305: AWS Breaks Up with Unpopular Services - "It's Not You, It's Me"

[00:00:00] Speaker A: Foreign. [00:00:06] Speaker B: Welcome to the Cloud Pod where the forecast is always cloudy. We talk weekly about all things aws, GCP and Azure. [00:00:14] Speaker C: We are your hosts, Justin, Jonathan, Ryan and Matthew. [00:00:19] Speaker A: All right, well we are doing episode 305 recorded for May 20, 2025. AWS breaks up with unpopular services. It's not you, it's me. Good evening, Ryan and Matt. How you doing? [00:00:30] Speaker D: Doing good, how are you? [00:00:33] Speaker A: Well, there's a lot of show notes this week because someone in the world said let's take Google I O and Microsoft Build and Oracle is actually happening right now too, although we don't really care about that all in the same week. [00:00:49] Speaker D: So we have a summit. [00:00:51] Speaker A: That's true. [00:00:51] Speaker D: AWS likes us. [00:00:53] Speaker A: Yeah, Amazon was like we're just not going to talk much. That's unfortunately not true though. They still had a bunch of news. Yeah, well let's, let's jump right into it. This week with AI is how machine learning makes money. Google IO has quite a bit of stuff. First up, Jules, which probably is going to drive Matt crazy because that is technically the name of someone very close to him. Jules is their new autonomous AI coding agent that can read code, understand tent and make code changes on its own. It goes beyond AI coding assistance to operate independently. It clones code into a secure Google Cloud vm, allowing it to understand the full context of your project and then enables you to write tests, build features, fix bugs and much, much more. Joules operates asynchronously in the background, presenting its plan and reasoning when complete and this allows developers to focus on other tasks while it does its thing. Integration with GitHub enables Jules to work directly in existing workflows without extra setup or context switching and developers can steer and get feedback throughout the process. For cloud developers, joules demonstrates a rapid advancement of AI for coding movement, moving from prototype to product as cloud based parallel execution enables efficient handling of complex multi file changes. While it's in public beta, Joules is free with some usage limits and this allows developers to experiment with cutting edge AI coding agent and understand the future potential of how it's going to take your job. [00:02:07] Speaker C: Yeah, no kidding. Yeah, I mean more and more as tools get released, like it's just going to change the way anything gets written. I mean it's, it's crazy how fast it's, you know, in my life and how much I use it in a day to day basis and it's just getting more and more capable. Pretty crazy. [00:02:29] Speaker A: It's pretty crazy. [00:02:31] Speaker C: This is interesting. This is one of my favorite announcements from Next was A very similar. And so this is interesting. They're taking it even to a next level. [00:02:39] Speaker A: Yeah, well, I mean, Google Code Assist does quite a bit of this too. And this is like, its whole, you know, thing on top of GitHub. And I think it's actually built so you can. You don't even have to have the agent installed on your laptop. [00:02:51] Speaker D: It's free now. And I'm just waiting for, at one point all these things to. [00:02:55] Speaker A: When they start charging you lots of money. [00:02:57] Speaker D: Yeah. And at one point it's like, well, make more than I'm paying for them. It might make sense to have the AI bots do my job and I'll just go play video games. [00:03:06] Speaker C: Well, how many times? Like, I have, like, a list of ideas that I always want to do. [00:03:10] Speaker A: Right. [00:03:10] Speaker C: It's like, I don't have time to do them all. [00:03:13] Speaker D: So, like, it'll be a good, like, mvp. Like, go write me a whole program that will do this. And, you know, I mean, I wrote. [00:03:21] Speaker A: A whole program this week of AI. So, yeah, we'll talk about it next week because we don't have enough time with all the announcements this week. But, yeah, I wrote a whole bunch of AI this week for a bunch of different projects. And it's fun. Like, it's super fun. Like, is it good code? Absolutely not. [00:03:36] Speaker C: Yeah. [00:03:36] Speaker A: I literally was just working on a very simple chatbot application for Slack for the one, you know, the. The fun one that we have. Not the chat, not the clapboard one where all of our friends, you hang out, and we. We make fun of each other and, you know, have a good time, send memes and dad jokes, et cetera. And so we're making a chatbot for that one. It has two commands right now. Two commands, and it created 92 files. I was like, okay, I could see. I could see like, you know, some baseline stuff. I could see. Okay, each command maybe has a little bit of code and then you got some testing. So, like, I. I could see 20 maybe, but 92. So I'm like, I already just asked the AI before we started recording. I was like, hey, why don't you go see if you can optimize this? Because this seems excessive. So it's a. It's analyzing what it did to see if it can make that better because. [00:04:23] Speaker C: Yeah, just crazy. [00:04:24] Speaker D: It'll just put it all in one file. [00:04:26] Speaker C: Yeah. [00:04:26] Speaker D: And solve it that way. [00:04:28] Speaker C: Yeah, yeah. I'll point it at my half finished bot that I made and see if it, you know, compares itself. [00:04:36] Speaker A: Yeah, nice. [00:04:37] Speaker D: It'll probably better than that. [00:04:39] Speaker A: Compare, compare this code to this code. Which one is worse? And like, well, you know. Anyways, that's something to look forward to. Next week I will share my venture of building our new show Note Chatbot Cool. Introducing Flow Google's AI filmmaking tool designed for veo. Flow is an AI powered filmmaking tool custom designed for Google's advanced video image and language models. I think I told you guys before that I made eight six second videos and it cost me $45 of VO. Did I tell you that? Yeah, yeah, just. This is not a tool for the faint of heart or about a budget. The tool leverages cloud AI capabilities to make AI video generation more accessible but not cheaper. Creators can describe their vision in plain language and bring their own image video assets. Key features include camera controls, scene editing, extensions, asset management and a library of example clips. Zims to enable a new wave of AI assisted filmmaking. Flow is an evolution of Google's earlier video effects experiment now productized for Google as I Google's AI cloud subscribers. An example of an applied ML moving from research into cloud products and services or how they make money. Potential use cases include storyboarding, pre visualization and final render clips for both amateurs and professional filmmakers. And early collaborations demonstrate applications in many short films. So yeah, this is quite good if you're in the video thing and you're into doing a lot of editing and you have a budget text, check this out because it could help you do a lot of VX virtual effects and all kinds of other things. [00:06:02] Speaker C: I mean it's gonna, this is another area. Just like it's changing coding, it's gonna change movie making and directing. And because it's, you know that there's not only do you have the vision in your head, but you have to be good at, you know, the prompt engineering to get it out. Pretty wild. But it's gonna step up by, you know, birthday videos that I make for. [00:06:21] Speaker A: My kids or it'll make your birthday videos better. [00:06:24] Speaker C: That's what I'm saying. It's definitely going to make them better. You should see them now. It's not good. [00:06:31] Speaker D: It's like a PowerPoint is what he makes for them. [00:06:34] Speaker C: Everything is a star wipe transition. [00:06:37] Speaker D: There's others. Don't look at my corporate decks. [00:06:44] Speaker A: Google is extending its multimodal foundation model Gemini 2.5 into a world model that can understand, simulate, plan and imagine. Like the human brain allegedly. Gemini is showing emerging capabilities to simulate environments, understand physics and enable robots to grasp and follow instructions. Goal is to make Gemini a Universal AI assistant that can perform tasks, handle admin, make recommendations and enrich people's lives across any device. Google is integrating live AI capabilities into Project Astro, like video understanding, screen sharing and memory into products like Gemini Live Search and Live API. Project Mariner is a research prototype exploring agenda capabilities with a system of agents that can multitask on up to 10 different things like looking up info, making bookings and purchases, etc. These developments aim to make AI more personal, proactive and powerful, to boost productivity and usher in a new era of discovery. I mean, I can't wait for an assistant. [00:07:35] Speaker C: Yeah, I mean this is, this is. Ever since, you know, the first Iron Man, I think this is what everyone's been waiting for. [00:07:41] Speaker A: Yes. Personal Jarvis. [00:07:43] Speaker C: Yeah, yeah, because it's, you know, and they're just now, you know, AI is getting capable and the agentic model allows sort of that separation of units between models and their, the communications coming with all the, the new protocols and this is fantastic. And, and little scary, little, little like, oh, we're not. Humans are going to be extra. [00:08:09] Speaker D: I mean this is really, I feel like getting enabled with the agent agent, the MCP stuff that's coming out, you know, everyone's adding it and having all the bots be able to talk to each other and kind of delegate it out. I feel like it's going to be like, hey, Gemini, I have a party for my 40th birthday that I want my 50 friends to come to find me a restaurant reservation in San Francisco on this date that can handle it. It'll just go out, do the research, find it, book it, send the invites. It'll be amazing when we get there. And also very terrifying. [00:08:45] Speaker A: All right, and next up at Google I O they announced major Updates to Gemini 2.5 large language models, including the 2.5 Pro and 2.5 Flash versions, which are leading benchmarks for coding, reasoning, learning and much more. The new capabilities include native audio output for more natural conversations, advanced security safeguards against prompt injection attacks, and ability to use tools and access computers. Experimental Deep Think mode enables enhanced reasoning for highly complex math encoding tasks and the developer experience. Improvements include thought summaries for transparency, adjustable thinking budgets for cost control, and support for model Context protocol or MCP tools. The models are available of course, via Vertex AI and the Gemini API for businesses and developers to build intelligent apps. The rapid progress and expanding capabilities of large language models have major implications for unlocking new AI use cases and experiences across industries, which is pretty nice. [00:09:32] Speaker C: Yeah, it's crazy. Like the, the model improvements and the velocity of them. Right. Like it's, I was using Gemini. I don't even know if it wasn't the Pro or Flash versions, but it's the 2.5, I guess normal. And it's, you know, world's better than it used to be. And it's, it's funny, I'm finding myself hopping from model to model lately as I, you know, as they make improvements in the market and like they get a little bit better, it's kind of, it's strange to hop around. It's. And it's changing the way I sort of like think about doing business and because it's like you couldn't really buy a single tool, you know, and be able to hop around models to models. Like it's a little strange trying to figure that all out. [00:10:18] Speaker A: I mean, more natural conversations. I guess we're still going to be out of work soon here at the podcast. [00:10:23] Speaker C: I mean none of my conversations are natural, so getting stuff out of my brain into the world is a difficult journey. Lots of pitfalls. [00:10:33] Speaker A: Well, apparently on the research and development side, DeepMind is building a something called Alpha Evolve, which is a new AI coding agent for the Gemini language model which with addition of an evolutionary approach to evaluate and improve algorithms. It uses an automatic evaluation system to generate multiple solutions to a problem, analyze each one and iteratively focus on the end and refine the best Solution. Unlike previous DeepMind AIs trained extensively on single knowledge domain. Alpha Evolve is a general purpose AI to aid research on any programming or algorithmic problem. Google has already started deploying Alpha Evolve across its business with positive results for cloud computing. Alpha Evolve could enable more intelligent, efficient and robust cloud services and applications by optimizing underlying algorithms and architectures. And businesses and developers could leverage Alpha Evolve to tackle complex problems and accelerate R and D in fields like scientific computing, analytics, aiml, etc, on the cloud. There you go. That's great news for us who want to have be employed. [00:11:29] Speaker C: Not even sure I fully understand this one, you know, and I'm going to get replaced by it. [00:11:33] Speaker A: Yeah, I think basically if you thought you could, you're like, well, AI can't create new algorithms, so I'm good on that. Like, no, Nope, you can too. [00:11:43] Speaker D: I do really like, you know, playing with all the different models and hearing what's coming is, you know, both fascinating and terrifying because you're like, cool, I can go play with this. But you're also like, how many things do I do now that I've like retrained my body and brain how to think in a different way in order to be able to actually build this out or you leverage the new tool. [00:12:07] Speaker A: I mean, hopefully. I mean, unless they don't need us anymore because the other AI is doing all the programming work. This is creating the new algorithms and then we're going to get quantum computer that's just going to figure out all possibilities and figure out that you're just going to die at this point. Rehoboam in Westworld. Oh no, you're basically going to die five years from now based on all the AI inputs that we have and we're not investing in you. Goodbye. That's the worry. There's things in Westworld you're like, oh, this is starting to feel a little concerning. [00:12:37] Speaker C: Too close to home. [00:12:39] Speaker D: Yeah, I think we're very grim today in the show. [00:12:41] Speaker C: Yeah, I don't think recording later is doing us any favors. [00:12:44] Speaker D: No, no, see, go back to early. [00:12:50] Speaker A: Well, OpenAI. I didn't put it in the show notes this week because it's just a purchase announcement. They bought Windsurf if you guys were paying attention and then released their own competitor, Windsurf, which was sort of weird. I'm like, well why did you buy them if you had your own thing? They've released Codex, an AI agent that can generate production ready code based on natural language prompts from developers. Codex runs in a containerized environment that mirrors the user's code base and development setup, and developers can provide an agent's MD file to give Codex additional context and guidance on project standards. Codex is built on the Codex 1 model, a variant of OpenAI's O3 reasoning model that was trained via reinforcement learning on a broad set of coding tasks. For cloud developers, Codex could automate routine probing work, boosting productivity. Businesses could leverage Codex rapidly prototype cloud applications and services. And Codex represents a major step toward AI systems becoming fully fledged software development partners alongside human programmers. While still in the Research Preview, Codex provides points to a future where AI is deeply integrated in the cloud application development lifecycle. But it will be quite a while before we get to this one because you have to have the $100 license to even try it. [00:13:51] Speaker D: Yep. [00:13:53] Speaker A: And I while I am now paying $100 to Anthropic, I refuse to pay $100 to OpenAI. [00:14:00] Speaker D: Don't worry, there's more later on that you spend more money on it. [00:14:03] Speaker A: Yeah, we'll talk about the one I'm not definitely not spending money on later. [00:14:07] Speaker D: The only thing that makes me feel good about this is a developer has to figure out how to run their application within a container. And I've yet to meet a developer that can do that easily themselves without some assistance. So I feel like we're still good for a little bit. I have a job because I have to set up a container to then let the developers, you know, go go at it. [00:14:27] Speaker C: I think that's just the back end, man. I don't think that just make me myself feel good. [00:14:33] Speaker D: We're trying to be optimistic this show after we've been so pessimistic thus far. Then there's an Azure conference, so you know it's got to go downhill one way or the other. [00:14:44] Speaker A: Yeah, like how about Hashi Ibim my name for hash IBM but hashtag Hashicorp, which is the subsidiary of IBM, apparently has new validated patterns to provide pre built validated solutions for common use cases using HashiCorp tools like Terraform, Vault Consul and Nomad. They help accelerate time to value by providing a starting point for building and deploying production ready infrastructure and apps in the cloud. Patterns cover core use cases like service networking, Zero Trust Security, Multi cloud deployments, Kubernetes deployments and much much more and integrate with major cloud platforms including aws, Azure and Google Cloud. What in Oracle Validated patterns solve the problem of figuring out best practices and recommended architectures when using HashiCorp tools for common scenarios. Patterns are fully open source and customizable, allowing users to adapt them to their specific needs. And this matters for you because it makes it faster and easier to properly implement HashiCorp tools in production by leveraging curated validated solutions. So blueprints but for HashiCorp I applaud this one. Thank you. [00:15:42] Speaker D: I did until I looked a little bit more like into the article and I clicked through and like find all of them but like they're like cool. Terraform with Prisma Cloud by Palo Alto Networks maybe that's a good idea. I don't know. I just feel like there's going to be someone that runs a terraform destroyer takes down your Titan Prisma cloud feels like a bad life choice. Or you know I just saw Apache Kafka and immediately just went I don't really want to do that. So you know I've had other choices in there or Cloud AWS cloudformation templates to terraform configurations. [00:16:16] Speaker A: I have noticed that the AI has a tendency to want to use cloudformation and I always had to remind it no, no, please use Terraform. [00:16:23] Speaker C: That's interesting. [00:16:24] Speaker D: Are you using Q is why? [00:16:26] Speaker A: No, I'm not Using Q, that's the worst part. There's so much more document I mean you have to think about. There's so much more documentation that exists for Amazon services that's cloudformation because of whatever that this is just how we end up where we're at. [00:16:39] Speaker D: So the only tools that are going to get used in the future and be tools that have good documentation. [00:16:45] Speaker C: Yep, yep. [00:16:46] Speaker D: That's terrifying. [00:16:49] Speaker A: All right, let's move on to aws who was blessedly quiet ish I guess comparatively to everyone else who's announcing everything in the product sun. First up they decided that they are going to kill some more products which you know, I'm not mad about it because they're products I don't use so it's easy not to be mad about it. They're ending support for several services including Amazon Pinpoint, AWS IQ, IoT Analytics, IoT IoT Events, SimSpace, Weaver Panorama Inspector, Classic Connect, Voice ID and the DMS Fleet Advisor. This basically end support means these services will be no longer be available after specific announced dates and AWS will provide customers with detailed migration guidance and support to transition alternative services. Some services like AWS private 5G and data sync Discovery have already reached end of support and are no longer accessible already. And this announcement matters because ending support for services can significantly impact you and if you rely on them. But I don't. So I don't care. In addition to this, they have announced publicly, finally after a lot of chastising from our friend of the show, Corey Quinn, an AWS product Lifecycle page, which is a page you can now bookmark, which will tell you which services are being deprecated. Services closing access to new customers ends up ending support or reaching full closure status as well. Now this page is limited, missing all the ones they previously announced, so hopefully those get added sometime in the future. But hey, it's a good step forward in being more transparent about the services that you are now starting to deprecate. So appreciate that. [00:18:17] Speaker C: I mean there's a reason why we don't make fun of this like we used to make fun of Killed by Google. Right. And it's a lot of it has to handle with the way that you sunset these services versus the very abruptness of, you know, some of the Google killings of the past with a lot more usage than these. Like I, I haven't really heard of any of the AWS services impacting anyone I know, right. That services they're killing I think largely are not used very much and there's probably customers out there using it, but it probably doesn't have enough of a foothold to make enough noise, so it's kind of interesting. But I like it, and I definitely like the new service. [00:18:57] Speaker D: The one I was surprised by was Pinpoint, because I thought that was supposed to, like. I felt like they were pushing that really hard when it first came out. And I assume people adopt it. No one in my circle, but I just felt like they were like, it's a service. It was email, it was targeted. It probably was a tool they used internally. I was surprised by that one. [00:19:18] Speaker C: Yeah. I mean, Sip Space Weaver seemed like a big investment as well. [00:19:22] Speaker A: It was cool, but I don't know what I would do with it. [00:19:25] Speaker C: I figured someone would figure it out. I never did, but apparently the. No one did. [00:19:31] Speaker A: I mean, I now understand the digital twins. Like, that took me a long time to get my head wrapped around, but now at least I have that and how I think about a bunch of these things. But, yeah, sometimes they build stuff to see if it sticks on the wall, and maybe it does for one or two customers they built it for, but then no one else is interested. And I think that's kind of a death knell for a lot of these things. [00:19:50] Speaker C: Well, didn't Panorama come out of, like, the COVID AI sort of video work that they did in their warehouses? Like, that's, you know, like, that's kind of a crazy thing, you know, like, so, I mean, that's just, I guess, you know, Panorama is just one of those things. Like, the technologies moved on and probably left it behind. Yeah, I guess. [00:20:10] Speaker D: I mean, like, Inspector Classic makes sense, you know, And I could see them doing it for a lot of these, like, V1s of things. Like mace. Like, Macy had, like, that V1 before they, like, redid it all. Like, I feel like there's probably a lot of those that we'll start to see. Kind of tack on some of them. Just some of these services. I forgot about, like, private 5G. I was like, oh, yeah, I forgot. It was like, build Your own private 5G network inside your, like, warehouse. That. [00:20:38] Speaker A: Yeah, that's back in the day when 5G was going to save the world. [00:20:42] Speaker C: Yes. [00:20:43] Speaker A: It was going to solve all problems, all issues, you know, like, 5G is going to do all the stuff. Like, I'm like, it's just connectivity, guys. Like, what are they really going to do for us? And then, like, they were like, oh, we have private offer 5G. I'm like, okay, what do I want that for? Like, oh, my warehouse in the middle of Nowhere now has 5G coverage. I'm like, I don't get it. So yeah, I don't, I don't think a lot of people got it. Which is the problem, other than other than the cell phone providers who probably love this stuff because they made their ability to deploy 5G faster. All right, well, moving on to the new Strands agent. An open source AI agent SDK directly from Amaz on the SDK is to simplify building AI agents by leveraging advanced language models to plan chain thoughts, call tools and reflect. Developers can define an agent with just a prompt and a list of tools. Integrates with Amazon Bedrock. Well, there's a strike against it that supports tool use and streaming as well as models for anthropic, Metis, Llama and other providers. Strands can run anywhere. Apparently. The model driven approach of Strands reduces complexity compared to frameworks requiring complex agent workflows and orchestration. This enables faster development and narration on AI agents use cases, including conversational agents, event triggered agents, scheduled agents, and continuously running agents. Strands provides deployment examples for AWS, Lambda, Fargate and EC2. Right now this is all price on all of the underlying models and services that you're using. This is all based on pre release pricing information that they've given us. So it might charge you something for in the future, but right now it seems like it's just open source and you can use it anywhere. [00:22:11] Speaker C: I hope we don't get too many more of these to be honest, because now like OpenAI has one, Google has one, Amazon now has one. Like it feels like great, we've got a whole bunch of open source options that do the same thing, right? And it's like, okay, they're all trying to, instead of like collaborating in the open space, in the open source market, they're creating their own competing versions of it and it's going to make things diverge, which I don't like. Cloudflare even has their own agent SDK, so. [00:22:40] Speaker D: Sweet Jesus. [00:22:41] Speaker A: Yeah, it just came out. I almost put it in the show notes this week and I was like, I can't. I. I think I thought even Digital Ocean had something coming out about agents and I was like, okay, like we're going to table those for next week when I have more time to read. But you know, everyone, everyone's got to do AI because that's, that's how you get money if you're on the public market or you're trying to raise money from investors. [00:23:03] Speaker C: Well, I mean, I get the AI thing, but these are specifically SDKs. Right. So it's like it's for. And the reason why they're all open source is just, you know, to make hitting their existing APIs easier. Which makes sense. I mean that's, that's always good. But it's. I don't know, the, it's the. I wish there was more, a little bit more uniformity, you know, Like I like the things that are taking off that are more generalized like the ncp and these are just getting kind of strange to me. [00:23:33] Speaker D: I feel like they're just. You're kind of gonna end up with like, you know, 50 of them out there and you'll kind of. The market will naturally select like two or three that everything will fall into is at least kind of the way I'm viewing it or hoping that it happens. And obviously I'll choose the wrong one to learn and then I'll be mad in about a year from now. Yeah, just making sure we're all clear what's gonna happen. [00:23:54] Speaker C: Yeah, that's what. [00:23:55] Speaker D: So don't do whatever I choose. [00:23:57] Speaker A: Well, AWS apparently has a. A bone to pick with Microsoft. It used to be Oracle they hated, but I'm starting to think they don't like Microsoft that much either. Maybe because they make Azure in our competitor, but they originally competitor, I promise you. But if you go back, they had Babelfish which was supposed to help you get off a SQL Server into postgres using a layer. I don't know if I actually use that thing, but they definitely talked about it a lot. But now they've got a new one. AWS Transform service is also available for mainframes and VMware, but we killed that story and just focusing on this one because it's Net. So AWS Transform for. Net is a new AI powered service that automates porting. NET framework applications to cross platform Net making modernization faster. For those who don't know Net. That's Net Full NET to full Net Core NET Core runs on Linux. It was ported and so this makes your modernizations faster and less error prone. This matters because ported apps are 40% cheaper to run on Linux and have 1 1/2 to 2. X better performance and 50% better scalability because you can run them in a container. It integrates with source code repositories like GitHub, GitLab, BitBucket and provides experiences through web UI for large scale portfolio transformations and a Visual Studio extension for individual projects. New capabilities include support for private NuGet packages, supporting MVC razor views and executing ported unit Tests cross repo dependency detection and detailed transformation reports. Enterprises with large portfolios of legacy. NET frameworks that want to modernize to Linux, reduce costs and improve performance capability will benefit the most from this tool. And I got a lot of requests for access to this this week. Hey, can I get to that? Yeah, sure, do that for you. [00:25:33] Speaker D: Yeah, this is on my list of things to look at, you know, for my day job because if I can move any of the legacy apps off full blown Windows into a container or Linux or anything else, my life will be easier. [00:25:46] Speaker C: Yeah, I don't think Amazon has a thing for Microsoft much as more is just like there's a ton of businesses that are completely stuck in this position and have no way to get out. And much like Babelfish, this doesn't solve a lot of those concerns. Like, you know, Babelfish never really took off because you couldn't just plug and play replace it. There was enough deep, you know, Microsoft SQL things that, you know, people had coded into the app that made it functional that Babelfish couldn't do a good job replicating. I mean, judging by the adoption rates. I wish it could. I don't code in. NET and don't use Microsoft SQL Server because I'm not made of money. So I don't know. But developers have to embrace this as well. They have to use it, they have to change their development patterns and not only their development patterns, but their entire IDEs, their ways of working, their ability to test things locally run code, you know, is all change has to change with these types of paradigms. And so you still have to have sort of a developer team that's on board. Even though you've got AI rewriting it for you and doing the hard work, you know, it's still, for now anyway, developers still need to do a thing soon AI will make that unnecessary as well. So. But I think that's the problem with these types of services. This is the magic button we've all been waiting for to get off of these, these old, older platforms. But it still requires a little bit of human capital in the meantime. [00:27:15] Speaker D: Yeah, but there's also still I found with any of these services, they're a good starting point, but there's still a lot of legwork that needs to be done later on. And looking Even at the. NET1, there's some sharp edges like application has to be in NET8. Not all, I guarantee you, not all companies even are at net 8. Windows 32LLs do not have core compatible libraries like it will work for I think to kind of start you in the right direction and kind of the way, you know, for lack of better term vibe coding is done of like 80% of the way there. So that can be pretty. It's like Microsoft front page making HTML. It's not going to be pretty. It will work. But if you need to get under the hood and tweak anything, good luck. [00:28:01] Speaker C: Well and there's just so many things that people don't really they take for granted for changing to. NET Core. Sure you can start running it in a container, but then how are you managing sessions? How are you killing sessions? All of these things that are very different and are huge sharp edges when you change that paradigm. This is great. I just really hope that it gets adoption whereas the previous attempts at it hasn't. [00:28:30] Speaker D: I think the problem is people try to bite off more than they can chew. Okay, cool. We've taken our Windows code and shoved it into Linux. So let's immediately go to a container. No, let's not do that. That's going to break everything. Why don't you at least go to a static Linux box first? Because you only know how to manage a static Windows box and then from there jump to a container like so if you do the more intermediary steps I think you'll have more success on it. But people want to go, cool, we immediately can go to containers. And it's like managing a container platform is a completely different skill than managing a Windows scale set. They're just not the same. [00:29:07] Speaker A: I would say that even you want to take even a smaller baby step, make auto healing for a single server work. [00:29:15] Speaker D: Yeah, it's Windows, it takes too long. [00:29:18] Speaker A: But if you figure that out then you can get down more towards that path. But I know one of the things even when I only need one server because I don't care about ha or I don't care about. I don't want to save spend money like I always build it to auto scaling group. That way I have an easy way like I'm always thinking about. And when I do need to scale up it's not actually that much of a lift because I've already done a bunch of the heavy lifting. But I don't have to like get into like what do I do about session across multiple boxes. A bunch of things that are a little more complicated. But it's, that's a good baby step too I think. [00:29:45] Speaker D: Yeah, that was like one of my first things that you know when I came into my current job was like even our jump boxes that we leverage our auto healing boxes for that reason don't care. They don't need ha. But like I still want still need them in that way. Yeah so I can you know I I understand probably a phrase and things that you can't say anymore but probably has been that I've been in the cloud too long but like I want to manage pet cattle not pet. I want to shoot that jump server in the head and let just build me a new one. I I don't care. [00:30:20] Speaker C: Yeah I think the trick is with these that there's a lot of business initiative right to save money to do all those things and then there's the internal friction of change and that and they you have to marry the both right. Because the motivation has to be there for the right reasons and the people have to be involved to make this happen. So like I really, I'm just really hopeful that this takes off and I guess that's why I'm more cautious with this because I don't want to see it like be the solve all for everything and then land on its face. [00:30:57] Speaker B: There are a lot of cloud cost management tools out there, but only Archera provides cloud commitment insurance. It sounds fancy, but it's really simple. Archera gives you the cost savings of a one or three year AWS savings plan with a commitment as short as 30 days. If you don't use all the cloud resources you've committed to, they will literally put the money back in your bank account to cover the difference. Other cost management tools may say they offer commitment insurance, but remember to ask will you actually give me my money back? Artero will click the link in the Show Notes to check them out on the AWS Marketplace. [00:31:36] Speaker A: All right, let's move on to another way to run Docker in Amazon. [00:31:42] Speaker C: Another. [00:31:42] Speaker A: Another one? Yeah. AWS code builds New Docker Server Capability Provisions A dedicated persistent Docker server within a code build project to accelerate Docker image builds. It centralizes image building to remote host with a consistent caching, reducing wait time and increasing efficiency up to 98% faster builds. In example, the persistent Docker server maintains layer caches between builds, especially beneficial for large complex Docker images. With many layers, it integrates seamlessly with existing code build projects, simply enabling the Docker server option when creating or editing a project. Supports both x86, Linux and ARM architectures and ideal for CI C pipelines that frequently build and deploy Docker images, dramatically improving your throughput. Pricing on this will vary depending on the Docker Server compute type, check the code build pricing page for all of the details. But, yeah, that's nice. I definitely use a lot of Docker builds. [00:32:30] Speaker C: I do, too. But I don't quite understand where this fits. So, like, you, you're building Docker images. [00:32:39] Speaker A: So right now, if you have code build and you want to build on a Docker server, you have to connect to an ECS or Fargate instance that's inside of a VPC elsewhere. So you have to do peering to where your code build environments. And now this is. You can basically run this as a fully managed Docker server inside the code build environment. [00:32:54] Speaker C: Okay. [00:32:55] Speaker A: So you don't have to do all that extra connectivity steps. That's the advantage here. [00:32:58] Speaker C: Thank you. [00:32:59] Speaker A: You're welcome. [00:33:00] Speaker C: Because this is awesome. Now that I know that that's. Yeah, that's always. The. Getting started with code build is always hard. [00:33:06] Speaker D: Right. [00:33:07] Speaker C: Because there's so much setup that you have to do to get. It's great. Once you get the infrastructure up and running, then it's almost like a top. It sounds like this is solving that problem. That's really cool. [00:33:18] Speaker A: All right, let's move on to our next story here. Amazon Inspector is enhancing security of Inspector inspection of Amazon ECR container images to running containers and Amazon ECS and eks, providing visibility to which images are actively deployed. So, you know when you get that message from security saying, you know, you've had 12,000 vulnerabilities in your Docker containers, and you're like, well, the one that's actually running has no, you know, has no vulnerabilities because we have a good pipeline. All those are old ones. You can now help filter that down for your security team and say, those are old ones. Yes, we can delete them so your numbers get better. But calm down. Close the seven incident, please. [00:33:54] Speaker C: There's also a cost implication. Right. I've always run into this when managing. [00:33:59] Speaker A: A repository and there's no more space. [00:34:01] Speaker C: Which ones are the ones we need? [00:34:04] Speaker D: You never know. [00:34:05] Speaker C: I've tagged them all as slash dev. Are we good? So this is great. I like to see this. I think it'll help a lot. It's a nightmare. And I can't wait. I hope this goes into many other patterns and technologies, even more than ecr. [00:34:21] Speaker A: Yeah, I mean, I can see a bunch of, like, lambda functions. [00:34:24] Speaker D: Like Ammis would be great, too. [00:34:26] Speaker C: Yeah. [00:34:26] Speaker D: Layers. [00:34:27] Speaker A: Yep. [00:34:28] Speaker D: Yeah. Everywhere would be great. This feels like something that's, I want to say really basic, but I also understand how complicated it can be. Especially because day one, they already have works across single account, multiple accounts and organizations. So like that's a pretty good day one feature where I feel like that normally wouldn't be for aws. [00:34:51] Speaker A: All right, let's move on to gcp. So nothing to say to Matt? Sorry, I heard your question and I was like, I don't know what to say on that. Like I agree with you. So sorry I left you hanging. Google AI Ultra is a new premium subscription plan providing access to Google's most advanced AI models and features including Gemini, Flow, Whisk, Notebook LM and many more. Offering the highest usage limits and easy and early access to cutting edge KPIs like VO3 video generation and DeepThink 2.5 Pro enhanced reasoning mode, it integrates Google AI directly into apps like Gmail, Docs, Chrome Browser, all for seamless AI assistance. It also includes YouTube Premium and 30 terabytes of Google One storage, which. Okay, you're just trying to justify this cost because this is targeting filmmakers, developers, researchers and power users doing the best Google AI has to offer. So I guess if this makes my 6 second videos part of this monthly fee, this isn't so bad. But if you're thinking to yourself, well it's going to be like Anthropic and Claude at $100 a month, you'd be wrong. Yep, it's 249.99amonth with a 50 offer intro for the first three months. So the first three months it'll cost you half of that amount. 125. And then basically expanding Google's AI offering to compete with Microsoft, Amazon, OpenAI and others in the rapidly growing case. So yes, if you want more AI from Gemini and you don't want to pay for the drip of the API, you can now pay 250 months for all these things which again, if you're doing VO might be the right way to go. [00:36:15] Speaker C: Right. Well, I think you still, I mean I don't think you get all of the that VO content for free, Right? [00:36:22] Speaker A: No, I mean I imagine there's still. [00:36:24] Speaker D: Some like some limit charge. You don't worry. [00:36:26] Speaker C: Yeah, yeah, yeah. I was shocked at the price of this and then and laughing at the inclusion of YouTube Premium just because it's like I was waiting, I was fully expecting this as I clicked the article that this was going to be 100 bucks and it's not. It's more than, you know, like it's interesting. [00:36:45] Speaker A: So they, this is a Google One plan which is typically where you buy extra storage. I happen to have one of these. I think it cost Me a hundred dollars a year. And so this says it's 250 a year. Maybe that's a little better. I don't feel quite as bad about this, but it does say it's only VO2 on the product page. [00:37:01] Speaker C: Wait, 250? I thought it was 250. [00:37:03] Speaker D: It's pretty much. [00:37:05] Speaker A: Hold on. Maybe that. Oh, that's the one. No, sorry, I. This. They sent me to a link to another part of the Google 249amonth. [00:37:12] Speaker D: Per month. [00:37:12] Speaker A: Per month, yep. But 125 for the first three months, correct? Yeah, sorry, that was my confusion because of the number one thing threw me off. [00:37:20] Speaker C: Yeah. [00:37:20] Speaker A: But. Yeah, so that's. Yeah, pricey. But let's see. I don't see. Yeah. VO2 and early access to the groundbreaking VO3 model. So it's not the latest and greatest stuff in all cases. And then Flow, that new tool we talked about a little bit earlier. [00:37:35] Speaker C: Yeah. They don't do them a service like in the article. They have a comparison with the Google AI Pro, which is their free tier for one month, and then 20 bucks a month after. And there's just not a whole lot in addition. Right. For the 20 bucks a month, you get all the Gemini Access, Whisk, Flow, Notebook, LM Gemini in your Google Apps, and two terabytes of storage. And so it's like they, you know, there's definitely some improvements, but Project Mariner, is that enough to justify this cost? [00:38:08] Speaker A: I mean, Project Mariner is so young. Like, it's hard to say that's going to be valuable yet. Yeah, no. So I'm. I'm just looking. Okay, so YouTube Premium is basically 13.99amonth. And then the storage, 3 terabytes is the Google One plan. That's where I got confused. That's 99 a year, which is $8. That's. So it's $21. So basically, of the $250, 25 of it is just the storage and the YouTube part, and then the rest is supposed to be all this other stuff. [00:38:34] Speaker C: Well, but that's 25 bucks a year for the. So you have to divide that by 12. [00:38:39] Speaker A: I broke it to the monthly cost. [00:38:40] Speaker C: Oh, you did? [00:38:41] Speaker A: Okay. Yeah, yeah. $13.99 a month. And eight. And then the 99 drive by 12. So, yeah, that's how I got the YouTube. [00:38:46] Speaker C: But yeah. [00:38:48] Speaker D: All right. [00:38:49] Speaker A: Yeah. [00:38:49] Speaker C: I mean, I don't know. These things are expensive to do. I know they're expensive to run. I imagine the margins on these are still really slim, but I mean, I. [00:38:59] Speaker A: Look forward to someday making someone upset about a cost of inference because clearly it must be much higher than I realize. Yeah, crazy. Well, Database center is an AI powered unified fleet management solution for Google Cloud SQL. It is now generally available for you, so congratulations to the Google team for getting that out. It proactively identifies risks and provides intelligent recommendations to optimize performance, reliability, cost compliance and security. It introduces an AI powered natural language chat interface to ask questions, resolve issues and get optimization recommendations. And it leverages of course, Gemini to do all that work that allows you to create custom views, tracking historical data on database resources and issues, and centralizing all of your database alerts across your massive fleet. Competes with database managed offerings from AWS and Azure, but differentiates with AI powered insights and tight integrations with GCP's databases and AI ML services. No additional cost for the core features. Premium features like Gemini based performance cost recommendations required Gemini Cloud Assist, but you do pay on the API layer. Advanced security requires Security Command center, which is very pricey, so do be cautious that if you don't have some of these other things this may not be as great of a value to you as you might hope. [00:40:07] Speaker C: And while I really like this feature, I want to make fun of it just because it's it'll be like a lot of the other Google services where it'll just be very confusing to the end user where they won't really know which service they're using under the covers. They'll click a button, they'll set up a whole bunch of stuff up and then they'll get a bill that has AlloyDB on it and they'll be like I don't understand what this is at all, so I look forward to that conversation. [00:40:32] Speaker A: GKE data cache is now generally available, which accelerates your stateful app. This accelerates read heavy stateful apps on GKE by intelligently caching data from persistent disks on high Speed Local SSD SSDS and you provide up to 480% higher transactions per second and 80% lower latency for PostgreSQL on GKE simplifies, including a high performance cache layer versus a complex manual step, and competes with offerings like Amazon elasticache and Azure Cache for Redis, but is more tightly integrated with GKE and GKE's persistent disks. Potential cost savings are here by allowing you to use smaller persistent disks and less memory while still achieving high reperformance. Remember those disks, those local ones? They go away when the server dies, so they're not stateful. They definitely will Die key use cases of this will be your databases, your analytics platforms, your content management systems, developer environments all but need fast startup. This is actually boosts the performance of business critical staple apps on GKE with minimal effort unlocking new use cases for your business. This is based on local SSD usage and varies by the configuration. So a 375 gigabyte local SSD is only $95 a month. So your experience will vary. [00:41:38] Speaker C: Don't even think about it, Justin. Don't even think about it. [00:41:41] Speaker A: I'm not thinking about it. [00:41:42] Speaker C: I know you are. I know. [00:41:43] Speaker A: Nope, nope. Not until you said it. [00:41:44] Speaker D: Don't check your email. Don't check your email right now. [00:41:49] Speaker C: Yeah, I'm sure there's already an action. [00:41:51] Speaker D: No. [00:41:52] Speaker A: You know how long it took me to get rid of the ephemeral disks on all of our SQL servers? I'm not going back, I'm not going backwards, Ryan. [00:41:59] Speaker C: But you could run it in containers. [00:42:02] Speaker D: I won't tell him to run SQL and containers. I thought I got him off this. [00:42:06] Speaker C: I. I don't think he understood what this was. I think he would have been all over this. It would have been. [00:42:10] Speaker A: No, I knew. I know what it is but I. You guys have brow beating me so hard on this that I just don't, I don't even try anymore. And like my goal right now is to get SQL to Linux and then if I get there, I'll be happy. Like that's as far as I'm going to go on this path. Because you guys all hate the for. For good reasons. Even though Azure runs it at scale on Kubernetes, so. Just saying. And so does Google and Amazon. [00:42:33] Speaker D: I think it's not always very well. [00:42:35] Speaker C: Yeah, I think it's definitely possible to do and I do think announcements like this, like, you know, really speak to some of the, you know, the hard edges of that. You know, like there. Because there is a lot of running persistent disks and you know, stateful apps that doesn't really fit with the Kubernetes sort of ecosystem. So it is sort of a square peg and a round hole and you get things like, you know, disk latency and things that make running applications hard and then very difficult to troubleshoot and figure out what's going wrong. So like a lot of my complaints and beating away of ideas of running SQL Server and Kubernetes is largely because of, not because of the functionality of the application, but those weird performance concerns that you're going to run into that are going to be much harder to sort out when you're going to end up running a single container for kubernetes node trying to fix it. [00:43:32] Speaker A: Well, we just talked about the cost of inference and Google is introducing LLM dash D, an open source project that enhances the VLLM inference engine to enable distributed and disaggregated inference for large language models in a Kubernetes native way. LLM-D makes inference more cost effective and easier to scale by incorporating a VLM Aware inference scheduler, support for disaggregated serving to handle large requests and multi tier KV cache for intermediate values. Early tests by Google Cloud using LMD show 2x improvements in time to first token for use cases like code completion. LMD is a collaboration between Google Cloud, Red Hat, IBM Research, Nvidia Core Weave, AMD, Cisco Hugging Face, Intel Lambda and Mistral AI. Leveraging Google's proven technology and securely serving AI at scale, it works across Pytorch and JAX frameworks and supports both GPU and Google Cloud TPU accelerators providing flexibility and choice. Of course Amazon's not listed there, so they don't support any of the Amazon inference chips, but maybe Amazon will steal this because it is open source, so maybe that'll happen. Deploying LL D on Google Cloud enables low latency high performance inference by integrating with Google's, Google's global network, GKA AI capabilities and AI hyper computer across software and hardware technologies. This is great for gentic AI workflows and reasoning models that require highly scalable and efficient inference. No pricing was provided, but most likely will vary by usage and underlying compute resources of your GKE clusters. [00:44:53] Speaker C: Well, it's open source, right? So it's, it's more just like software that you run. [00:44:56] Speaker A: It's just software project that distributes the. [00:44:59] Speaker C: The, the inference load. I mean it's, it's interesting to see. I, I like reading about all these problems I don't have in my day job yet. Uh, you know, like the, the problems that they're solving because by the time I actually do get there, hopefully they'll all be addressed. So like, you know, speed of response is a big one and I think that tools like this are, are looking to address that by, you know, sort of doing what we used to do, you know, tuning the performance of web servers and trying like, oh, what's better than one web server, two web servers? And I think that this is, it's kind of neat to watch it happen all over again. [00:45:33] Speaker D: Yeah, you're watching it, you know, at new levels and seeing how they're going to solve these problems is interesting too because like we've solved them over here in this way, but they're slightly different because they're different types of problems and the collaboration effort between all those companies and companies that are producing their own models and everything is interesting. And hopefully, like we said earlier, we're going to hopefully solve these problems here and then everyone else kind of just grab them versus everyone coming up with their own solutions. [00:46:10] Speaker C: Yeah, I wonder if this was capitalizing on like, did the, did the community sort of look at like vertex AI and some of the things that they've, they've sort of productized and be like, how are you doing it? And then sort of started the collaboration that way. Or I wonder like it'd be, it'd be kind of fun to be a fly on the wall and how this was made. [00:46:28] Speaker D: Honestly, my assumption from like, you know, me, you, me managing some of these types of things is like just speed of response in general. Like, and this is even around like the inference but like just making API calls to Azure or you know, I've done it myself in cloud and other places. Like you sit there and you watch it spin and you're like, wait, did I do this right? Especially the first couple times you do it and it's processing, but you don't really have that near real time feedback that you're kind of expecting. [00:47:00] Speaker A: All right, well Ryan mentioned that, you know, the things mature without him and then when he gets there that he would like to use them. And so I'm pleased to give you, Ryan, Spring AI. So if you ever decide to adopt Spring Boot, we now got AI for you. Spring AI 1.0 enables seamless integration of AI capabilities into Java applications running on Spring Boot, allowing enterprises to leverage AI without complex integrations towards various AI models for image generation, audio transcription, semantic search and your chatbots. Provides tools to enhance chat models with memory, external functions, private data injection, vector stores, accuracy evaluation and cross service connectivity via the Model Context protocol. Integrates with Google Cloud's Vertex AI platform and Gemini models through specific comparisons to other cloud AI offerings not provided. Utilizes Google Cloud's LODB or Cloud SQL for scalable highly available postgres databases with PG vector capabilities to support vector similarity searches and key use cases. Include modernizing enterprise Java apps with AI so you're welcome. [00:48:01] Speaker C: Yeah, I mean, I guess Spring Boot's better as a framework for Java apps than things that have come before it, I think. Yeah. And I, you know, like it's done a good job standardizing a whole lot of like Java startup and a whole bunch of stuff that used to be a little bit more here or there. And so I guess if you do the same thing for AI integration, perhaps it'll be a little easier for the communities. I don't know. This is one of those ones where I think it's good but I hate it just because of like old scars supporting these in production. [00:48:38] Speaker D: All I can think of still is linking this to the prior article. I'm like, oh, Java cold start issues. [00:48:46] Speaker C: No for sure. [00:48:48] Speaker A: Who doesn't love good Java cold start issue? It's like a lambda cold start. [00:48:51] Speaker D: Yeah, well yeah, and they did solve Java cold start lambdas, but they never solved it for everyone else. Or maybe they did. [00:48:58] Speaker A: Well, Google did Google solve it or I remember Amazon tried to solve it multiple. I think Google did try. Now that I think about it for more than a second, I think I. [00:49:05] Speaker D: Thought Amazon had like cold boot for Lambda. [00:49:08] Speaker C: Amazon built it into Lambda. [00:49:10] Speaker A: Yeah, yeah, it's built into Lambda for sure. But I do think am I Do you think Google did something in containers for Java? I just remember what it was to make it boot faster. [00:49:19] Speaker D: I mean can they do it for Windows too? [00:49:21] Speaker C: No, AI will just move it all to. [00:49:25] Speaker D: Net Core or transform. Perfect circle of this podcast today. [00:49:33] Speaker A: Moving on to AI Studio now allows deploying apps directly to cloud run with one click, making it faster and easier to go from idea to shareable app. I did see Ryan visibly cringe when I said click Gemma. Three models can be deployed from AI Studio to cloud run, enabling easy scaling of demo projects to production on serverless infrastructure with GPU support. The new cloud run MCP server lets MCP compatible AI agents like Claude Copilot, Google Gen AI and SDK deploy apps to cloud run, empowering AI assisted development. These integrations streamline the AI app deployment development workflow on gcp. From building and testing an AI studio to production deployments on cloud run scalable serverless platform Cloud runs granular billing and Fury tier makes hosting AI Studio apps very cost effective. Basically from $0 a month with 2 million free requests then you pay for all the requests after that. Automated deployments for AI agents like the MCP server is a different trader versus other clouds that were doing GCP strength in AI key use cases. You might want for this rapid prototyping deployment of AI powered apps, scaling Dremel LLM workloads and AI agent based development. And I just realized reading through this that MCP is like the new click. [00:50:35] Speaker C: Ops in some ways. I mean I'm hoping That that's not exactly what it turns into, but it definitely has that capability in the sense of like, easy button to configure all your integrations. And maybe it is, but yeah, I mean, this is definitely something I want to play around with because it is sort of like I. I like serverless and I'm. I like, you know, coding my applications to run on serverless, so it seems like right up my alley, but I'll have to figure it out. [00:51:08] Speaker A: Yeah, I can see you getting into this one. All right. And Then Google at IO mentioned or announced Imagen 4v03 and Lyria 2 models on Vertex AI. They were able to before, but just on a vertex, which enables high quality creative content generation from text prompts. These models introduce technical improvements in quality, multilingual support, speech audio integration, and greater user control over output compared to the previous versions. Several of these are still in previews, but Lyria 2 is now generally available, so at least one of them got GA'd. And the models integrate with Google's Vertex AI platform and leverage DeepMind's Synth ID watermarking so you can identify any AI generated content along with customizable safety filters. [00:51:48] Speaker C: Ooh, that's interesting. [00:51:51] Speaker A: The watermarking. [00:51:52] Speaker C: Yeah, I hadn't really thought about that. It was like, you know, I. I assumed that they were going to have to do something like that for AI generated content, but the fact they're just building it into these AI platforms, so not at the model level, but in the model. [00:52:06] Speaker D: It's going to have to be there. [00:52:07] Speaker C: Yeah. I mean, it's great. Yeah. [00:52:09] Speaker D: And then you'll have another one that, you know, you'll load and say, remove the watermark, of course. And then it will just remove it because you'll run it locally with, you know, LL Studio and solve the problem. [00:52:21] Speaker C: But maybe it'll. That's where you add back like all the seven fingers or something, you know, just so you still know it's AI created. Yeah. [00:52:30] Speaker A: So I got two Google try to get two Gemini stories in this week. Not going to. You already talked about Pro and Flash. Go away. [00:52:38] Speaker C: We get it. You like it. [00:52:40] Speaker A: Yeah, I know they like to do that. We announced the same thing in three different ways. I missed that one when I was going through shows. All right, Azure is up next. And so build happened, keynote happened. Satya Nadella got on stage and talked and he invited Elon onto stage. And Nvidia CEO. I'm forgetting his name at the moment. And then there was also another guest as well. I only watched about 15 minutes of it before I was done. But thank God for Gemini because gemini built into YouTube, answers all the questions. So I was able to literally just ask Gemini like, hey, in this video, did they announce enhancements to GitHub Copilot that allow gentech code developments and agentic tasks for Ryan's first prediction? And Gemini said yes twice. Full coding agent and agent mode and GitHub copilot. I wouldn't give you a point for GitHub copilot because I did. Hadn't we announced that the week before? They just happened to announce it also in the keynote, but they had not announced the full coding agent prior to this. So that was a point for you, Ryan. Congratulations. [00:53:34] Speaker C: It would make sense. My first prediction that I've ever gotten right would be on build cloud platform I don't use. [00:53:40] Speaker D: You're welcome. Yeah, you're welcome, Ryan. [00:53:43] Speaker A: And also that you it would be AI, which you're not a huge fan of yet. So that's that. It was kind of like bitter. It was kind of awesome. Like I'm like, oh, this is fantastic. Yeah, they did mention the quantum chip, but they gave you no new technology or capabilities or anything that would make you a point here. And I did look for. Did they mention augmented or virtual reality for teams? They said there was no teams announcements. [00:54:05] Speaker C: We can all breathe cyber. [00:54:07] Speaker A: There was a mention of that team. You know that Microsoft 365 updates is the most exciting update since Teams was announced. That was said. That was the closest we got to teams anything. [00:54:18] Speaker C: I think we're talking more about it in a future show. But you know the fact that Google made an announcement for their version, it's pretty funny. Like, oh, I was right subject, wrong cloud. [00:54:27] Speaker A: Yep, wrong cloud. Matt scored a big goose egg. They did not release a new version of the ARM processor, Cobalt. They did give you general availability of Cobalt VMS though, which I don't think I realized was not generally available yet. So you got that? [00:54:40] Speaker D: I don't know that I realized that they weren't generally available. I knew that they were like private or I knew they weren't available in all regions or whatnot. I didn't realize that they were privately still. [00:54:52] Speaker A: Yeah, yeah. So that. So that's not John available. So you can now move all your workloads to Cobalt, a new generation of service hardware. They killed some Surface hardware, but they did not. They didn't talk about the keynote. It was a non keynote item. But they did kill the most interesting Surface tablet they had, which was the Transformer one that would do the tablet and the computer. They did kill that form factor. But no new Surface hardware was announced in the presentation. So swing and miss and then a major update to App Services platform in Azure. Gemini did try to give you a lot of help here and it was trying to like make like, me think that some AI stuff was what you were talking about. But you said app services platform and none of those articles mentioned that. And so I did not give you a point. [00:55:29] Speaker D: No. [00:55:29] Speaker A: If you like to argue that contention, I will let you argue with Gemini because he was strongly batting for you on this one. [00:55:35] Speaker D: There was nothing in there around app service plans or app services. It felt like where they were going with given, you know, hey, we're just going to code for you and everything. I kept waiting for it. [00:55:45] Speaker C: I was like, no, Yep, definitely. [00:55:48] Speaker A: There was like lead up to something, but nothing happened. I had Microsoft launch their own LLM Swing and miss. I would have put money on it almost, but it's a little too early. I still feel it's coming. It just. Maybe not now, but. But I think by the end of this year it's going to happen. It might be its own press event, it might be its own thing. I don't know. [00:56:08] Speaker D: It could be their second conference of the year. [00:56:10] Speaker A: Microsoft Ignite, which is Ignite. Yes. [00:56:13] Speaker C: It's a large breakup with OpenAI. So it's. It's interesting. We'll see. [00:56:16] Speaker A: Well, it's a slow breakup. Right. It's like we weren't. We're sleeping in separate houses. We haven't filed for divorce quite yet. But we're definitely talking about how we might split the assets. You know, there's definitely some conversations, but nothing is. [00:56:27] Speaker C: But we're sticking together for the kids right now. [00:56:29] Speaker A: Yeah, we're sticking together for the kids. Yeah. [00:56:31] Speaker C: And there was 13 billion kids, I think, or what was there a large number of investments. [00:56:38] Speaker A: I said Microsoft Office Copilot will get upgraded with MCP inclusion in it. I got this. And Windows got MCP too. I like, who knew that was going to happen? That Windows now has an MCP connector for it. So you can do all kinds of terribleness to your Windows computer. I can't wait to see how that's going to go badly with hackers. [00:56:54] Speaker C: Oh, I can't wait. [00:56:55] Speaker A: It's going to be awesome. [00:56:56] Speaker C: It's going to be my kids computers, but, you know, not mine. [00:56:59] Speaker A: Yeah. So I got a point there. So now Ryan and I are tied. So things are getting interesting. Ryan dun dun dun. [00:57:05] Speaker C: And Matt, who uses this platform day to day. [00:57:08] Speaker D: I feel like the person that uses it never gets it correct. [00:57:11] Speaker A: This is. [00:57:12] Speaker C: This is true to form and I love it. [00:57:14] Speaker A: I mean, I was pretty good at AWS when we were using AWS to nail these pretty well. But yeah, it does seem to be. If you don't. If you use that cloud, you seem to miss the force of the tree. But yeah, then the last one I have is an Agent Spaces or Glean type competitor. So they did sort of announce this. It already existed though. So this is where I messed up. So Microsoft 365 copilot, I tried when it first launched, which was just like an Office Copilot assistant and it was terrible. And I just have written off in my mind that Microsoft 365 is just garbage for Office and so it's not per the video I watched, it's basically Agent Spaces. But for Microsoft technology, all the Agent Studio stuff, all the ability to do control AI things to connect to all your enterprise business systems, it's all in Microsoft 365 Office 365. So it already existed. I just haven't been paying attention to it because I wrote it off and this is on me. So if I was not an idiot and I had properly known the micro 65, I probably could have got the connectors because they did add all the connectors to different apps because before this week is only connecting to Microsoft apps, but they did add, you know, the model library and you connect to Salesforce and all the things. So I would have, if I had known and I had written this properly, I could have got there, but I got no point on this one, so. But that means we're tied, Ryan. And so we go to the tiebreaker, which was how many times did they say copilot of any form? Microsoft Copilot. GitHub, Copilot Copilot on its own, you know, whatever one you want to mention. They mentioned it exactly 69 times, which I chuckled about because I'm 12. [00:58:45] Speaker C: Yeah, yeah, yeah. [00:58:46] Speaker A: And then I'll do the numbers. And Ryan, that means you have won the Azure first ever Azure prediction show with 62. And again, we do Price is Right rules. So you are the closest without going over. And congratulations. Wow. [00:58:59] Speaker C: Thank you. I. I'm glad it's an audio podcast because my smug face is not very nice. It's. [00:59:05] Speaker A: Yeah. [00:59:07] Speaker D: Congratulations on being able to guess what Microsoft is going to announce. Not sure if that's good or not. [00:59:14] Speaker A: Yeah, secretly. He's secretly a Windows fanboy. We just didn't know. Watch out for Ryan on Azure predictions. He's strong. [00:59:22] Speaker C: Here we are. Yeah. [00:59:25] Speaker A: All right, well, let's get into a bunch of stuff they announced at Build so Azure AI Foundry, your AI app and agent factory, is the end to end platform for building and deploying AI apps and agents. It provides a unified development experience across Visual Studio code collaboration in GitHub and Azure Cloud. It offers a growing catalog of state of the art AI models including Grok 3, Flex Pro, Sora and 10,000 plus open source models from Hugging Face, a new model router optimizing model selection. The Azure AI Foundry Agent service is now generally available enabling designing, deploying and scaling production grade AI agents and integrates with over 1400 enterprise data sources and platforms like Microsoft 365, Slack and Twilio. What no teams Multi agent orchestration allows agents to collaborate on complex workflows across clouds. Agentic retrieval and Azure AI search improves answers relevance by 40% for multiple questions and enterprise great features like end to end observability, first class identity management via Microsoft Entre Agent ID and built in responsible AI guardrails. Foundry Local is a new runtime for building offline cross platform AI apps on Windows and Mac. Integrating with Azure ARC enables central management of edge AI compared to AWS and gcp. Azure AI Foundry offers tighter integration with Microsoft's developer tools and enterprise platforms and it targets customers building enterprise AI workflows. Pricing details were not provided but likely varies based on usage of computing resources, API calls and data processing. And some features like fine tuning do have a free tier included. So that's one of the things they get summarized here. But yeah, you can actually fine tune the model using your own data, which is kind of nice. So as a user, so basically I can log into my five and say like this is, this is important to me in my context of my job and I want you to fine tune your results to me. [01:00:59] Speaker C: Interesting. So is that like, is that. Are they. They prioritize the rag ability. Like that's kind of. Kind of. [01:01:05] Speaker A: It's not rag. This is, this is reinforcement learning against it based on like hey, here's your email box and here's all the emails you send to the people you talk to and the context that you have around that and how do I use that inside of my AI contexts? Like that. So yeah, that's cool. [01:01:18] Speaker D: Like emails from your CEO you can say are more important than emails from your cto. [01:01:21] Speaker A: Yeah. Or like you know, you always email technology people and you do these things and you have a certain way you write your emails that you want the model to follow the process or you know, things like that you can now do so. Yeah, yeah. Impressive for sure. [01:01:35] Speaker C: This is exactly what I've been looking for is these types of changes, right? Specifically around the email and Office365 workspace stuff. Just because that's both in my corporate life and my personal life. That's what we use. And so I want all of these things to be automated away. Cool. [01:01:58] Speaker A: Well Powering the next AI frontier with Microsoft Fabric and Azure Data Portfolio for those of you who are not familiar with Fabric, it's basically Power Bi and some bunch of stuff that looks very similar to BigQuery and Snowflake and data services are a bunch of APIs to access that data and do transforms and ETLs. [01:02:15] Speaker C: And Justin is tired of me asking what fabric is every single I know. [01:02:19] Speaker A: That's why I'm now doing it first Data servers are being enhanced the power of the next generation of AI applications that combine analytical, transactional and operational data in structured and unstructured forms. Cosmos DB NoSQL database is now available in Microsoft Fabric to handle semi structured data for AI apps. In addition to SQL databases starting at only 25 cents an hour for serverless instances, there's a new Digital Twin Builder low code tool allowing creating virtual replicas of physical and logical entities to enable analytics, simulations and process automation. Power BI is getting a new Copilot experience to allow users to chat with their data and ask questions and this will also be integrated into Microsoft 365 copilot. SQL Server 2025 Preview is adding vector database capabilities and integrations with AI frameworks like LangChain to power intelligent apps. Pricing varies on this one by Core and Addition. The Postgres extensions for VS code now include GitHub Copilot for AI assistance writing queries and Azure database for PostgreSQL adds high performance vector indexing as well. Azure, Cosmos DB and Azure Databricks now integrate with Azure AI Foundry to store conversation data and power AI solutions. And Microsoft is partnering with SAP on SAP Business Data Data Cloud and SAP Databricks on Azure initiatives to help customers innovate on SAP data and these enhancements. Azure as a leader in converging databases, analytics and AI compared to point solutions from AWS and GCP throwing shade targeting enterprise customers building next gen AI applications. [01:03:39] Speaker D: I mean the big thing here is. [01:03:41] Speaker A: Cosmos DP is now part of Fabric. [01:03:44] Speaker D: Yeah like that to me felt like a little bit of a gap in the past like it was. They were so focused on structured data but the unstructured data piece still is something that's critical in my opinion. So them missing that, you know, and finally adding it, I mean finally it's like fabric's only like 2 years old. Feels pretty big. [01:04:05] Speaker A: Ryan, still don't know what fabric is. [01:04:07] Speaker C: I got. I still don't know what fabric is and how I would use it or what I would use it for. I do find it funny with all these things they mentioned, you know, like SAP business and other things. It's like the, the next generation of cloud stuff. Yeah, I don't think so. [01:04:22] Speaker A: All right, next up is Microsoft Discovery, which is a new enterprise AI platform that aims to accelerate research and development R and D. By nailing scientists to collaborate with specialized AI agents and graph based knowledge engines. It can help drive scientific outcomes faster and more accurately. Discovery integrates with Azure infrastructure and services to provide enterprise grade trust, compliance, governance and extensibility. And researchers can bring their own models, tools and data sets. It also leverages innovations from Microsoft Research and will integrate future capabilities like reliable quantum computing. This is where you mention the keynote. Well your discovery in quantum. But the platform introduces a new agentic AI paradigm where people and AI agents cooperatively refine knowledge and experimentation iteratively in real time. The AI can deeply reason over nuanced scientific data specialized across domains and learn and adapt. Aws, NGDP offer some AI ML tools for research. Micro Discovery appears to be more comprehensive specialized platform focusing on the full R and D lifecycle and scientific reasoning. And the agentic AI approach here does seem to be interesting. [01:05:18] Speaker C: Yeah, it's going to be years before we realize like the. These improvements. Right. Because it's the R and D side of these things are going to, is the advancements are going to come at a faster rate just in general technology AI side and it's going to be neat to see how those things change, you know, those industries or I guess not really industries but those things change as things go on where I think that'll be easier for scientific discovery. [01:05:45] Speaker D: Yeah, I mean this whole discovery product, whatever it's called, you know, platform is, is interesting among itself because the way they even presented it on the keynote was you know, how it like self discovered a new lubricant and like giving it, you know, a segment of stuff and telling it to kind of help do that deep investigation of R and D research and learning and then kind of come up with new molecules or come up with new ideas that you've never even thought of and kind of working it in that way. It's Very interesting, you know, way the AI is going and kind of going into AI self learning and self discovery things. [01:06:27] Speaker C: All I want is for AI to cure cancer. [01:06:30] Speaker D: It'll get there, don't worry. I'm sure then it will kill all of us in a different way, but don't worry about that. [01:06:35] Speaker A: Yeah, well, and then, and then finally they decided to come after us with Agentic DevOps so GitHub Copilot is evolving into an AI powered coding assistant that collaborates with developers across the entire software development lifecycle from planning to production. The new agentic DevOps approach reimagines DevOps by having intelligent agents automate and optimize each stage while keeping developers in control. Agent mode, Good luck. Agent mode and getup Copilot can analyze code bases, make multiple file edits, generate tests and fix bugs, and suggest commands based on prompts. The new coding agent and Copilot acts as a peer programmer taking on code reviews, tests, bug fixes and feature specs so developers can focus on high value work. And Azure is adding app modernization capabilities co to assess, update and remediate legacy java.net and mainframe apps. Reduced technical debt. The new Azure site Reliability Engineer agent monitors production apps 24 by 7, responding to incidents and troubleshooting autonomously to improve reliability. And GitHub models makes it easy to experiment with and deploy AI models for various providers right from GitHub with enterprise guardrails. Microsoft is also open sourcing the GitHub Copilot extension in VS code reflecting their commitment to transparency and community driven AI development. And these agentic AI capabilities remove friction, reduce complexity and change the cost structure of software development while enabling developer creativity. So yeah, lots of DevOps type features here on this one. [01:07:49] Speaker D: Yeah, I mean during the keynote they talked about like hey, there's a production outage and it automatically goes and scales and fixes it and then makes an issue that then it can self fix with their GitHub copilot agent. I'm like it's really terrifying that it's going to do it and you're going to wake up all of a sudden to an Azure bill of like $400,000 because be like hey, there's a problem with your SQL so let's just grow your SQL so your SQL's not a problem. Cool. All of a sudden I'm running 128 VCORES on my SQL hyperscale cluster because you know, someone's ddosing me feel like there's going to be things he's going to miss. [01:08:29] Speaker C: That's how we, you know, we, we ran into those rough edges with auto scaling to begin with. Right? Like it's like, oh, it's scaling out of control. And this is going to scale out of control unless we give it the prompts that say, hey, be reasonable. [01:08:41] Speaker D: I think it just does it though. I don't think this is like a prompt thing. It just fixes it itself. [01:08:48] Speaker C: No, I mean it's whether it's a prompt or whether it's built into models, like it's, it'll be fixed because true, you know, Azure is going to have to like give a whole bunch of money back, you know, for, for things that scaled out of, out of craziness. And so, you know, they'll, they'll want to fix it from the product side and then customers will want sort of the guardrails. But, you know, I think that this is, this is the future that I'm here for, which is, you know, the SRE agents responding to on call so I don't have to, you know, like, that's fantastic. And everyone knows I wasn't going to answer that page anyway. [01:09:24] Speaker D: We know we have a hard enough time trying to get you for the podcast. [01:09:28] Speaker A: Only if he's napping, which is most time. All right, we've made it to the last story of the night. It was a long one. I know. Oracle is our last story of the night. Finished off nicely, right? Oracle is launching E6 standard compute powered by AMD Epyc. They're claiming up to 55% better price performance compared to similar compute offerings. But specifics are vague and comparisons likely cherry picked as usual from the new for the new E6 standard compute instance. [01:09:58] Speaker C: Or from Oracle in general. [01:10:00] Speaker A: E6 instances are supposedly ideal for workloads like web servers, application servers, batch processing and distributed analytics. But these are generic use cases that any major cloud provider can easily handle. Oracle touts security benefits from using in house design servers with built in firmware level security and improvement but likely table stakes compared to security from all of the other cloud providers. By the way, AI wrote this and I did tell it to be cynical. [01:10:19] Speaker C: About Oracle, so I was going to say like this. [01:10:22] Speaker A: I'm surprised by the tone this is. AI might help on this one. E6 instances offer up to 128Ocpus two two two terabytes of RAM and one petabyte of remote block storage specs that match or trail other cloud providers. Oracle claims E6 is the best price performance in the industry for scale out workloads. A bold claim. It warns Deep skepticism. Okay, maybe we need to tone that down a little bit. [01:10:43] Speaker D: No, no. [01:10:45] Speaker C: I don't know. I mean, I think it's the right level. It's just I was surprised from AI, but now that I know that it was specifically told to, you know, I. [01:10:52] Speaker A: Told it to be cynical and emulate us. Yeah, yeah. To be us. Yeah, that's the goal here. [01:10:56] Speaker C: Now this is about the right level. [01:11:00] Speaker A: But yeah, so that's it. Anyways, new new instances, 75.075 per OCPU hour is basically the first starting place, which will get you nothing. And then it'll go up from there quite quickly to bajillions of dollars. Especially if you add Oracle Server to it, Oracle Database to it. [01:11:16] Speaker D: What else are you going to run on? [01:11:17] Speaker A: Oracle Financial ERP on top of database. I mean, the thing about all Oracle products is it's just Oracle Database all the way down. Different thing on top of it at some point. But yeah, you're right, it's just Oracle. All right, gentlemen. Well, that is it for this week. That was a lot. [01:11:37] Speaker C: That was a lot. [01:11:37] Speaker A: I appreciate you guys hanging in there with us on that. [01:11:39] Speaker D: Yeah, next time it's a lot. Let's not make a bad life choice and start late. [01:11:46] Speaker C: Yeah. This is a culmination of errors this week. [01:11:49] Speaker A: Yeah, this is. Justin is traveling. Matt has a new baby. There's sleep schedules. He's traveling back from relatives, Ryan's sleeping too late. You know, just all the bad things. He conspired against us all. We appreciate it. All right, guys, we'll see you next week here in the Cloud. Hopefully well refreshed and rested after our 3D weekend. [01:12:09] Speaker C: Absolutely. Bye, everybody. Bye. [01:12:12] Speaker A: Bye. [01:12:12] Speaker D: Bye, everyone. [01:12:15] Speaker B: And that's all for this week in Cloud. We'd like to thank our sponsor, Archera. Be sure to click the link in our show notes to learn more about their services. While you're at it, head over to our [email protected] where you can subscribe to our newsletter, join our Slack community, send us your feedback and ask any questions you might have. Thanks for listening and we'll catch you on the next episode.