioned. So you can log all that into analytical database or a vector database. And then the last step is you can have this dashboard where, add your experts on whatever agent. Output or agent task is created. What is your business value? You can have those humans evaluation to analyze what's working well, what's not working well, and you can update and optimize to make sure your agents are actually getting better with time, right? So users will provide feedback and agent application get updated, and as they get updated, they'll get way better, but. Okay, let's talk about what is Amazon Bedrock Agents? So Amazon Bedrock Agents provides you with a managed capability for actually managing and running agents at scale. So if we think about the different tasks that you and your development team would be required to run by creating agents, you probably will know that you need to provide some tool capabilities for. Your agents, right? Within Bedrock agents, you have the ability to create tools with action groups, which can call potentially your own APIs. You can integrate with knowledge basis, you can actually have code interpreters, so your agents can generate code. They can run those codes in a sandbox environment that Bedrock provides to your agents, and the output of that code can then be consumed for. Potentially as a response to your user or later on collaboration with other agents. I already mentioned about memory. But one interesting thing that we're gonna do with them, which today is this idea of multi-agent collaboration for complex workflows. It's important to say that the less one before last item, the comprehensive trace debug and observability is very important and critical for bad. Okay. By default using these capabilities of bedrock agent, you get comprehensive tra trace and debugging logs and metrics that you can look and analyze each stage and each step that your agents are taking. And, there is building word rails for security and compliance control which you know, will provide a peace of mind. For you and your enterprise as you're building new capabilities with ai. So think about Amazon Bedrock agents as a powerful foundational model that becomes even stronger when combined with the full suite of bedrock agent capabilities, from knowledge base to flows, to guard rails, all working together to create a more capable, secure, and reliable agent application. I just wanna, I apologize, I just wanna. Drill down a little bit more on the new capability that we release late in 2024, which is the Amazon Bedrock agent's multi-agent collaboration. One of the most powerful capabilities is the moody agent collaboration. Let me walk you through how this works. First, you can easily assemble agents. And connect e each of those specific agents with their own specific knowledge base. You can give access to specific tools specific APIs. These agents on its own can plan, execute and comple complex tests togethers. Think of, multi-agent collaboration as a team of specialists that are working coordination. For example, one agent might handle data analysis while another manages customer communication, right? And they can all report into a specific supervisor that is planning and delegating different tasks and aggregating the results for these coordinator agents, right? Amazon Bedrock have built intent classification to help unify conversation across agents, ensuring that is smooth. Handoffs and consistent user experience across agents are kept. As your need grows, you can scale those agent experience efficiently because it has built team. Observability that lets you monitor how each agent are performing and interacting with your user. And of course, all of these runs on Amazon, bedrock Guardrails, and AWS enterprise security and privacy control by default. So to summarize a little bit here, what Amazon Bedrock Agents allows you to do is to provide the Lego blocks, the building blocks. For you to create a powerful agent AI solution for your own organization. I already talked about some of these configurations, some of these features I haven't mentioned session handling. So you can have session, then you can have memory management. All of these are secure within the Amazon Bedrock ecosystem. So this becomes even stronger capability. Because the point I mentioned before that development teams can get very overwhelming with this very fast paced environment of agent tech solutions with Amazon Bedrock agents. You can expect more and more capabilities are gonna be added in the future that will follow the same approach of managing that for you while you focus on your business needs. I'll give an example of something that Amazon have been very invested and started releasing some of these capabilities, and hopefully you see more and more coming the next few months and a year, which is the model context protocol. I. You probably have already heard about MCP, which stands for Modello Context Protocol. MCP was created by Tropic. It's an open source protocol that is standardized the way models share and understand context to each other and connect to external data sources such as tools, right? M making your application context aware of what tools are available for them. Okay, currently Amazon bedrock inline agent, which is a way for you to create agents with code inline already support MCP servers because, you can build clients. The agent. Agent is an MCP client that can connect to MCP servers. So you can have your MCP servers as action groups in Amazon Bedrock age, the same way you do when you're building a single agent, or when potentially building a multi agent collaboration that allows you to enable connecting agents to external tools and data sources via MCP servers. You'll support passing information between models and tools in a multi step workflow. You can expect more capabilities, not only within the bedrock inline agent, are gonna potentially become more and more native to this offering on Amazon Bedrock. Now, before I do my demo, what I wanna pause here for a few seconds is if you have found so far that Amazon Bedrock agents is interesting and you wanna know more, please. Scandi secure code to potentially go to our Bedrock agents page. Or if you wanna run a workshop and you're already an AWS customer, you can request a customized workshop where we, as part of the AWS team will, potentially virtually host a workshop for you, for agents, or potentially your account team might fly or meet you in your office with your team and actually host these specific workshop where you're gonna be able to create agents and play around with Amazon Bedrock. And of course, there are free courses that you can do on your own pace. Just to scan the last Cure code here now. Before I I'll talk about my demo, but I just wanna show a diagram of what, how my demo works because what I'm gonna show you on the demo is a web UI that is abstracting all these very powerful investment research and agent, which is actually using behind the scenes, the moody agent collaboration capability of Amazon bedroom. So I have created three different agents that are collaborators from my supervisor agent, which is called Investment Research Assistant. The goal of this assistant is to help, analyst, investors to do research about the specific investments, right? About the investment work. I first have one agent that is specialized on analyzing you. News, very specific agent is tasked with a different set of tools available for you. The first tool that is available is the ability to call an API to do web search. So if I ask about Amazon News Amazon Stock News this agent will be invo and they'll be actually calling web search APIs. And retrieving information about a specific requests. In this case that I said Amazon News. The other tool that this actually model has, sorry, this agent has, is the connection with proprietary financial data. In this case, for the demo is just Amazon earning reports, 10 Qs and 10 Ks that have been uploaded and prepared by bedrock data automation. And what happened here, I have created the Amazon bedroom knowledge base that ingest in, embed, and creates vectors into my open search serverless index. And then my agent can actually query using natural language to actually retrieve a specific chunks or of specific asks for example, Amazon earning reports. So that is one agent. The second agent is quantitative analysis agent. What this agent does, it has a action group that has tools capability to actually call a Yahoo Finance API. So you can call different APIs to check, different prices, different stocks, different date range, and that can be real time. And you return to the model to the age of top. Now the last agent is the smart summarized agent. This agent is prompt and capable of summarizing a specific set of informations potentially to give the response back to the main supervisor agent. The cool thing here is I've provided a specific guidance for each of those agents on how they should behave. But because Amazon Bedrock multi-agent collaboration has a lot of the orchestrating intelligence using, years and decades of machine learning experience of Amazon, a lot of that intent classification is handled automatically for me. And everything that I'm gonna do here is gonna I'm gonna show in the demo. I'll ask a prompt here to do something that can retrieve data and give a complex investment analysis. And you see that. Each step will be traced, and you'll be orchestrated by the supervisor agent with, which is this agent that you see here, right? The investment research assistant. And as needed, you actually call this collaborator agents, right? And each step will be able to see the input and the output for those agents until we finally, once it's satisfied because the planning. And reasoning. Once it's satisfied with the answer and the data that is collected, you get the response back to the user. So you can think, you can see that this is just more than just a simple agent, a simple chat bot, right? Even though this is a natural language conversation with the bot behind the scenes, you're using the power of connecting different data sources, different actions that a foundational model on its own has no capability. And everything you're gonna see on this demo, I send a single, after I've created all these agents, either by A-P-I-S-D-K, cloud formation, or the console, I can just re, I can just send AVO agent API, and I can use this capability. And because this is all managed on AWS, all the scaling, all the security, all the logging, all the valuation is taken care for me by Amazon Bedrock. So this is really powerful. So what you're gonna do now is we are gonna jump into the, into the a web browser and I'm gonna show you the demo on how this agent works and hopefully we're gonna get excited and potentially start building some of these capabilities on your own as well. We'll see you in a moment. Okay, welcome back. Here is just a very simple ui. Right behind the scenes I have already created the agent, the multiagent collaboration that I showed you. So here is me as the user. We are gonna send a prompt through this very simple ui. With investment research assistant, which is a supervisor. And then I'm gonna show you the traces based on my question. What type of the different agency requests to collaborate with, how does he actually get the data back? How does he keep integrating? Iterating until it gets to a final stage where it satisfy and actually returns the response. So this is like a, it can take, a minute or two, three minutes to actually get some of this response back because it's doing a lot of different steps. It's taking a lot of different action. Given the planning that the supervisor agent that is part of a Moody agent collaboration on Amazon Bedrock is actually doing right. So this is again, it's an investment research system. The question I'm gonna ask here is given the recent Amazon 10 Q, sorry, 10 K, what are the, some, what are some of the growth opportunities highlighted? Then I want you to search the news related to those opportunities, and I want you to list competitors for each of those areas. Finally give me a stock price range expected within the next 12 months, right? I wanna learn the growth opportunities. I wanna see what the market's saying in terms of expected price range from different sources and then search the current stock price and how much has fluctuated in the last 30 days, right? So in order for me to do this as part, as a human, or if as an analyst, you take me quite a lot of time. As I send this request behind the scenes, it's actually invoking my agent. So the first thing it does here invokes the investment research assistant, which is a supervisor. And then we are gonna see now multiple different steps. So the first step is the orchestration. I'm, and I'm gonna try to now go through all them because it'll be a lot, but I'm gonna show some of them. So the first one is just, you can expand here and you can see. You are an investment research assistant responsible for overseeing and synthesizing, synthesizing financial research for a specialized agent. You're always to coordinate the subagent to produce. Anyway, it's a very prompt, fine tuned prompt here for my supervisor agent, but what it does here is the supervisor agent will actually do the first reasoning. And the Asian reason is saying first what I need to do is get growth opportunities from Amazon. 10 K. 10 K. Then I need to find recent news about these opportunities. I need to identify the competitors from different from these opportunities I need. I need to get analyst price targets, and then I need to get the current stock data and recent volatility. What he has done is, okay, now that is the plan that I need to take and I need to take multiple steps to get the plan. So in the step four, what I realize is okay, I need to request the collaborator agent, which is the financial use agent, to go and retrieve some data, right? So the first one is go fetch from the knowledge base. Then if needed, fetch latest relevant news from. From a given stock base. So you would try to do the knowledge base. My knowledge base in this example only has Amazon on your reports, like the 10 K that you're talking here. So you can see here that the financial year is okay, I'll first search the knowledge base. From growth opportunities. So he has hand off the task from the supervisor into the financial news agent and to say okay, I'll search the knowledge base for growth opportunities matching in Amazon recent 10 k and related information. So he goes, queries the knowledge base on Amazon Bedrock knowledge base. He gets some of the chunks, right? Using the vector database here he then he orchestrates the response back to my supervisor, so orchestrates back my response to the supervisor, but now it's okay. I need to get more data. So it's like you're saying, okay, let me now get the recent news about Amazon growth initiatives and competitors. So he collected the information and the growth opportunities that are listed on 10 K, and now it's actually going and getting the. Growth Initiatives and competitors from Neils, right? So you can see here it's actually running. It's now finally doing something that is called the tool use, which is using the action group functionality on Amazon Bedrock, and is doing a search query using an API saying Amazon Growth Initiatives, AI Cloud. Retail healthcare competitive strategy plans because that's what probably was listed on than 10 K, right? It has returned that, returned some of those specific news and now it's saying, let me search specific for analyst price targets and competitor information. So you can see that it's taking it step as that is thinking is step at us at each step adds each step in order to obtain my final request, which was multiple, tasks, which prosecutions at the same time. So now he's saying, okay get the news about the price target and competitor market share for cloud retail, AI and healthcare. So he has gotten into that and it's based on the knowledge based information in the news, I can now provide a comprehensive update on Amazon's growth opportunity related development. So the financial news agent has now done within itself, has now done take multiple steps to finally. Achieve the single goal that the supervisor agent requested, right? So you can see how this becomes very complex. Remember the multiagent collaboration here? The good analogy is have a team of specialists working together, and this is one of the specialists just taking a lot of the actions, right? So the final response from the final news agents okay, key growth areas from the CEO recent communication is ai. Core focus is shopping code, personal assistant. The growth opportunities will be major infrastructure, AI infuse Alexa, voice assistant healthcare service expansion, showing positive customer sentiment. You can see some of the competitors here, and then you can see the analyst coverage. What are the, the price target that it has? Share that, that they have. Decided to put each of different kind of companies and research institutions. So it goes back to the supervisor and be like, okay, here's the data supervisor. What do I need to do? And the supervisor decides some information here and we could reel down, but I don't wanna take too much time and says, okay, now we need to actually get another collaborator here, the quantitative analyst agent to go and see. And you can see here, you'll try to, if you go here, the. Agent reasoning now for the quantitative analyst agent to be like, I'll retrieve Amazon stock data and analyze the recent price movements and patterns. So it goes and makes a stock data lookup API, using the tool function to use on the action group so it gets the data right, and then it's okay, analyze the current price range and key pattern over the past month. It gets all the data so it gets from current stock. A price for today, 30 days. Price lines, volatility, and the pre price. 11 trading partners. So it talks about some of the resistance and some more like the, terminology fundamental understanding of stock itself. It goes back to the agent, right? And the agent says, okay, let me synthesize this information to a comprehensive analysis of Amazon growth opportunities and Mark lookup. And then finally, if you see here on step 43, so this multi-agent collaboration has taken 43 different steps using different agents, using different tools, using different knowledge base. To achieve the response. So you can see that based on my prompt says here is the detail analysis of Amazon growth initiatives and market outlook. So the create growth opportunities. One is artificial intelligence scale, 1000 generative AI applications in developments within Amazon. These are the focus areas. Here's the competitors. Now it's cloud service, current growth opportunity. Current growth is 25 year over year AI contribution, three to 5% now. Expected to be 10% this year. Year. Remember the 10 Ks from 2024? That's site 2025 is the expectation now, infrastructure, metadata, center expansion, some of the competitors. Third growth opportunity is healthcare Talks about Amazon Pharmacy. One medical integration showing positive customer adoption. Then it goes to Project ky, which is the satellite internet service for Amazon. Then it talks about different competitors and finally goes into the current trading metrics. What is the price? What has been the 30 day range? What is the support level? What is the resistance level? And here is the analyst projection. So some of analyst projections are saying, okay, like Wolf research saying it's 200 notable j JP Morgan saying it's two 10 and recent trading partner has been. Higher lows since April 24th. Five consecutive days of gain, harvest rate trading was April four to nine. So it seems the wide ranging analyst targets 200 to $261, reflects varying opinions of AI initiative, distribution success, AI AWS growth, sustainability, healthcare expansion potential, a competitive ion impact. So I know it was a lot and I was getting more into the response. But this is the answer for this specific prompt that is sent. And if you were just to prompt a model without having these tools, it'll be way, way hard. Again, these only run for three minutes, but how about if you have a very complex task that involves, dozens of agents and you'll be running for hours, right? To do a very in-depth research. You could do that during your use case. And I just called a single API. This was just a single API call that I made. And that API call had all the infrastructure managed for me, all the foundational models using the power of the AWS generative AI stack. So again, my name is Sam bfi. If you work curious or wanna talk to me, please reach out on LinkedIn. Could just search by my name Sam bfi. Hopefully this demo was useful. Thank you so much for taking the time and watching and I'll hope you see you soon. Have a great one. Great one everyone. Bye bye.