Transcription
Good afternoon, everybody. Let's continue with the agenda. Thank you for joining this session. My name is Antonio Rodriguez. I work in Amazon Web Services and I'm also a developer as many of you, and also a cloud whisperer, as I like to say. We are really just whispering models these days, not really writing code.
And uh, in this session, we are going to be talking about how to take the applications that you are building in CL code from prototypes that you might have in your computer to fully production-ready applications that you can have in the cloud, in this case, with Amazon Web Services. And uh, I think the story that uh, we are telling today is a story about teamwork, a story about better together. Because uh, I remember I joined the team and started working uh, with Anthropic more than three years ago. I remember we were still launching Claude 2 as the big amazing model that was uh, you know, changing and shifting the ground uh, below us. And uh, since then, we have tried to combine the best frontier models that Anthropic has been developing with the best cloud provider uh, that we had in Amazon Web Services. And uh, this session, by the way, is a hands-on session. So what we are going to do is that I'm going to give you a brief introduction into how do we work together, Anthropic and AWS, and how you can use Claude in AWS in many different ways that we have available today. And then we are going to give you some accounts in AWS so you can, free of charge and with no limitations, test Claude in in in Bedrock, in this case, and also in in general in AWS so you can see uh, how to use it. You have guided instructions and a workshop, and again, we are going to give you an environment so you can play around with it.
So, um, again, this collaboration started and uh, has been continued as a very long-term relationship and um, we have uh, proven this year by year. Amazon has invested uh, has done a multi-billion investment in Anthropic uh, because we really believe that we are building the future of AI together. And at the same time, we are the primary cloud provider for Anthropic. And uh, Anthropic has uh, committed uh, more than 100 billion on usage in AWS as well, in the infrastructure that we are offering for powering the models that you are using every day with Claude. And uh, an example of the this investment is Project Reineer. Project Reineer is actually uh, one of the largest uh, AI compute infrastructure that was built for training and hosting the Claude models that you are using today with Anthropic. We are also using uh, custom and purpose-built chipsets that are specifically made for uh, the Claude models. In the chipsets that we call Trinium. So we have a our third generation of Trinium chipsets already built uh, by Amazon. So that you can actually get the best speed and also use it in the most cost-effective way uh, when you are expending uh, tokens with the with Claude. And uh, I think that if we start diving into the technical part of uh, this collaboration, there are three angles that we basically offer you when you use Claude through AWS. I think the first one is that you have a very comprehensive list of features that you can use uh, for let's say, accompanying the model that you are using. So in Bedrock, you have a full platform uh, so that you can combine and fine-tune the models as well. Uh, we are the only provider that allows you to fine-tune Haiku, for example, in the cloud. And uh, we have full integrations with the rest of the features that we have available in Bedrock and the AWS ecosystem in general.
The second and very important point is the security. So we can make sure that you keep control of your data at all times. We avoid this hopping of data going to the public internet. You can keep everything private in your AWS boundary. So we make sure that you uh, always align with the compliance and security uh, best practices that you might have in your companies or or the regulations. If you are working in banking or healthcare, u, these kind of industries, it's highly regulated industries. It's very important uh, that you actually have these options uh, for controlling everything that you're doing with Claude. And uh, obviously, the data privacy part. Uh, we work uh, nowadays in the latest generation inference engine that we have in Bedrock. Uh, we have zero operator access uh, in the platform, which means that no one from Amazon or from Anthropic can actually get access to those instances. So your data remains fully private. We are not sharing that data with anyone. No one is using that data for training models or anything like that. So you keep a full restriction. And again, this is because we need uh, to align with a lot of compliance regulations that uh, our customers need. There are hundreds of thousands of customers who are using these models every day, and therefore, it's very important for us to align on those.
Last but not least, the scalability. The using those models, Claude models through Bedrock gives you u, practically infinite scalability in the sense that you can uh, really decide how do you want to do the deployments. You have full flexibility for choosing regions as well. For example, you could decide that you want to deploy Claude only in the London region in AWS or only in the European region, for example, if you have to align with GDPR and and these kind of regulations.
Now, if we see the whole ecosystem, who who is familiar with Amazon Bedrock by the way in this room? Okay. So I was suspecting less hands, to be honest, because uh, this is really a Claude event, but uh, it's great to hear that you are using Bedrock or you know about Bedrock. And this and Bedrock has been growing a lot since uh, the the beginning, since the last uh, three or four years when we released the service. And today, apart from the core, which is obviously the foundation models that you have access to, and obviously uh, Anthropic Claude models are a very important part of that. You also have other features that help you in the full journey as a developer. So you have uh, features that help you doing evaluation, that help you doing prompt optimization, uh, that help you fine-tuning the models, as we mentioned before, or uh, doing model distillation as well, if you need it. We also give you some features that help you connecting. If you have RAG use cases, you can have knowledge bases that are fully managed so that you can attach to it as well. We have uh, Guardrails, which is a way in which you can apply content filters, you can deny topics, uh, you can uh, protect PII data or sensitive data by doing automatic masking and so on. And uh, you can also control hallucinations from the models by doing uh, proper grounding, automated reasoning checks. And uh, even more important in the last uh, two years, I would say, tools for building with agentic AI. And this, if you are using Claude agents, then awesome. But if you are using the Claude agent SDK and you are looking for an infrastructure for hosting your agents in a secure way in the cloud, then you can also use Amazon Bedrock Agent Core, which is uh, again, fully compatible with any open framework that you might be using for agentic today, like LangChain, like CrewAI, and obviously Claude agent SDK as well.
Now, if I have to summarize in a single slide everything that I'm saying so far on the why the better together story, why is it better to uh, use Claude in AWS, I would say this is probably the slide, right? So first of all, as I mentioned, the data that you can get when you are using Claude in AWS, you have a full control of the environment that you are setting up and where you are using it. You also have centralized billing from AWS, so you can actually have a single bill. If your companies are already using AWS or you are using AWS for other things, you can have a consolidated, a single bill that is actually having the usage of Claude included with everything else that you might be doing today. You have full observability through our services like CloudWatch or CloudTrail as well, in which you can actually see details on metrics, logs, you can see the traces of everything that has happened. And uh, we also offer SLAs, so we make sure that you have a guaranteed availability on the services, so you have an uptime that is uh, secure for your applications, and you can go uh, safe to production and make sure that you can scale at the at the cloud scale uh, safely.
Um, another very important point is that we have built-in OAuth integration. So if you are using single sign-on or if you are doing Open Authorization, OAuth, or if you are using IAM roles in AWS, for example, um, you can directly integrate that with your identity provider. You could be using Microsoft Entra ID, you could be using Okta, you could be using any of those uh, identity providers today. And that is a built-in integration that you can do also for using Claude in Bedrock. And uh, last point, you can use PrivateLink, which means that all the communication can remain fully private. So you make sure that no data is being sent to the public internet at any point, uh, when you are using Claude in Bedrock. As I mentioned, I mentioned about compliance before. These are some of the compliance regulations that we uh, support and we are uh, good for you to use as well, like, for example, FedRAMP, HIPAA, and so on and so forth. There is a full list online if you want to check it out and uh, and more information about all of these topics.
Now, the good news is that today you have three ways of using Claude in AWS. And actually, this is the the most options that you are going to find in the whole industry right now. The first one is you can use Claude models directly through Amazon Bedrock, as I mentioned before. So basically, everything that I was saying so far, in which you have the same uh, Haiku, Claude experience that you might be getting directly, but in this case, fully through the Amazon Bedrock uh, APIs that are offered to you. We have recently added Claude 3.5 Sonnet and included the Messages API as well, so that you can directly invoke with the Messages API if that's a method that you prefer.
Now, we have also recently launched Claude Platform on AWS. Claude Platform on AWS is actually generally available since a few days ago, and this allows you to have the same exact experience that you might be using today directly with Anthropic, but in this case with a consolidated billing through AWS and having the access control fully done on AWS as well. We act as a gateway in this case, and we send all the inference requests to Anthropic so that they can process the request for you. So it's giving you the best of both worlds. So you get again the security, the billing, the consolidated access control that we are giving in AWS, and at the same time, you have all the feature parity with Anthropic, including everything that we have been hearing in this event and the latest features that you might be using today, including web search, files, agents, and so on and so forth.
Now, the third uh, way in which you can use it is also if you are not using programmatically Claude, but you are rather using Claude Desktop. So you might be using CoPilot, you might be using uh, Claude Code through the Claude Desktop application. You can also use it directly from AWS. And again, you can pay in a consolidated bill in AWS. Uh, and that you have two ways of uh, doing that as well. One is through the Claude Enterprise in in the AWS Marketplace, in which you pretty much subscribe to the number of developers that you have using this application as well. But you now also have the option of using it uh, and and let's say, point directly the Claude 3.5 Sonnet, which is one of the latest additions also that the Anthropic team did for uh, connecting Claude Desktop to third parties. In this case, you can connect Claude Desktop to Bedrock, or you can connect even Claude Desktop to the Claude Platform on AWS if you prefer to do so. So again, you have all sorts of combinations to make your life easier and uh, pretty much meet you where you are on your needs on this.
Perfect. Let's dive into the actual workshop that we are going to be building together. Um, I just want to uh, make a recap that first of all, you need two prerequisites for running this workshop. One is you need an AWS account. So that's we got you covered. We are going to give you again a workshop account that has uh, an environment that you can play around with, and uh, you can pretty much be uh, free of charge, free of any concern of using that environment. And then the other thing is that you obviously need to have Claude Code installed uh, as an application. And uh, we have a few options for you. So if uh, the folks on the back help me switching to my laptop, I can show you what is exactly what we are doing.
Um, you are going to find that in the accounts that we are setting up for you, we are having the workshop that is called Introduction to Claude Code on AWS. And uh, this workshop has some instructions to how to set up everything that we are doing. We are pretty much uh, playing around with the ScalyDraw uh, tool with the repository of the ScalyDraw. So you can see how Claude Code can help you playing around with a ScalyDraw and drawing architecture diagrams for you. For example, um, this is a basic workshop. If you are already an advanced user who who is a getting started with Claude Code in this room, all right? Who would say that is a ninja in Claude Code? So you have all the skills on Claude Code. Who is in the middle? All right. Who's not answering? So uh, yeah, I guess most of the room is probably in the middle. So you will find that probably module one is basic for you. So it's a just using Claude Code on AWS and learning how to set it up and point to ScalyDraw and so on. Uh, module two and module three are probably more interesting for you. We are going to start uh, playing with the context. We are going to start using Playwright, MCP, so we can actually uh, take screenshots of the diagrams that we are drawing and making uh, modifications automated through Claude Code on it. We are also playing around with the uh, Git workflows and on Claude Code and so on. And in the module three, we are going to play with sub-agents and plugins, and we are also creating custom skills, hooks, and some advanced uh, parameters for configuration.
I just want to mention that we have an even more advanced workshop that is also available for you if you look for this URL. And uh, basically, there you have other things that are very interesting for teams who wants to move to production in a safe way. So here you can learn how to set up uh, team standards, how to implement and and distribute uh, advanced workflows as well, how to do a scalability, cost control, and so on. So, for example, you could see uh, how to set up dashboards like this one for making sure that you can control all the usage that you are doing of uh, Claude Code tokens uh, in your accounts and so on and so forth. Developer productivity measurements, measuring return of investment of your applications and so on. And last, you have also options for using uh, the Claude agent SDK on the Amazon Bedrock agent core. So you can learn how to build the agents connected to the Claude infrastructure uh, that we have uh, here in in AWS as well.
We're coming back to our workshop again. Uh, you have different ways of uh, doing it. You will see that at the very beginning, you will uh, see a screen that looks like this with some terms and conditions. I just want to remind everybody that those are workshop accounts. It's uh, individual for you, but make sure that you don't upload confidential information, personal information, or anything like that. You will have to agree to the terms and conditions and click on the join event. And from that point on, uh, you are going to see pretty much this experience that you have in the instructions here, where you actually going to have a an environment fully deployed for you. And uh, you will see a URL here that will take you to a terminal and a pretty much a Visual Studio Code UI where you can actually work on all the instructions that we are doing in the workshop.
Let's say that you don't want to follow this path and you want to do it in your own machine. Then you can do it. You can use Claude directly in a terminal. If you prefer, you can configure it for pointing to Bedrock uh, with instructions that we are giving you in the workshop. Let's say that you want to use CoPilot uh, or or the Claude Desktop app. Uh, you can do it as well. Again, here I'm pointing to my Bedrock account. So I have a Claude Desktop pointing to Bedrock, and uh, here I'm doing the ScalyDraw uh, application and the and the workshop that we have for you uh, over there as well. And uh, there is more information at the end. We are going to be sharing with you uh, more on a guidance, for example, for how to use Claude Code and CoPilot on Amazon Bedrock. There is a full blog from my colleagues uh, who is covering all of this in detail and so on and so forth. All right, let's go back to the slides. Uh, please on the back. Thank you.
Perfect. So, um, the instructions for accessing the accounts are very easy. All you have to do is go to join.workshop.aws and you have to input that access code that you see on the screen. I think my colleagues uh, have been around over there. You have the two uh, Solutions Architects from AWS who can help you with questions. If you have any questions, just raise your hand and we can get uh, close and we can help you out. We also have been handing over some pieces of paper that has this information so that you can get it uh, easier for copy and pasting in in your uh, laptops.
Um, one last comment uh, with regards to advanced configurations. If you want to use things like, for example, uh, configuring the settings that you are using in Claude Code manually, um, this is the kind of parameters that you will need to set up. You might take a picture now if you want. So the here is how to set up uh, the region, point it to Bedrock, choose the default model, and so on and so forth. And then on the right hand side, you also have uh, some parameters for reducing the token usage or doing rate limiting and how to enable auto reporting for telemetry and uh, things like that. You also have some information here about uh, uh, how to configure extensions in the case that you're using Visual Studio. But again, it is all about asking Claude to configure itself, or you can do it manually if you prefer through the extensions that we have there. It helps you with some things like, for example, disable the login prompt that is uh, a bit annoying every time that you start, or things like uh, uh, hiding the onboarding uh, message that you always get in Claude Code and so on and so forth.