iTranslated by AI
re:Invent 2025: GenAI Fundamentals and AI Agent Utilization with Amazon Bedrock and Amazon Q
Introduction
By transcribing various overseas lectures into Japanese articles, we aim to make hidden valuable information more accessible. The presentation we are featuring in this project, based on this concept, is here!
For re:Invent 2025 transcription articles, information is compiled in this Spreadsheet. Please check it as well.
📖re:Invent 2025: AWS re:Invent 2025 - Quick catch up to GenAI on AWS (GBL103)
In this video, Shimizu, AWS Senior Manager of Architects, explains
the theme of "Quick Catch up to GenAI on AWS." Starting with the concept of Foundation Models, he explains how they differ from traditional task-specific models by being versatile and adaptable to flexible uses. He introduces services like Amazon Bedrock and Amazon Nova, which allow users to select the optimal model from various vendors. He particularly emphasizes the importance of AI agents, explaining how agents can autonomously handle error processing that traditionally required human intervention in workflows. He also mentions the prediction that 20-30% of business operations will be replaced by agent technology by 2030. He presents specific use cases, including analyzing organizational data using Amazon Q, streamlining software development with Amazon Q Developer, from code generation to documentation creation, and modernizing legacy systems.
※ This article is automatically generated while maintaining the content of the original lecture as much as possible. Please note that there may be typos or incorrect information.
Main Content
GenAI Fundamentals and Leveraging Foundation Models with Amazon Bedrock
Hello everyone. Today's session is "Quick Catch up to GenAI on AWS," and thank you for staying with us until the end.
First, let me introduce myself. My name is Shimizu. I'm a Senior Manager of Architects, and I'm in charge of customers in social infrastructure. My hobby is making things, and as you may have seen at AWS Summit Tokyo, I created things like a robot arm that automatically plays the piano based on what was captured by a camera. By the way, I'm very grateful that this was well received at the AWS Summit Tokyo venue as well.
I'd like to dive right into this session, but first, I want to touch briefly on the broader framework of generative AI. Regarding the broader framework of generative AI, first there is deep learning, and within that, there is Generative AI. These foundation models that continue to learn using terabyte-scale data and macro data, and have trillions of parameters, are exactly what we call foundation models.
What are these foundation models? Traditionally, models were created by training them using labeled data, but they were limited to specific tasks. They were limited to things like text summarization, writing some text, or specific content.
On the other hand, these foundation models are not limited to specific tasks, but can be used for general purposes, so they can achieve flexibility closer to humans. These foundation models can train neural networks with various types of data.
For example, if you train a neural network on text, you can create applications that are ongoing and digital, or if you train a neural network on text, you can create applications such as text summarization and meeting minutes generation. You can train it on various types of data such as text, sample program code, images, and videos.
For example, this is about generating video from text. By providing information in the form of a cinematic dolly shot filmed on a diner table with a juicy cheeseburger, melted cheese, french fries, and cola, you can generate a video like a hamburger.
There is a service called Amazon SageMaker, and for example, you can decorate areas using things like Inference and APIs. Depending on your skills, you can use models, and there is this service called Amazon Bedrock. Within this, it's already something that you all commonly use.
By using Amazon Bedrock, you can easily use various models. Not just Amazon, you can select the optimal one from models of various vendors. Each model has its own characteristics. For example, some are large, some are small, and some can be customized. You can choose from perspectives such as security, privacy, and how to use them with existing systems.
That being said, tasks are not that simple, and there are also agents for solving complex tasks. These can handle complex requirements. There are also multi-agents, so I would like to explain about those later as well.
So, how do we select a model? We need to choose the appropriate one from various models. First of all, there is pricing, so for example, you can choose from inexpensive to expensive options, including models that can handle more high-performance fields.
Also, among the models to choose appropriately, Amazon's own models are included. This is it. It's a model called Amazon Nova. You can select models that can handle various tasks, from low spec to high spec, with various specifications. It also supports images and videos, and this is the latest model from Amazon Nova.
The Potential of AI Agents and the Introduction of Amazon Q
First, let me introduce some of the features from Amazon's various offerings. For example, here you can do things like an assistant, such as summarizing text, or generating images. I think you can compile the highlights that were generated, but here on the left side of the image, there is something called an assistant.
Also, human intervention is still necessary. In that case, if there is some kind of error, if the assistant cannot make a judgment, you cannot proceed immediately without human support. What I'm showing you here now is the agent on the right side.
This agent can build advanced systems themselves. Agents can do more advanced things. First, they identify tasks in various places and organize what needs to be done. By building it this way, those tasks can be processed consecutively.
In conventional workflows, processes inevitably had to proceed according to human-crafted logic and rules. Therefore, if an error occurred, the process would stop right there, and human support would be required.
Indeed, when someone comes, support is needed to recover from the situation. On the other hand, AI agents can interact with their environment more flexibly. For example, a common scenario is that when a problem occurs due to an error, instead of just sitting and waiting, the agent itself can grasp the situation, and my agent can call the appropriate API to provide a better restaurant or a better option.
And such error reports can be sent, and it is said that the opportunities to block such errors are now accelerating with the introduction of agents. By 2030, according to some estimates, 20% to 30% of business operations will be replaced by such agent technology. Also, by 2028, 15% of daily tasks are expected to be handled by such agent technology. This report came out three months ago. And now, to quickly promote the use of such agents in your organization, the first thing I'd like to introduce is Amazon Q.
First, I believe a lot of data is accumulated across your organization, and Amazon Q makes it easier to leverage and analyze this data. This quickness includes a major AI agent feature. This means it allows you to work alongside it as a team member, as a colleague. Furthermore, it securely protects private data and provides these as the best answers. And the agents I will introduce next allow you to create customized ones as you like. And with just a click, you can now integrate with external systems, which was previously very difficult. Moreover,
I will also introduce Amazon Q. This can also automate complex business processes. And furthermore, it can foster team collaboration. By utilizing such agents, you know, this is very complex, but I think it's very useful today, so I'd like to tell it as such a story. There are those options.
Practical Application of Generative AI Technology in Software Development
Those options are, look to the left. The traditional way of developing AI solutions is like this: you teach it how to write, like sit down or stand up, or write codes, and you create something that 'takes it'. This involves a human making a judgment and spending time working on that program.
Uh, testing and bugs, summarizing, costs, budget, security, whatever, summarizing, stake, code analysis, code generation, testing, and also documentation, energy release, techniques, can be utilized.
I will introduce the practical application of generative AI technology. So, let's promptly introduce some stories about software development related to AI development. First, regarding the services provided by Amazon Developers, we recommend it when you want to ask if there's any support available. It means you can perform tests on the codebase and use various functions.
Then, regarding Quick Code, there are also various agent functions. Some use these as an IDE for the development environment. Until now, software development sometimes required coding, so frequent typing errors were common. Quick Code can elevate this level to not only prototypes but also production development.
Regarding agents, by using them to define tasks before starting development, you can reduce development costs. You can literally take code from the
on-type to the production level. Also, there are stories about such features, so you can automatically create documents. Although creating user documents can be difficult, this can solve that problem.
And the third is Transform. This is about using generative AI as a culture, focusing on modernizing the legacy world and security services. I introduced this in this session, but please take another look at the portfolio. You can use Amazon services like Amazon Bedrock.
There are various services including Amazon Quick for agents. I'd like you to take a look at what we've transformed using Amazon Q all together. There's a lot of information available, so I think some of you may not have seen it yet, but please do search for it. We've compiled the story using slides, so please take a look.
And finally, we'll be holding a Japan session from 2 PM to 4 PM on Thursday, so please join us. We've compiled a lot of information in Japanese, so we hope you'll participate in the session. It will be a session in Japanese. So with that, I've introduced everything. Thank you very much for today.
※ This article is automatically generated using Amazon Bedrock, while maintaining the information from the original video as much as possible.
Discussion