#731: AWS News: Kiro, Amazon Bedrock AgentCore, and Lots More

0:00This is episode 731 of the AWS podcast, released on July 28th, 2025. five. Hello everyone and welcome back to the Adibus podcast. I'm Liz Sheebly. Great to have you back. And I'm joined by my co-host Jillian Ford. G'day Jillian. How are you doing? Always awesome to be here, Simon. I love that you've got like the old school t-shirt with like new school topics, which we're going to be talking about today. For those who are listening, of course it is an audio show. I'm wearing a The Goodies podcast, which is a 70s UK-based comedy show that was very popular in Australia and the UK, but not much else.

0:38So it's always fun to wear the goodies t-shirt. We're not joined by Shruti today. She is unavailable, but because we have three availability zones and two are functioning, we can carry the load just fine. And my goodness, there's a lot to talk about. Oh, yeah. Lots of good stuff, both on the AI side and the non-AI side as well, sort of comes in two categories these days when you're talking about IT. But Jillian, so one of the first things off the rank is a new development experience called Kiro, K-I-R-O. When I first read it, I called it Kiro, but I'm told it's called Kiro. And it's a IDE-based development experience.

1:13And I've been using it for a little while, actually, myself. And I'm really enjoying it because it takes a different approach to software development being augmented by AI. It's less of a autocomplete ask for answers type thing. and far more of a create me a set of requirements based upon a set of prompts, iterate on those and then do the technical design based on those requirements and then go and build. And I think as we've been talking about generative AI and its use in software development, the creation of the spec is actually far more important than it's ever been before because the machine will go do what the machine is told and if the spec is not great, you're going to get not great.

1:56If you get a really robust, clear spec, you'll be able to build really well. So this is kind of interesting. I don't know, Jillian, what you've seen and what you've been experiencing with it. Yeah, I think it's really cool that because in the past, like generative AI has been incorporated into IDEs based on how we've been programming pre-generative AI. And I like that the Kiro team has thought about, well, how can we really integrate generative AI as part of the entire development experience? And what does that look like? And I think that is something you'll see as a difference between using Kiro versus if you were to use whatever your favorite IDE is.

2:37And then maybe just use like your like maybe an LLM or whatever to help doing your coding assistant. So I think that's really cool. Like MCP servers, of course, is something that you'll see. The agentic chat, which is definitely something that you're probably already used to if this is something that's already integrated in your workflow. I think the hooks is also really interesting. So these are agent hooks that are going to help with automation as part of your like integrative development experience. But now it's a GenTech type of hooks versus maybe just automation that you're probably used to already in your development.

3:19Yeah, it's all really, really similar stuff. But the nice thing is I was a VS Code user. And if you're a VS Code user, the migration into Kero is literally one click. Complete compatibility with all your plugins and everything else. It's all easy, which is great. So it's a very familiar experience. you just get like some extra capabilities like oh this lets me build quicker which is what i'm always looking for and better i like it now there was also more and i think quite a substantially important uh iteration in the world of generative ai agents um just recently and this is in the form of amazon bedrock agent call and you took a bit of a deep dive on this jillian tell us about what you see in this new capability kind of what it is and where it fits yeah this is super cool and i think it's going to really uh change how people think about building ai agentic applications overall at scale um so let's talk about what it is um so this is going to help really streamline the process of taking your ai agent applications that you're currently building in your proof of concept stage and then getting it into production, whether that's how do you set up observability?

4:41How do you make sure that it's secure? How do you make sure that reliability is also included as part of the overall application? And you can use a bunch of different open source frameworks that maybe customers have already started to use. You can now use that within AgentCore and it's going to be hosted within the Amazon Bedrock ecosystem. So it just makes it much simpler to go from proof of concept to production. It does. One of the things that really appealed to me too, I guess, from an enterprise user perspective is the fact that there's this thing called the agent call gateway, which really simplifies that tool integration, that discovery approach and understanding what's going on.

5:25This seems to be the pattern, I guess, that's emerging in terms of, well, how do we let all these agents go do this stuff, but also maintain some degree of control and understanding of what's going on? Yeah, absolutely. And some other things also, in addition to the security aspects that really come to mind, I like that you've got short-term and long-term memory as an option, and long-term lives literally eight hours. And so if you can think about like, wait, eight hours, there's a lot of things you can do in an application, a workflow, if you're having eight hours as a memory context and then hundreds of tools.

6:01So the ideas of what you could possibly build are almost endless. Yeah, I think it's exciting. Speaking of exciting, let's get into some of the other updates we've had. Firstly, for the AWS Marketplace, speaking of agents, we are now introducing AI agents and tools in the AWS Marketplace. So this lets agents and tools from AWS Partners be purchased from the Marketplace. So it's a streamlined procurement and deployment experience, and you can find what you need more quickly and get up and running fast. So again, this is a new way to, I guess, purchase and consume these capabilities. And again, it ties into, I think, where things are moving in the future.

6:41And you can also, as a partner, you can categorize your offerings and highlight MCP support, A2A protocol support, et cetera. So as people are growing out in what they're doing, and this all integrates into Amazon Bedrock Agent Core as well, it becomes easy to pick and choose what makes sense for your business. Now we've got some updates in the world of analytics. AWS Glue now supports zero ETL integrations from Amazon DynamoDB and eight applications to S3 tables. And so for those who are new to the zero ETL, I'll do a quick primer. So zero ETL is what it sounds like. That means you're not actually doing the ETL.

7:24AWS has created these integrations with services. services. So data that starts in point A can easily be viewed, read in point B. So let's say like RDS, for an example, maybe you have a Redshift data warehouse. And fun fact, there's even a landing page. If you want to look up zero ETL, it's amazing the number of different zero ETL integrations that are out there within AWS. So I definitely suggest taking a look at that as you're thinking about ways of being able to move and share data since there might be zero ETL integrations that are built and you don't have to actually do that yourself. So now how does this translate into what we're talking about today?

8:09So with DynamoDB and AWS Glue and especially Amazon S3 tables, you can now automate the extraction and loading the data into S3 tables from DynamoDB and other applications like salesforce sap surface now and zendesk um so now uh with the zero with uh you can enable s3 tables to also work with lake formation to support other services as well athena emr red shift um and glue so yeah this is definitely a time saver lots of integrations lots of lots of for sure. And more MCP. We're also announcing Amazon MSK has MCP-based server that allows customers to interact with their Amazon MSK clusters using a standardized natural language interface and agentic applications.

9:08Amazon Redshift now supports automatic refresh of materialized views that are defined on external Apache Iceberg tables in the Amazon S3 data lake. So with this update, Amazon Redshift is going to automatically refresh a materialized view when defined on Apache Iceberg tables that reside in the Amazon S3 bucket of the data lake. Amazon Redshift now supports cascading refresh of nested materialized views that are defined on local Amazon Redshift tables and external streaming sources such as Amazon Kinesis Data Streams, Amazon Managed Streaming for Apache Kafka, or Confluent Cloud. With this update, customers can now run cascading refresh of nested materialized views with a single option to specify cascade or restrict.

10:00And now we've got application integration. We also have the MCP server for the AWS API. I can only imagine the amount of things that you could probably go wild with, with this type of control. So literally the AWS API MCP server allows MCP clients to discover supported AWS APIs and make calls to them through the host of foundation models, enabling actions such as inspecting, creating, and modifying AWS resources. Like, wow. That is like... Yeah, let's be careful on that last one, folks. Or else you're going to hear from Simon. The phrase dry run is always an important one. But don't worry, because the server provides secure access through IAM credentials and is pre-configured with API permissions.

10:58So that does ensure that foundation models can only access or perform authorized actions on permitted ADOS APIs. But definitely still agree with Simon. You definitely want to be wise. Scott damn. Yes, with that one. Amazon EventBridge now supports logging to Amazon CloudWatch Logs, Amazon S3, and Amazon Kinesis Data Firehose, improving observability and simplifying debugging for your event-driven applications. Amazon Coreto has announced quarterly security and critical updates for long-term supported and feature release versions of OpenJDK. Let's talk about artificial intelligence, and we're happy to introduce the new AWS AI League.

11:49This is a program that helps organizations upskill their workforce by combining a fun competition with hands-on learning using AI services like Amazon SageMaker AI and Amazon Bedrock. The program offers unique opportunities for both enterprises and developers to get valuable and practical skills in fine-tuning, model customization, and prompt engineering. Now, you can apply to receive AWS credits to host internal AWS AI League competitions, which helps foster a culture of innovation. Individual developers can also participate in the league at select AWS summits and at reInvent, which means you can compete with lots of others.

12:25There is up to$2 million of AWS credits and a championship prize pool of$25 ,000 to reward top performers at AWS reInvent. So get a look at it. I know that when folks used to do the Deep Racer League, they got a lot out of it. This is kind of the next version of that in the brand new world. image to video generation support for lumas ai ray 2 is now available in amazon bedrock this new feature expands upon the text to video generation capabilities that came out in january which gives you even more powerful tools for creating dynamic video content you can now transform static jpeg and png images of up to 25 meg into videos wow i've got to try it okay this is cool this is like you know because i always have trouble starting the picture but if i have the picture, I can then make the video from the picture.

13:14It's nice. I like it. Well, we can create like image to videos of like us, the podcast hosts. Yeah, let's not do that. That's a poor use of the technology I would suggest. Some new foundation models available in Amazon Bedrock 12 Labs, Marengo 2.7 and Pegasus 1.2 multimodal foundation models are available. Marengo 2.7 is a video embedding model proficient at performing tasks like search and classification. While Pegasus 1.2 is a video language model that can generate text based on your video data. Amazon Nova Sonic has added language support for French, Italian, and German. Now this is a speech-to-speech foundation model that delivers real-time human-like voice conversations with low latency.

13:58I know folks who've used this are like, it's pretty cool. So now it supports more languages. We're announcing on-demand deployment for custom Amazon Nova models in Amazon Bedrock. So this enables Bedrock customers to reduce their costs by processing requests in real time without requiring pre-provisioned compute resources. You can now customize Amazon Nova in Amazon SageMaker AI. So available as ready-to-use recipes. This allows your customers to adapt Nova Micro, Nova Lite and Nova Pro across the model training lifecycle, including pre-training, supervised fine-tuning and alignment. Amazon SageMaker has announced integration with Amazon QuickSight, so you can now launch your Amazon QuickSight directly from the Unified Studio.

14:44Amazon SageMaker Catalog has added support for Amazon S3 general purpose buckets, so this makes it easy for you to discover and access datasets. And Amazon SageMaker HyperPod has accelerated OpenWeight's model deployment. So this lets you seamlessly train, fine-tune and deploy models on the same HyperPod compute resources, which means you maximize your resource utilization across the whole model lifecycle. Building models is hard and it's a big deal and there's a lot of computing. I know HyperPod helps a lot of folks working in that space. And related to that, a few updates for HyperPod in fact.

15:20We're announcing a new observability capability. So now you can understand what's going on and take away the manual work of collecting hundreds of metrics from across the stack. And HyperPod has also introduced CLI and SDK for AI workflows. So you can now build things out in a far more intuitive and automated way. Amazon SageMaker Studio now also supports a remote connections from Visual Studio Code. So if that's what you like to use, also things like JupyterLab and Code Editor based on code OSS, you can get access to that. And Amazon SageMaker, not finished yet. They've introduced a visual workflows builder.

15:59So this is a drag and drop interface for building workflows, which makes authoring and scheduling much easier. And Amazon SageMaker also now supports data processing jobs. So you can author, manage, monitor, and troubleshoot your data processing workloads as well. And it is also now simplified data management with automated lake house onboarding and metadata ingestion. So you can automatically ingest your metadata for your data sets, things like Blue Data Catalog Tables into your SageMaker Catalog. So there's no manual IAM permissions, undifferentiated heavy lifting goes away. SageMaker has also streamlined the S3 Tables workflow experience.

16:38So it makes it easy for you to query, create and join Amazon S3 Tables with data in S3 general purpose buckets. and fully managed ml flow 3 is also now available on amazon sage maker ai so this allows you to do more experimentation and it transforms your managed ml flow from an experiment tracking to providing end-to-end observability which reduces your time to market and makes it easy to figure out what's going on in the experiments that you're running so lots of updates there a quick update for business applications, custom integrations is now available for topics in Amazon Q in QuickSight. Custom instructions enable author professionals to curate Amazon Q's responses to questions by adding domain-specific knowledge that can't be captured through a topic's metadata settings, like synonyms or semantic types.

17:33Next up is compute. AWS Deadline Cloud has expanded support for Unreal Engine in its service-managed fleets. With this new feature, you can submit Unreal Engine 5.4, 5.5, or 5.6 projects to Deadline Cloud for rendering without needing to configure or manage compute infrastructure. AWS Deadline Cloud now supports usage-based licensing for Chaos V-Ray, so you can seamlessly leverage the cloud to render and access flexible V-Ray licensing. And we've got a new update on the free tier. AWS Free Tier now offers$200 in credits and six-month free plan to explore AWS at no cost. I think we need a sound for this one.

18:19I had no warning that a sound was required. Let me see if I can do this. There you go. That's the best I can do in short notice. All right. This is a big change to the way the free tier works. And I think it gives you a lot more flexibility because you get your credits and you can do what you want rather than having to kind of track what each particular service has a free tiering. That's a great call out. And I also want to call out that Amazon Bedrock is included as one of the services that you can use these$200 in credits for. And that goes a long way when you're just building a proof of concept for an AI application.

19:04Not to bring up AI again. but exactly and yeah and the way it works is you know you get a hundred dollars in credits straight up when you sign up and then you get an additional hundred dollars in credits if you use services like amazon ec2 or amazon so you get sort of the way to go there also there are still 30 always free services so these are services that are always free we'll not list them for you it's in the the free plan details but it means you can get up and running and do meaningful work very, very quickly. And yes, it's available in all eight of us regions. Exactly, except U.S. GovClad regions and the China regions.

19:40Good call, good call. And Amazon EKS announces support for up to 100 ,000 worker nodes in a cluster, enabling you to run ultra-scale AI ML training and inference workloads in a single cluster. That's a lot. That is a lot. Let's talk about databases. Amazon DocumentDB with MongoDB compatibility now introduces support for up to 10 secondary region clusters. So that's awesome if you're building anything that needs scalability, availability, and is globally distributed. So this gives you disaster recovery from region-wide outages, and it gives you fast local reads for globally distributed applications.

20:19So previously we had a limit of five, now it's up to 10. So twice as good as it was before. Amazon RDS Custom for SQL Server now supports change data capture. Amazon RDS for SQL Server also now supports the cumulative update 19 for SQL Server 2022, which allows me to remind you to, let's see if I can get this right, haven't done it for a little while, pack your stuff. Sounds a little different this time. There was a bigger echo before. I know, that was the megaphone, whereas I could have done this one. Oh, that's a new one. I like that. I think that one will definitely creep people out to get them moving.

21:06Now let's talk about the Internet of Things. AWS announces the release of AWS IoT Greengrass version 2.15, introducing significant updates to both Nucleus and Nucleus Lite Core components. I don't know what that means, but I'm sure if you're probably, if you're an IoT person, you probably are in the know and know what that actually means. And that's the thing. And these are all about resource constrained devices. So we're talking five meg RAM type devices. So that's funny. On the one hand, as IT professionals, we get to work with like these machines that have got extraordinary amounts of storage on them now.

21:42And on the other hand, there's a whole bunch of folks who are like, yeah, I've got five meg to work with here. Throw me your bone. Let's talk about management and governance. AWS cost anomaly detection improves accuracy with model enhancements. Now, if you're not using AWS cost anomaly detection, and I do, it gives you a heads up if things are happening that are out of traditional behaviors. Now, this update provides you with a more consistent and reliable cost monitoring while handling historical cost variation. So it understands the patterns that your organization typically goes through. we're happy to announce the preview of the adibus knowledge model context protocol mcp server this is a new tool that surfaces authoritative adibus knowledge in an llm compatible format so you're getting documentation blog posts what's new announcements and well architected best practices and you don't have to worry about keeping it up to date because you know it's always up to date so this is going to be a go-to for me in terms of my little mcp collections that I'm going to run.

22:42The AWS price list API now supports four new query filters. So you can search product data more easily. These new filters help simplify product discovery, which means you can get up and running quickly. So you can now search products by exactly managing attributes, using substring matches and creating include and exclude lists for more targeted results. And if we weren't doing enough in the world of price lists, we now have an MCP server for the out of this price list. Can you get a trend going on here, folks? The world is MCPing itself. So now you can leverage your AI assistance to access pricing and product information across regions and make data-driven decisions all through natural language conversations.

23:24I'm excited to use this one too. And Amazon CloudWatch adds generative AI observability in preview as well. So this gives you end-to-end prompt tracing of components like knowledge bases, tools, and models. This is compatible with things like strands agents, LAN chain, LAN graph, etc. So it can use the flexibility of choice. But it means you get to understand telemetry across these increasingly complicated systems as well. Next up is migration and transfer. ADOS DataSync now supports IPv6. The year of IPv6 continues, Gillian. It really is. ADOS Transform from Mainframe introduces enhanced code refactoring and business logic capabilities.

24:08The reforged capability in Adibus Transform for mainframe, this is going to enhance Java code by restructuring complex methods and it's going to add descriptive comments, optimize variable usage, and improve the overall code flow. So this is going to result in more readable and maintainable code for developers. And Adibus Transform for mainframe's business logic extraction capability now provides application-level insights from high-level summaries to detailed business function analysis, which is going to complement the existing file-level business logic extraction to help users better understand their legacy applications.

24:51Like that. Nothing worse than not knowing what's going on in your application. What is this thing doing? Yep. Next topic is networking and content delivery. Amazon VPC CNI, or Container Network Interface Plugin, now supports higher bandwidth and network performance per pod. So this is really useful for AI high-performance computing use cases. And this interface is going to handle all of the incoming and outgoing traffic per pod. So if you need to scale a network performance for your Kubernetes-based workloads, this is definitely something that you should check out. Now let's talk about storage.

25:39Amazon EBS now provides visibility into EBS volume initialization status. Oh, this is a cool one. I'm very happy. You can use this status to determine when your volume becomes fully initialized when restoring from a snapshot and is fully ready to support latency-sensitive applications. So EBS volumes that are created from EBS snapshots undergo volume initialization in which the storage blocks from the snapshot must be downloaded from Amazon S3 and written to the volume before you can access them. Now the volume initialization rate fluctuates throughout the initialization process which can make completion times unpredictable.

26:14So sometimes you get increased IO latency and reduced performance. Now you can see where you're at using this status. This is a big deal for me as an old timer Amazonian cloud user, because we'd always tell folks, you know, if you're using an EBS volume that's created from a snapshot, it's kind of lazy loading in the background. So things can be unpredictable. Having visibility is really cool. Amazon S3 metadata now supports existing objects and reduces prices by up to 33%. So now you can discover data that's in your S3 data. And previously S3 metadata supported new and updated objects. Now it will also create and manage metadata for your existing S3 data.

26:59So you can write a SQL query across metadata for any amount of S3 storage. As I've always said, folks, SQL, learn it. You just get to use it for everything. The other thing is that we're reducing the journal table price by 33 % to make real-time change tracking and backfilling more cost-effective for large data sets. So the cost reductions continue. And then more of those to come in a second. Amazon S3 Tables, though, now supports, guess what? Model Context Protocol MCP Server. So this integration lets AI-assisted data management happen because now your AI code assistants have a contextual understanding of S3 table capabilities and operations so you can accelerate what you're doing.

27:43I mentioned there'd be more cost-effective stuff, well guess what? Amazon S3 tables now offers more cost-effective compaction operations for Apache Iceberg tables with processing fees reduced by up to 90%. S3 tables provide storage that's optimized for analytic workloads which automates maintenance operations like compaction to continuously improve query performance and reduce storage costs. with these compaction price reductions the per object price is now 50 percent lower while the per byte processing prices are 90 percent lower for bin pack compaction and 80 percent lower for sort and z order compaction so this is pretty nice amazon s3 also now supports compaction of apache avro and orc formats for apache iceberg tables and another big deal is amazon s3 vectors is now available in preview.

28:36This is the first cloud object storage with native support for storing and querying vectors. Now, this is important because it reduces the cost of uploading, storing, and querying your vectors by up to 90%. So this allows it to be cost-effective to create and use really large vector data sets. It's designed to provide the same elasticity, scale, and durability as Amazon S3, but this lets you store and search data with sub-second query performance. S3 Vectors gives you a simple and flexible API for operations like finding similar scenes in petabyte scale video archives, identifying collections of related business documents, or detecting rare patterns in diagnostic collections of millions of medical images.

29:19So this is really, really cool. It's natively integrated with Amazon Bedrock Knowledge Bases, so you can reduce the cost of using large vector data sets for RAG type work. You can also use it with the Amazon OpenSearch service to lower your storage cost for infrequently queried vectors and then quickly move them to OpenSearch as demands increase or to enhance your search capabilities. Lots and lots of choice here. I was speaking to someone the other day and they said to me, Simon, this is where the vector should have been in the first place. It's hard to disagree. And finally, the Amazon S3 console now displays an external access summary for all your buckets.