🦄 Making great presentations more accessible.
This project aims to enhances multilingual accessibility and discoverability while maintaining the integrity of original content. Detailed transcriptions and keyframes preserve the nuances and technical insights that make each session compelling.
Overview
📖 AWS re:Invent 2025 - From idea to impact: Harness AI agents and tools in AWS Marketplace (AIM3318)
In this video, Mike Levy and Idan Meislitz from AWS Marketplace demonstrate how to discover and deploy AI agents through AWS Marketplace and Amazon Bedrock AgentCore. They explain agentic AI fundamentals, address challenges in moving from POC to production, and showcase the AI Agents and Tools category launched in July with over 900 solutions. The session includes a detailed demo of subscribing to Articul8 LLM-iQ Agent on AWS Marketplace, deploying it via AgentCore Gateway, and integrating it with Kiro CLI coding agent to recommend optimal LLMs for generative AI applications. They highlight real-world use cases across legal, marketing, engineering, and R&D domains, emphasizing how AgentCore Gateway simplifies third-party tool integration while providing security and access control for enterprise agentic workflows.
; This article is entirely auto-generated while preserving the original presentation content as much as possible. Please note that there may be typos or inaccuracies.
Main Part
Introduction to Agentic AI and AWS Marketplace Integration with Amazon Bedrock AgentCore
Thank you all for joining us this afternoon and welcome. My name is Mike Levy and I lead business development for AI agents and tools in AWS Marketplace. I work with our partners and customers to help them realize the value of Marketplace. I'm joined today by Idan. Idan, would you like to introduce yourself?
Hey everyone, my name is Idan Meislitz. I'm a product manager on the AWS Marketplace team where I own our AI and ML based offerings. We're super excited to talk to you today. Our goal is to make agentic AI something real and tangible. We're going to cut through the hype and show you how you can discover agentic solutions through AWS Marketplace and deploy them in your workflow through Amazon Bedrock AgentCore.
Just a quick warm up: who here has used an AI agent in production today? Awesome. Whether you've used an agent in production or you're just beginning with some POCs, we have an exciting session for you today. We're going to help you understand how to quickly find and deploy AI agent solutions through Marketplace and Bedrock AgentCore. So let's dig into it.
I'll walk you through what we'll cover in the session today. First, I'll do a brief overview on agentic AI so we have a shared understanding and are on the same page. Then we'll talk through some of the challenges we've heard from customers on building AI agents, particularly getting them from POC into production. I'm going to walk through AI agents and tools in Marketplace and how we've helped customers get to production faster using trusted solutions from our third party partners and deploying them directly in Amazon Bedrock AgentCore.
Idan is going to walk you through some real life examples about how customers are using some of these third party solutions in their workflows. Finally, what we're most excited about, we're going to walk through a detailed demo on how to discover, use, and deploy an agentic solution using Marketplace and Agent Core together. So just to start with the basics: what is agentic AI? There are a number of different definitions floating around in the industry. From an AWS perspective, AI agents are autonomous software systems that leverage AI to reason, plan, and complete tasks on behalf of humans or systems.
Why does that matter? Why is it exciting? Agents are powerful in their ability to think iteratively. They can create a plan, iterate, and call different tools in order to accomplish a plan or take an action. There are a number of underlying ingredients to an AI agent. Obviously there's an LLM that functions as the brain to help decide which tools to call and which actions to take in order to ultimately accomplish a task. The AI agent helps combine various tools with other ingredients like observability, memory, context, and guardrails to help bring together the full AI agent in production. All these underlying components are part of what Agent Core offers customers today.
It's been fascinating to see the evolution from generative AI into agentic AI, but how are customers really using agentic AI in their workflow? When we talk to customers, it's fundamentally three primary outcomes that they're trying to drive. They're trying to automate business processes, improve their productivity, or reduce their costs. This is why Gartner thinks that by 2028, over 33 percent of enterprise software is going to include agentic AI, up from just 1 percent today.
Some of the main areas where we've seen development are things like business workflows, such as financial reporting and demand forecasting. Security and incident management is a popular one as well. There are things like workplace productivity, so think about integrations with tools like Zoom and Asana. Innovation and research is another area, in fields like engineering, media and entertainment, and life sciences, anywhere we're automating some of the complex data analysis. These tasks can free up humans and other resources to work on higher order things like strategic ideation rather than technical execution. These are some exciting developments, but what are some challenges that customers are facing in developing these solutions?
Fundamentally, with any technological innovation, many of our customers are facing a build versus buy decision. With agentic AI, that's no different. Customers face potentially increasing costs, time, security, and integration requirements that are hampering their ability to embrace the agentic AI trend. Marketplace, which was launched in 2012, has really been helping customers answer that build versus buy question for every technological development since then. Even when you're developing AI agents, getting from POC to production has become a challenge.
Gartner even predicts that over 40 percent of agentic AI product projects may be canceled by 2027 because folks are not seeing the business outcome or they have increasing costs or business value. We've seen challenges that customers have from experimentation to full scale production, particularly around scaling to multi-agent architectures and moving from POC into production.
That is why we helped our customers by introducing Amazon Bedrock AgentCore. For those of you who don't know, Amazon Bedrock AgentCore is a set of modular services that helps customers get everything they need to get agents to production. All the underlying components that we talked about—the ingredients to a recipe of an AI agent, things like memory, observability, guardrails, and runtime—are all part of the AgentCore building blocks that our customers can now use to get to production. You heard Matt Garman talk about a few new components that were released in preview today that we're excited about: policy and evaluations. This really represents a step function change for our customers that are trying to build agentic workflows.
Ultimately, what you can do with AgentCore is not just build your own workflows, but you can incorporate partner solutions as well. That's why we introduced AI Agents and Tools in AWS Marketplace as a category along with the Amazon Bedrock AgentCore launch in July. AI Agents and Tools in the marketplace really enables customers to have a one-stop shop with everything they need for successful AI agent implementations, whether that's vendor or customer hosted agents, whether it's all the underlying tools and ingredients like MCP servers, guardrails, knowledge bases, or SaaS with embedded agents, which is an important part of the category, as well as professional services to really help our customers tie together all of their agentic solutions in one workflow regardless of where you are on your agentic journey.
We launched in July with over 900 solutions from several hundred partners, and since then we've seen tremendous growth in the number of agentic solutions, whether those are AI agents, agent development platforms, SaaS that includes embedded agents, and professional services, which is a big piece of that. Through the marketplace and making these partner solutions available to you, our customers, you really have faster access to the right solutions. You have the ability to build POCs because there are features of Marketplace like demos and free trials to help you get started faster. You can purchase like any marketplace solution with your AWS account, so you benefit from a procurement perspective that really extends to AI agents and tools solutions, and also faster deployment, which is exciting for our partners who are looking to get started with integration with Amazon Bedrock AgentCore runtime and Gateway.
What does that integration look like between Amazon Bedrock AgentCore and Marketplace? It looks like this. Customers can deploy and run their agentic solutions using Bedrock AgentCore. On the left hand side, you'll see some examples of container-based solutions, whether they are agents or MCP servers that can be deployed in our customers' environment in their VPC and integrated with Amazon Bedrock AgentCore runtime. Runtime is a component that helps provide specialized compute resources purpose-built for agents. On the right hand side, you'll see vendor hosted solutions—think API-based solutions or servers, agents, knowledge bases, and so on—that are accessible through Amazon Bedrock AgentCore Gateway. Gateway is something that Edan is going to talk a little bit more about.
Real-World Applications and Understanding AgentCore Gateway Architecture
Speaking of Edan, he's excited to talk about a number of different real world solutions that we've seen with partners that are helping our customers today. Let's talk about a few real world examples where we are seeing customers using different solutions to build agentic workflows. Let's start with legal. Legal workflows usually involve a lot of processing documents, processing contracts, and transcribing legal proceedings. We have tools from vendors, as you can see here on the slide, to help build those types of agentic workflows. If you're in marketing or if you are building e-commerce websites and you need personalization, you need to generate content, you need to generate images, you could use products from these partners and many more to help you with that.
On the engineering and IT side, we also have a bunch of solutions that can help you with observability, coding, and many other use cases. And then lastly, research and development. Today when you use models, a lot of the time those models don't have all of the data they need, and you want to give them more data to use, whether it's real-time data or specific data from specific vendors. You can use vendors like these to help you with that. Now to the fun part—let's get to our demo. In our demo, I'm going to be showing you a use case, a coding use case, where you're building a coding agent and your coding agent is building types of applications, such as generative AI applications, and you want to make sure that whatever your coding agent is building, it is using the best LLM for the job, whether it's performance or price. I'm going to show you that full use case.
I'll show you how you can discover a relevant solution on AWS Marketplace, how you would subscribe to it, and then we will deploy and launch the solution on Bedrock AgentCore called Gateway. We are going to be using Kiro CLI as our coding agent that will use that tool to build out those generative AI applications.
Before I get into the actual demo, I just want to spend 30 seconds talking about AWS Marketplace and AgentCore Gateway. For those of you who don't know what AgentCore Gateway is and what it provides, basically AgentCore Gateway, together with Marketplace, allows you to simply integrate third-party products and tools with your agents. It also provides a security layer to make sure that when your agents are calling tools, they have the necessary credentials and permissions. You can also limit which agents are able to call which tools using Gateway. Additionally, you have the ability to create multiple gateways where each gateway is essentially a tool set that specific use-case-specific agents can use.
So how does it look? On your left, you basically have an agent and then you want to connect your agent to resources, to skills, to tools. Those can be APIs, those can be MCP servers, those can be Lambda functions. Some of those can be products that you purchase on AWS Marketplace. AgentCore Gateway comes in as a layer in between your agents and your tools, your resources, and your skills. Your agents with Gateway only need to connect to one MCP server, which is actually the Gateway, and when they connect to that Gateway, they get access to all of the underlying tools that exist within that Gateway. You can create tens of thousands of gateways, and for each gateway, you can select specific tools and create use-case-specific gateways so that specific agents have only the tools that they need.
Live Demo: Deploying Articul8 LLM-iQ Agent Through AWS Marketplace and Bedrock AgentCore Gateway
Now for the demo, the use case is a coding agent which we wanted to get some knowledge about which LLMs are best for which use cases. The way to start is basically to go to this web page, which is our AI agents and tools solutions page where you can describe your use case in natural language, and we will recommend agents and tools for your use case. The first step is you type in your use case, and then once you're ready, you click submit. Now we are going to present a few options for your use case. For this use case, we see that the Articul8 LLM-iQ Agent is one that fits well. It can help us select which LLMs are best for which tools, and that looks like exactly what we need. So we are going to dive deeper into it.
Now, this is the product descriptions page where you can see all of the details about the product. You can see pricing, you can see supported use cases, and you can also see that it supports Bedrock AgentCore. You can also see how to use it, so you can evaluate the product. It looks good. The next step is to subscribe to it on AWS Marketplace. In this screen, you'll see all of the relevant information regarding pricing, and then you will click subscribe.
Now once we've subscribed, the next step would be to launch the agent and to put it behind the Gateway. We're going to go to our launch wizard and as you can see at the top there were two options: an API option and a Bedrock AgentCore option. We're going to select the Bedrock AgentCore deployment. We're going to continue through the wizard. The first step is to get the access keys from the seller. When someone signs up, they are redirected to the seller, and at this point the seller is going to provision the API keys directly to your AWS account since they use a feature called Quick Launch on AWS Marketplace so that you don't have to do that by yourself.
The next step to create a Gateway would be to set up the authentication. Basically, we need to tell Gateway that when it's calling this specific tool, it needs to use this specific API key. We're going to copy our API key into this identity screen in AgentCore, and we're going to paste in the API key and add it. Once we've added it, if we scroll down, we will see that we have an API key ready to use. In the next step, when we create the Gateway, we'll use this key. Now we are going to click this button, add product to new Gateway, and now we are going to be redirected to the AgentCore Gateway console to set that up.
Now you're on the Bedrock AgentCore console. Because of the integration with AWS Marketplace, all of the required details are pre-filled. You provide a name, select authentication, and we're going to use Cognito for easy authentication. A target is basically a tool, so we're going to name the target. As you can see, the OpenAPI spec that is required for this integration is already provided. That's because of the Marketplace-based integration. Now we're going to select the relevant API key to use. We clicked create gateway and now we are waiting for the gateway to be ready to use. There you go. The gateway is ready to use. Now clients can connect to this gateway.
That's going to be the next step. We're going to be connecting our Kiro CLI to this gateway. As you can see here is the target that we selected before. In order to connect an NCP client to the gateway, we need first to get an access key. We're going to use this command and run it in terminal, which will exchange our client ID and client secret for an access token. When we are calling the gateway, we will use that access token. I'm going to be replacing the Cognito link with the client ID and client secret. Now I have got an access token that I will use in a few seconds when I'm setting up the gateway connection.
The next step when you're setting up Kiro CLI is that, like most MCP clients, you have an MCP JSON file that you edit. This is what the MCP server JSON file looks like. I've already pasted a stub just to make it easy to insert the gateway. All we need now is the MCP URL and the access token. This is where you get the gateway MCP URL. We're going to copy that and paste it into this JSON. Now we are going to copy the access token that we got a few seconds ago. Basically now Kiro is going to use that access token to access this remote server on the gateway. We are going to save and then once we do that, we are going to launch Kiro CLI.
Now you can see Kiro CLI is attempting to connect to our MCP server, and as you can see, it is connected successfully to the RIVGateway that I configured a few minutes ago. Now we are just going to do slash tools to see that it has the tool connected, and as you can see here, via that RIVGateway, I can access this tool called Get Recommendation for an LLM. Now we are going to actually call it. Imagine your agent is building a RAG application and it wants to know which LLM to use that best balances price and performance. I'm going to write a prompt. If this was a coding agent, it would have written the prompt by itself. Once I click enter, Kiro CLI is going to ask me for authorization to call the tool. I'm going to allow it.
Once I get back a response, you will see a couple of options that we got from the LLM IQ agent for LLMs that suit best for RAG use cases, for example, Anthropic Nova and so on. Now your coding agent can make a good decision on which LLM to select. That's the end of our demo. You saw very simply and very quickly how it is to purchase a product and then launch it on the Bedrock AgentCore Gateway and then get started using the product on the gateway.
If you want to learn more about agents and tools on AWS Marketplace, you have these two QR codes here where you can learn more. Michael and I are going to be here for a couple more minutes if you have any questions, and please also leave us feedback on the mobile app. Thank you very much, and we hope you found this exciting.
; This article is entirely auto-generated using Amazon Bedrock.
Top comments (0)