#748: re:Invent 2025 - Matt Garman Keynote

0:00This is episode 748 of the AWS podcast, released on December 3rd, 2025. Hello, everyone, and welcome back to the AWS Podcast. Simon Lynch here with you. Great to have you back for our very special set of reInvent editions. And of course, I'm joined by Gillian Ford. G'day, Gillian. How are you doing? It is the most wonderful time of the year here. Indeed. Indeed. And unlike last year where we sort of took turns at doing these episodes, we're doing them together this year because that way we get to compare and contrast the things we think are cool. And were interesting. And as well as that, a call out.

0:40So this episode will focus on Matt Garman's keynote and all the cool things that came out during that keynote. We have two more episodes coming up this week that will include the other keynotes. We'll also be augmenting one of those episodes with kind of the pre-invent catch up of all the things that have been released during pre-invent as well. So you won't miss out on anything. So we've got our work cut out for us, Gillian. We really do. It is a lot, but I'm always excited for it. And it's always interesting to see what appeals to people and which things are interesting. So we're going to seek to try and unpack some of that during this episode, but let's get started.

1:18Firstly, we're going to start with AWS Clean Rooms. And AWS Clean Rooms has launched a privacy-enhancing synthetic dataset generation for ML model training. Now, obviously, training is really, really important, but also maintaining privacy of data sets that you may be using, particularly internally, is really, really important too. So this feature generates synthetic training data sets that preserve the statistical patterns of the original data without the model having access to original records. So this is really a new thing that I think it's going to be interesting to see how customers choose to use this in terms of being able to better secure their training data sets whilst maintaining really cool accuracy.

1:58Now, of course, unsurprisingly, there've been a huge amount of announcements related to AI and Gen AI and all those cool things. So let's run you through them. So we're also announcing Amazon Nova 2 Lite, which is a fast, cost-effective reasoning model for everyday workloads. So it gives you interesting leading performance. It gives you a really effective price point. And it now also supports extended thinking, including step-by-step reasoning and task decomposition before doing anything. It supports text, image, video, documents as inputs, and you get a 1 million token context window. That's really interesting because I'm sure many of you are finding yourself running out of context.

2:38And you can also customize this as well. And it also has two built-in tools, web grounding and code interpreter. Web grounding will give you publicly available information with citations and the code interpreter lets the model run and evaluate code within the same workflow. So this is a really interesting new one to take advantage of. Now we're also introducing Amazon Nova 2 Sonic, which is our new speech to text model for conversational AI. Now this gives you natural real-time voice conversations with your applications. And this delivers expressive voices, masculine and feminine voices in each of the supported languages.

3:16So lots of new languages. and it has natural expressivity, turn-taking, a seamless handling of user interruptions as well. And it shows that people, when we test this with customers, they actually like how it feels and how you interact it. There's much better understanding. The model now handles alphanumeric inputs, short utterances and 8 kilohertz telephity speech input as well. And it also now has English, French, Italian, German, Spanish Portuguese and Hindi as well and it can switch between languages in the same conversation which is something I can't do so it's a lot smarter than me this is really cool Jillian I think you've had a lot of customers who sort of been looking for this extra capability yeah I'm super excited for this I see customers especially in the healthcare industry who want to be able to add voice interactions like the patient doctor type of interaction so I think This is just going to be super helpful, especially when they have patients who are multilingual.

4:19If you're in the mood for training your own models, you can now build your own frontier models using Nova. This is called Amazon Nova Forge. And this allows you to have supervised fine tuning and reinforcement learning on top of these existing models. And this allows you to really start your development from early model checkpoints. And then you blend your data sets with the Amazon Nova curated training data, and you can host it all in AWS and you can build things out. So you can see how that anonymized data set earlier on comes into play. Now, there's a lot that's going on. So we're going to move through some of them, but we're also going to dive deep on some.

4:55And one that I think is really interesting is Amazon Nova Act is now generally available. And this allows AI agents to interact with user interfaces and automate complex workflows. And developers had been experimenting with this and they said, hey, we need this in production. but you can't just say, hey, give it access. You've got to do some stuff. So now basically this provides you with a complete interface workflow ability at creating this in a isolated way that you can then deploy into your environment. So I think this ability to interact with the web, if you like, is kind of important, Gillian.

5:35I think so too. And I also really like that it has this no code web UI that I think, especially as more teams are really thinking about evaluation-driven development and including people that are domain experts into the entire AI agentic process, having a really nice UI that just makes it really easy to be able to get started with making those browser types of automations and makes it a lot easier for them to be able to work with those AI agentic developers. Yeah. And doing it at scale too is always the trick. Now, speaking of trust, Amazon Bedrock AgentCore has added quality evaluations and policy controls for deploying trusted AI agents.

6:22This is huge because people want to not just deploy this technology, but deploy it with trust and assurance. So policy in AgentCore is in preview and it defines clear boundaries for agent actions by intercepting AgentCore gateway tool calls before they run using policies with fine-grained permissions. And then there's agent core evaluations. That's also in preview. And this monitors the quality of your agents based on real world behavior using built-in evaluators for dimensions like correctness and helpfulness and custom evaluators for your own business specific requirements. So this really allows you to have greater insight and control of what's going on.

7:00There's also some new features that are expanding what agents can do. So there's episodic functionality in agent core memory. So this is a new long-term strategy that lets agents learn from experiences and adapt solutions across similar situations for improved consistency and performance in similar future tasks. This is huge because if you've ever used an agent, it's like they have a complete memory wipe every time they run. They don't learn anything. They don't know anything new. This is kind of starting to address that. The other thing that's really interesting is bi-directional streaming in agent call runtime.

7:33So this deploys voice agents where both users and agents can speak simultaneously following a natural conversation flow. And this is really nice because I was actually planning to build something related to this. So I'm going to use this functionality. Now, vectors are a thing. I think, Jillian, we see vectors everywhere. You've seen this with your customers a lot. And Amazon S3 Vectors is now generally available and it has increased scale and performance. Now, how much scale, you may ask? Well, you can now store and search across up to 2 billion vectors in a single index, and up to 20 trillion vectors in a vector bucket, my goodness, and a 40x increase from the 50 million per index during preview.

8:19So, it means basically you can have one Hong Kong-grade index, which is kind of amazing. And query performance has been optimized, so infrequent queries continue to return results in under one second, with more frequent queries now resulting in latencies around 100 milliseconds or less. And you can also retrieve up to 100 search results per query, which is up from the 30 before. The write performance has also improved substantially, support up to 1000 put transactions per second when streaming single vector updates into your indexes. So you can really crank this up. And it's all serverless. So it's kind of a nice quality of life improvement and really making this robust for production.

8:57There's very few things in terms of like architecture patterns where you get like lower cost, lower latency, higher throughput. And I mean, this is just one of those, I think you just, it checks all the boxes. Yeah, yeah, it really does. And the, I don't have to do anything to get the benefit of it is always a big tick. Now, of course, there's always lots going on in Gen AI. And as we've noticed throughout the development of this technology that there's not one model to rule them all. You need to pick your model carefully, depending on your use case. And Amazon Bedrock has now added 18 fully managed open weight models, including the new Mistral Large 3 and a bunch of others, which I'll chill you about, into Bedrock.

9:38So what do you get? So you get new Mistral AI models, you get Mistral Large 3, Ministral 3 3B, which is a new one for me. Looks like a typo, but it's not. Ministrel 3 8B and Ministrel 3 14B as well. Also, you get from Google, you get Gemma 3 in three different model weights. You get Moonshot AI, you get Minimax AI, you get obviously NVIDIA as well, OpenAI and Quinn. So there are now over 100 serverless models available to you in Bedrock. And this is the power of Bedrock in that same API, same way of interacting, but you get to use the models that make sense for you. I'm really excited about this, Jillian, because I know about you, but I'm seeing a lot of customers sort of have to choose lots of different models for different tasks.

10:29And often it's a price performance tradeoff. Yeah, absolutely. I think more and more customers going into next year are going to really be wanting to experiment with different models that are out there. So model choice is always a win. absolutely another announcement is we're announcing the ability to use amazon sage maker ai with serverless ml flow so this is a serverless capability that eliminates infrastructure management and it transforms your experiments tracking into an immediate on-demand experience automatic scaling that means you don't have to do capacity planning so it's a kind of a zero infrastructure approach to sage maker ai

11:12let's talk about compute i don't think it would be a reinvent launch if we didn't talk about new instances it would make us sad it would definitely make us sad it really would yes um and we've got new memory optimized instances These are the EC2-X8AEDZ instances, powered by 5th Gen AMD EPYC processors for memory-intensive workloads. They deliver up to 5 GHz processor speeds and 3 TIB of memory. So these are going to be ideal for electronic design automation workloads and memory-intensive databases requiring high single-threaded performance. this one is kind of like a like a mind trick i feel like this one because we've got aws lambda managed instances which are like wait a minute like there's serverless there's ec2 there's no manager what does this mean don't worry i'm i'm we're also figuring this out together so the reason why we have this is it wanted to address some customer needs around accessing specialized compute options and also doing so in a way where you could optimize costs for steady state workloads without sacrificing the serverless development experience that you get when you're used to using Lambda instances.

12:44So you kind of get the best of both, really. So these are Lambda managed instances. I know I need to say this slowly because it is like a lot to wrap your head around if you're used to like picking one side or the other. So you get access to the latest generation of EC2 instances, and AWS handles all of the operational complexity, instance lifecycle management, OS patching, load balancing, and auto scaling. So this means that you can select compute profiles that are optimized for your specific workload requirements, like high bandwidth networking for data intensive applications, without taking on the operational burden of having to manage EC2 infrastructure.

13:27Another one that is really cool, it's like a mashup of like Lambda functions and like step functions, AI. So now you can be able to build multi-step applications and AI workflows with AWS Lambda durable functions. So this allows you to build these reliable multi-step applications. So those who are maybe familiar with like Lambda or like the step functions types of experience for these workflows. So now these durable functions that are regular Lambda functions with the same event handler that you're already used to, but you can now write sequential code in your preferred programming language and the durable functions are going to track the progress.

14:12They're going to automatically retry on failures and suspend execution for up to one year at defined points. That's a lot of suspension. Yeah. Yeah, this really is. I mean, it really, the paradigm of the types of applications that you can build now with like serverless, it really starts to bend a bit, which is pretty cool. And this is all without paying for idle compute during wait times. So that is super cool. So they use a checkpoint and replay mechanism known as durable execution to deliver these capabilities. I know all of you are like, how is this actually? How does this work? Yeah, so cool.

14:52So after enabling a function for a durable execution, you add the new open source durable execution SDK to your code. And Lambda is going to do a lot of like the heavy lifting for you. So, wow, this is super cool. So I know a lot of my serverless friends who are listening. This is definitely something you want to go and get your hands on. I think it is really interesting for those use cases where you just kind of needed a little bit of pausing. like it wasn't it was like you didn't need full step functions but you needed it to cope with a little bit of a delay or a break and this this is going to unlock a whole lot of new stuff i'm really excited too yeah and especially a lot of people who start off with lambda they love building in lambda functions and maybe they've had just some limitations before on like lambda and so now i think this really just continues to expand the use cases um as companies and like business requirements grow.

15:48So really cool. All right, next, we've got one in containers. So we're announcing Amazon EKS capabilities. This is an extensible set of Kubernetes native solutions that streamline workload orchestration. These are fully managed. They're integrated platform capabilities. So they include open source Kubernetes solutions that many customers are using today Argo CD, AWS Controllers for Kubernetes, and Kube Resource Orchestrator. So now you can build and scale Kubernetes applications without managing complex solution infrastructure. So let's talk about where your data lives, databases. Now, we've always had savings plans for EC2, but now we have database savings plans for AWS databases.

16:41So this helps you reduce your database costs by up to 35 % when you commit to a consistent amount of usage per hour over a one-year term. Savings automatically applied each hour to eligible usage across supported database services and any additional usage is just built at the on-demand rate. So this really helps you get more flexibility in your cost savings and just improving the cost profile of your workload. So definitely one worth having a look at. now let's talk about scaling and other capabilities we have new capabilities to optimize costs and improve scalability on amazon rds for sql server and for oracle so let's break this down firstly we now have sql server developer edition support this is huge because this gives you a free sql server edition that gives you all the enterprise edition functionalities specifically licensed for non-production workloads so you can build and test without paying SQL Server licensing costs.

17:39So this is a big deal for developers in those environments. You can now use M7i and R7i instances on Amazon RDS for SQL Server. This will give you significant cost savings over previous generation instances. And you can also understand the billing a lot better as well, because you can optimize the CPU capability to customize the number of vCPUs and that adjusts your licensing costs. So this is one of those financial engineering pieces. There is also additional storage volumes for RDS for Oracle and SQL Server. In fact, they now support up to 256 tibi-bytes of storage. That's four times more than they used to, which is pretty huge as far as I'm concerned.

18:20So you can store lots of stuff. I haven't met anyone with a 256 tibi-byte Oracle, or not Oracle, I have Oracle, but a SQL Server database. So that would be an interesting thing to deal with. let's also talk about the amazon open search service it has improved vector database performance and cost with gpu acceleration and auto optimization so you can build vector databases up to 10 times faster at a quarter of the indexing costs when compared to non-gpu acceleration so this means you can create billion scale vector databases in under an hour so you can get moving quicker and get up and running.

18:58There is now also auto-optimization, so you can find the best balance between search latency, quality, and memory requirements for your vector field without needing expertise. So again, better cost savings, better recalling rate, and this saves you weeks of hand-tuning that you would have had to do in the past. So big fan of that one. Gillian, we've got some new big changes in support. I know this is one close to your heart. break it down for us because it's kind of a big challenge yes this is and i know before of you before you all are just like wait i want to hear all like the engineering updates like stick with me because if you are using basic support developer support um there's just like some really cool stuff i would say here that i think can really help you sleep better at night um so there are some new and enhanced support plans.

19:48So one is Business Support Plus. So this transforms the developer, startup, small business experience by providing intelligent assistance powered by AI. You can choose to engage directly with AWS experts or start with AI-powered contextual recommendations that seamlessly transition to AWS experts when needed. AWS experts respond within 30 minutes for critical cases. One of the big things here, Jillian, is this is like response within 30 minutes for critical cases. That's twice as fast as previously, which is pretty cool. That is amazing. So if you're someone who in the past, like you had a business critical down issue, you're very smart, you would just go and try to figure it out on your own.

20:34Like there are people on the other side who probably have seen this before. So that's what I love about these support plans. Make sure you get those tickets into someone because when you open a support case, it gets routed to a subject matter expert on the actual service team. So they maybe have seen the issue that you are coming across. So that's one super cool. The next is enterprise support. This one I'm also a huge fan of. So with enterprise support, you get a designated technical account manager that's a part of your team. they do some really cool offerings to be able to really help you. So like an incident, a security incident response plan.

21:19I've just also seen them just be also really proactive with customers, whether that's cost management, helping them if they have like a really big event coming up, like Boxing Day, Black Friday, and they need to make sure that their architecture is really prepared for it. And if something were to happen, they can be able to swoop in right away. So what is also really cool is that, yes, there's now AI-powered assistants. So with up to 50-minute response times for production-critical issues, you can have support engineers if you need to be triaged like a real person. And there's also those AI agents that can also help you to be able to resolve issues faster.

21:59And the price point, I think, for both of these offerings is, I would say, it's just absolutely amazing. So Business Support Plus, that starts at$29 per month. So this is literally a 71 % savings over the previous business support monthly minimum. And then for Enterprise Support, that now starts at$5 ,000 a month. So this is a 67 % savings over the previous Enterprise Support offering. So I'm really excited for other customers to really be able to take advantage of these additional support offerings to really help them, especially in critical times where they need extra support to help them with their business.

22:37That's huge. But Jillian, what if I need more support? Ah, I'm so glad you asked. Well, there is another offering called Unified Operations Support, and this delivers our highest level of context-aware support through an expanded team of experts. So now with this core team, you get a technical account manager, a domain engineer, and a designated senior billing and account specialist. And depending on what your specific business needs are, like let's say you're going through a migration, maybe you're going through an incident management or security operations, these experts can really come in based on what your unique business requirements are.

23:20And really, so it's like really just super hands-on type of support that you can be able to get. There's also the AI-powered automation because I guess, I mean, this is the year of generative AI. This is when it all happens, exactly. And you also get five-minute response time as well with critical incidents. Five-minute response time. That's pretty crazy. Yeah, this is absolutely huge. So I'm really excited about all three of these programs for customers. And another one that also I think people have probably been waiting for is Amazon CloudWatch introduces unified data management and analytics for operations, security, and compliance.

24:06So this enhancement means that CloudWatch can automatically normalize and process data to offer consistency across sources with built-in support for open cybersecurity schema framework and open telemetry formats, so you can focus on analytics and insights. CloudWatch also introduces Apache Iceberg, compatible access to your data through S3 Table. So you can run analytics on Athena and SageMaker Unified Studio. Now let's shift into the wonderful world of migrations and modernization, because when I talk to customers, most of them, if not all of them, have some form of tech debt. And let me tell you, the bigger the organization, the bigger the debt.

24:51Some studies have shown that organizations will spend 20 % of their IT budget just on tech debt, I've met a lot of customers who dream of it only being 20%. It's often 80%. So it can be right around the wrong way. So how do we help with this? Well, we're happy to announce AWS Transform Custom. This is a new agent that fundamentally changes how organizations can approach modernization at scale. So this agent combines pre-built transformations for Java, Node.js and Python. And it also now has the ability to define custom transformations. So it means you can learn specific transformation patterns and automate them across entire code bases.

25:33So customers using AWS Transform Custom have achieved up to 80 % reduction in execution time in many cases, which frees up your developers. So what this means is that you can actually define transformations using your documentation, natural language descriptions and code samples, and then those get applied consistently across the patch. This whole thing allows you to, it's not going to just out of the box migrate everything for you with a press of a button. That's not the world today. But what it does do is dramatically compress the amount of work you have to do. And it means you can have less people working on it to get a far greater outcome.

26:10It's pretty important for any of you who have those things that have been niggling away. Be it just something as simple as, oh, I've got to upgrade all my old Python stuff. this will help you. Now, if you are living in the Windows world, I also have great news for you. AWS Transform now has full stack Windows modernization capability. So when we first announced the general availability of AWS Transform for.NET, it was great for.NET applications and customers loved that. But they said, hey, we also want to modernize SQL Server and legacy UI frameworks. So now you can do the whole shebang. So basically you can accelerate up to five times faster across applications, UI database and deployment layers.

26:57So along with porting.NET framework applications to cross-platform.NET, it migrates SQL server databases to Amazon Aurora Postgres SQL compatible edition with intelligent stored procedure conversion and dependent application code refactoring. For validation and testing, it also deploys the application to EC2 Linux or ECS and provides customization of cloud formation templates for production use. And it also has added capabilities to modernize ASP.NET web form UIs to Blazor as well. There's so much into this. If you're a migration geek, this is something to look at. Now, as I've mentioned many, many times in this podcast is I started my career working on mainframes and I always thought it would be my sort of superannuation retirement plan.

27:46I could be one of those grizzled old people who knew how to use COBOL, etc. But I'm in trouble because AWS Transform for Mainframe has introduced reimagined capabilities and automated testing functionality. So what is reimagining mainframe modernization? So this is a new AI-driven approach that completely reimagines the customer's application architecture using modern patterns or moving from batch process to real-time functions. So by combining the enhanced business logic extraction with new data lineage analysis and automated data dictionary generation from the legacy source code, customers can transform monolithic mainframe applications written in languages like COBOL into more modern architectural styles like microservices.

28:28This is important because building COBOL applications and the way we would build mainframe applications was completely defined by the processing model and modality of the mainframe. So just lifting and shifting to an x86 platform, for example, doesn't give you the benefit. It gets you off the iron, but it doesn't change the way the application works. This is about actually modernizing the actual application, which is really cool. Now, of course, testing is always something you need to do. And so there is now automated testing. You can now use automated test plan generation, test data collection scripts, and test case automation scripts as well.

29:05And AWS Transform for Mainframe also provides functional testing tools for data migration, results validation, and terminal connectivity. So this makes you more accurate in the migrations that you're doing. So there's been a fantastic update in the world of networking. We're announcing Amazon Route 53 Global Resolver. This is a new Amazon Route 53 service that gives you secure and reliable DNS resolution globally for queries from anywhere. This is in preview, but this allows you to resolve DNS queries to public domains on the internet and private domains associated with your Route 53 hosted zones.