[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.
[00:00:18] Speaker A: Episode 298 recorded for the week of March 25, 2025. BigQuery gets it with DevOps Good evening Ryan and Matt. How you doing?
[00:00:28] Speaker C: Hey there.
[00:00:29] Speaker D: Good, how are you?
[00:00:31] Speaker A: You know it's a lovely middle of the week. Ready for Friday already just got back from pto so you know, it's like five day work week shoved into as few days as possible, so it's tough.
[00:00:42] Speaker D: Yeah, short work weeks are a lie. You just do the same amount of work in most days.
[00:00:47] Speaker A: Yeah, pretty much. Which you know, is what it is.
Well, we have a bunch of stories though to get into, so I guess we should probably get right to it. I was saving for Jonathan to come back, but he's still out for more weeks here so Manus is a new AI agent from China and it's going viral and raising some big questions. Manus is being described as the first true autonomous AI agent from China, capable of completing weeks of professional work in hours. Developed by a team called Butterfly Effect, with offices in Beijing and Wuhan, Manus can function as a truly autonomous agent that independently analyzes plans, executes complex tasks. System uses a multi agent architecture powered by several distinct AI models including Anthropic's Cloud 3.5 Sonnet and fine tuned versions of Alibaba's Quinn. Unlike traditional chatbots, Manus can work on different tasks without needing frequent step by step instructions. Continue to work in the background even when users close their computers. A unique feature is the Manus computer window, which allows users to observe what the agent is doing and intervene at any point. The company claims that Manus outperforms OpenAI's deep research tool on the Gaia benchmark, a third party measure of general AI assistance. Early testing has shown Nick's results with some reviewers were impressed, others encountered bugs, error messages and failures on practical tasks like ordering food or booking flights. This remains difficult to access due to limited server capacity, creating a scramble for innovation invitation codes which were reportedly selling for thousands of dollars in the Chinese reseller apps. Well, that's great. Manus has announced a strategic partnership with Alibaba's Quinn team to help deal with surge in traffic and expand its user base. The emergence of Manus is raising questions with the global AI landscape, with some comparing it to January's Deep Seek moment and questioning whether China has leapfrogged the US and AI development. Privacy experts have raised concerns about data protection. So don't put your private datas in the Manus trap out.
[00:02:26] Speaker D: They're really giving ChatGPT all your private information. I mean, everyone's telling you, right? Like, sure, I don't want to give it to China, but, you know, I also don't like giving to OpenAI either.
[00:02:38] Speaker C: So, you know, yeah, I just like this new trend in the agentic AI. Like, this is kind of what I've been waiting for because it's. You'd start off with AI and you're like, oh, rewrite this and give me this, or, you know, do the tone analysis of this. And then you very quickly hit, like, the, you know, stumbling blocks of, you know, larger things. And so playing around with these new features where you can ask it to do a task and it goes off and does long term is really going to be, you know, what replaces us as, you know, worker bees. So cool.
[00:03:14] Speaker D: Yeah, I do kind of think that China's figuring out how to do all this on, in theory, lower quality, lower optimized, lower efficiency. Doing a podcast this late at night probably is not the best life choice, but lower, you know, speed, information, everything, you know, hardware and is doing it. So there is, you know, to me, I'm sitting here going, yeah, they might be starting to leapfrog because they're able to do a lot with a lot less.
You know, hopefully the clauds, the OpenAI and these companies look at what they're doing and kind of reverse engineer it and put it into these. And that way, you know, hopefully all these different companies learn from each other. So we're not recreating the wheel 15 times over and over again.
[00:03:58] Speaker C: Yeah, but my wheel's slightly better and you can pay me for it.
It's rounder.
[00:04:04] Speaker D: What does it say? Is your wheel around or is it a hexagon? I'm at Octagon, so I'm almost round.
[00:04:10] Speaker C: Yeah.
[00:04:12] Speaker A: All right. Cloudflare has an interesting way to fight those nasty AI bots. If you've hosting a website at all, you might have noticed that all of a sudden your website's being hit by a lot more bots trying to get information to answer things like deep research queries from Google Gemini. And so Cloudflare has announced a new feature called AI Labyrinth that aims to combat unit AI data scraping by serving fake AI generated content to bots, which is great. The tool will attempt to thwart AI companies that crawl websites without permission to collect training data for LLMs that power AI systems like ChatGPT. Instead of simply blocking the bots, Cloudflare's new system lures them into a maze of realistic looking but irrelevant pages, wasting the crawler's computing resources. The approach is a notable shift from the standard block and defend strategies by most website production services. And Cloudflare says blocking bots sometimes backfires because it alerts the crawler operators that they've been detected. When Cloudflare detects unauthorized crawling, rather than block the request, it will link the bot to a series of AI generated pages that are convincing enough to entice a crawler to traverse them. But while real looking, the content is not actually the content of the site they're protecting. So the crawler wastes time and resources. Data is automatically generated by its worker AI service, a commercial platform that runs AI tasks.
[00:05:17] Speaker D: So we're feeding, we're leveraging AI to give AI false information, which will then feed into the item that creates the AI fake content. Anybody else see a problem with this loop?
[00:05:33] Speaker A: Yeah, I mean, it's going to just poison to AI data later, which would be great. Yeah.
[00:05:37] Speaker C: Is the hallucination in the model or is it, you know, the bad data that's being fed? It'll be interesting because I guess we're going to need AI robots Txt, because this was a big, you know, fight in the early days of big search indexers.
[00:05:49] Speaker D: So do you guys remember there was a theory when, remember this from the beta of the AWS networking exam where there was questions about. It was called like DDoS, but it wasn't like a DDoS attack. It was like in order to cause you to spin up resources and kill all your computer, like, and make your bill go up. That's essentially what I feel like Cloudflare is doing is like a boss, you know, DOS attack to force these companies to just burn money.
[00:06:22] Speaker C: It totally is. I mean, but there's gotta, you have to, there has to be some sort of tool that you can protect your data and yourself. So I kind of, I feel I'm very conflicted. Right. Because I want my DC research query to go back and answer my questions. But I also, you know, I understand why people would want to implement this on their data and their site.
[00:06:40] Speaker A: Yeah, well, especially if you're like, you know, have put a bot block in place for this, to prevent this and it still does it. I mean, like, definitely that's a problem. You definitely don't want that to be the case.
[00:06:52] Speaker D: Yeah.
[00:06:54] Speaker A: All right. Well, OpenAI has long believed image generation to be a primary capability of their language models that's why they have built the most advanced image generator yet with the GPT4O. The result image generation that is beautiful but useful. Now I, while you guys are just talking, decided to throw in a prompt into ChatGPT using the new 4.0image and model. And I said create a fun cartoon version of the four main hosts of the cloud pod, excluding Peter. Put a tag under the amazing CloudPod hosts and then use this website for picture reference, which was our About Us page. And I'm now pasting into our show notes what they've generated.
[00:07:30] Speaker C: Oh, no, I'm so afraid so clearly.
[00:07:35] Speaker A: You didn't look at the website at all, which is fine. I kind of expected like I tried it with Gemini too, just out of curiosity. And it's that I can't recreate humans, which is a silly thing, but I know why they do that. But I love that this, this image which would now be the show. The show cover page.
[00:07:49] Speaker D: Yeah, of course, 100% you should be.
[00:07:51] Speaker A: Yeah. So I don't know who those people are, but they. Look, I don't know those people are either, but you know, I love that it's an amazing cloud pod and then hosts is like some weird AI ghost thing going on. And then it says excluding Peter.
She's the best.
There's a couple.
[00:08:08] Speaker D: One guy has like two mics coming.
[00:08:10] Speaker C: Out of the back of his head.
[00:08:11] Speaker A: Yeah, that's a real weird too.
[00:08:14] Speaker D: One guy doesn't have a mic and the guy on the left kind of reminds me of Peter, I'm not going to lie.
[00:08:18] Speaker A: Yeah, he does look a little bit like Peter. So, yeah, it sort of reminded me, but it was sort of hilarious to me that that's what joined. So, yeah, there's a new chat GPT4.0 image generation capability. Good luck.
[00:08:31] Speaker C: To be fair though, it actually spelled out words which it usually can't do very well.
It's not perfect, for sure.
[00:08:38] Speaker A: It's not perfect.
[00:08:38] Speaker C: Yeah.
[00:08:39] Speaker A: But in the actual blog article, you can see where they gave a bunch better instructions than I did. I did not give it nearly as many as they did. To create this person on the whiteboard with a reflection, you know, clearly that they're in a San Francisco office. There's a reflection on this glass whiteboard of, you know, the San Francisco Bridge, which. That's a golden gate in that picture and I know where that picture had to be taken from. And that's a park, but it's fine, whatever, you know. And then they have a basically, you know, do the same photo, but a selfie view from the photographer and it pops in a different fake human generated person. So, you know, definitely there's some things I could do that are cool. It has some more image consistency items. Has a couple other nice multi ability to say like, oh no, I like this cap. But you know, add a hat and a glasses to it and it'll do that for you. And you can do all kinds of different things. So it does look better than what they were giving you before. Just don't try to do real humans or pornography because that's also not allowed for the safety requirements.
Which makes sense.
[00:09:33] Speaker D: Now I'm looking at that picture being like it almost looks like it's on the wrong side of the bridge too. So I'm even more confused where it would be because that would have to be in the water. I think by Baker's speech.
Right. Because isn't that looking in. And also there's no bridge in there.
[00:09:48] Speaker A: No, it's definitely looking away from the city. Because you'd see the city skyscrapers, wouldn't you, in that?
[00:09:54] Speaker D: No. Unless it's looking towards him. Run.
[00:09:57] Speaker A: I mean, there's also the fact that there's no actual bridge deck on part of that.
[00:10:00] Speaker D: That's what I was actually looking at. No, no, there is. If you look closer. I zoomed in. There is. It's a little faint. I was a little confused at first.
[00:10:08] Speaker A: Yeah, I mean, I guess. Yeah, that could. I don't know. Yeah, we're taking too much time thinking about this.
[00:10:13] Speaker C: Yeah.
[00:10:15] Speaker A: It's already incorrect either way because yes, you're right. It's either Baker's beach or it's a middle of Golden Gate Park. It's one of the two. Both are bad.
[00:10:21] Speaker C: I think you guys got the wrong bridge.
[00:10:24] Speaker A: I think it's the Bay Bridge.
[00:10:25] Speaker D: Yeah, you think it's the Bay Bridge?
[00:10:27] Speaker C: I don't think that's the Golden Gate.
[00:10:28] Speaker A: I mean, it could be the Bay.
[00:10:30] Speaker D: Bridge because Golden Gate only has and.
[00:10:34] Speaker A: Yeah, that could be Treasure Island. That could make more sense.
[00:10:37] Speaker C: Could be Treasure Island.
[00:10:38] Speaker D: Right. Because there's only two on the two pilings on the Golden Gate Bridge, right?
[00:10:43] Speaker C: I think so, yeah.
[00:10:44] Speaker A: There's only two. This one has three. Three. Which I think. I think that Bay Bridge has four. That part of it. I mean, I don't think it's a.
[00:10:49] Speaker C: Real bridge, but I mean, that's possible.
[00:10:54] Speaker A: I mean, they do say overlooking the Bay Bridge. So it would be technically. So they didn't say the Golden Gate. So this stood technically be the correct bridge. Okay. That is their prompt so, so then.
[00:11:04] Speaker D: You'Re in the middle of the water.
[00:11:08] Speaker A: No, no, now you're at the Ferry Building. Yeah, I think it's from the Ferry Building area.
[00:11:12] Speaker D: I was thinking the reflection. So the reflection part is what screwed me. Yeah, the reflection. You'll be like fairy building ish. Yeah, okay.
[00:11:19] Speaker A: Yeah, Fairy Building ish, I would think would be where that photo's from. Yeah. All right. And again, moving on.
OpenAI is also launching a new Speech Text and Text Speech audio model in the API, making it possible to build more powerful, customizable and intelligent voice agents that offer real value. Latest Speech Text models set a new state of the art benchmark, outperforming existing solutions and accuracy and reliability, especially when dealing with accents, noisy environments and varying speech speeds. These enhancements are in the GPT4O transcribe and GPT4O mini transcribe models with improvements to word error rate and better language recognition and accuracy compared to the original whisper models, which if it can handle accents, that's really great.
I'm mostly excited because we use a transcript that we, you know, basically provide to our show note writer to help her write the show notes and then she'll, she'll pull quotes out of it for our show notes. And sometimes I'm looking at this, I'm like, that doesn't make any sense to me. What is that? And then I'm like, I figure, oh yeah, she just copied it from the transcript, which is fine, you know, more power to her. But the transcript was wrong. So I'm hoping maybe we have some less errors because I think Riverside uses ChatGPT for their AI stuff. So yeah, maybe we'll get some better quality and see.
All right, moving on to cloud tools. Valky's 8.1 performance gains disrupt in memory databases. Wow, that's a bold statement. Article on Valky this article on Valkey caught my eye as it's been a year since Redis announced they were dumping the BSD3 clause license and adopt the RSALV2 which is the Redis open source license and then the SSPLV one which we all know is the anti cloud one which birthed the VAL key fork. Apparently the Valkey fork is turning out to be highly successful for Percona research paper with 75% of surveyed Redis users are considering migration due to recent licensing changes and of those considering migration, 75% are already testing, considering or adopting Valky and third party Redis development companies like Retison are supporting both Redis and Valky now. It's not Just the licensing that's driving the change. But the Linux foundation member summit said the Valky is a far faster thanks to incorporating enhanced multi threading and scalability features. This wasn't the original plan as they wanted to keep the open source spirit, but also wanted the value to be more than just a fork. Initially, at the first tutorial summit in Seattle where they got the developers together, the users tried to figure out what this new product would look like. At the time it was expected to focus on caching, but users said they wanted more. With Valkey becoming a high performance database of all sorts of distributed workloads and although that would cause a lot of complexity, the new core team took it on because why not? You're not getting paid for this. Make it more complex. They were apparently successful though with Valky 8 redesigning the Redis single threaded event loop threading model, the more sophisticated multi threading approach to IO operations which resulted in a 3x improvement in performance as well as a 20% reduction in the size of the separate cache tables. Beyond that, they have been improving the core by adding rust to add memory safety as well as changing internal algorithms to improve reliability and failover times as well as they have rebuilt the key value store from scratch to take better advantage of modern hardware. Based on the work done at Google and others, a ton of this will be coming out as part of Valkey 8.1 which is shipping any day now.
[00:14:10] Speaker D: How to get a really good software package, Piss off a bunch of developers and tell them that they can't do better. And that's Valkyrie.
[00:14:19] Speaker A: Yep.
[00:14:20] Speaker C: I mean it's amazing to me just how fast they're, they're able to add this level of improvement. You know, like Redis has been around for a while and it's kind of crazy, but I guess that's, you know, the, the difference between small open source sort of contribution.
[00:14:37] Speaker A: Well, I mean occasionally, occasionally I'll find I'll get linked on Hacker News to a thread about some dispute that is happening in the Linux kernel open source community. And I've read through these email threads and you see the bureaucracy at work of these large massive things. And so it's very possible that there were people who wanted these things. Maybe Amazon wanted it, maybe Google wanted it, maybe Microsoft did. And you know, the Redis core community was. I know they're not supreme about yet but you know the, the core community, there's people who want one thing and people who don't and then basically answer is well if you don't like it, fork well, they didn't like it and they forked. And then people kind of thought, well we'll just keep doing what we're doing. And then people, those people who had to no voice before came in and said, well we're willing to put resources behind this thing and do that, then that's a win win for everybody. So I'm actually kind of glad to see this one worked out. Now I would say this is not common. I don't think this is happening with open tofu, I don't think this is happening with open bow, etc. Because in this case the Valky people who started this are all very highly paid engineers at Hyperscalers who have a vested interest in making this work and get paid to do it. Where a lot of open source projects are volunteer efforts.
[00:15:46] Speaker C: Yeah. So this is more. This is their day job. Yeah. For the various. Even though the contributions won't be IP in the end. That's. But that's, I mean, I think it's great because we're all sort of able to benefit from that and I love that model.
[00:15:59] Speaker D: And the performance improvements here are massive. Like, yeah, like Google had their, like a blog post we talked about insert some time ago between three months and a year I guess, where they were like, look, we're already seeing 30, 40% improvements and you know, here's all the details and they're expecting more. Like it's pretty amazing what they're able to do now with this and they keep improving. Redis is going to just slowly die off due to their own causes.
[00:16:30] Speaker A: I mean, sort of feels like elasticsearch.
[00:16:32] Speaker D: Talked about that a little later.
[00:16:34] Speaker C: Yeah, I mean, we'll see. Elasticsearch is trying to come back with changing the licensing model to be more. Yeah. Less restrictive.
But yeah, I keep seeing open search pop up and you know, other products.
[00:16:47] Speaker A: Yeah, well, I did see that the February 27th earnings report, they actually up 17% year over year on revenue at Elastic. So clearly it's not hurting them enough. But maybe in the future we'll see.
[00:17:01] Speaker C: Well, it's good. I mean I'd like to see that company succeed if they reverse course. Like because it's, you know, I, I do like the product. I do think it's cool.
[00:17:07] Speaker A: I mean if you, if you can reward like you made a bad choice, you went to sspl, everyone got mad and was running away and then you realize that was a bad call and you come back, you know, apologizing to the community and then you do. Well, I'm okay with that. But I don't think they apologize. That's the problem.
[00:17:20] Speaker C: I don't think they have either.
[00:17:23] Speaker A: Close, but no, no cigar. All right, let's move on to aws. First up, starting today, you can get more granular visibility of geography on the AWS website. Amazon says that due to data sovereignty, the need for more details is super important when talking about AWS regions and availability zones. And so they've added geography to the AWS regions and Availability zone page. And in case you didn't know, Virginia is in the United States of America, which is what the geography tells me. So this is a full blog post on this full blog post by Prasad Rao. I know no shame to Prasad, but literally this. All they did was add geography to the region and availability pages so you can see where the region, what country the region's in, basically. So if you didn't know, Jakarta region is in Indonesia, Yosaka region is in Japan. Now, I appreciate that the simplicity of this is on one webpage, in case I did not know this, because I. No shame to people they may not know, but also, this is not that impressive for a full blog post.
[00:18:25] Speaker C: I was waiting, you know, reading through this, reading through the show notes and reading through this, I was waiting for the cool thing, right? It turns out it's just as simple as I thought. So. Yeah.
[00:18:36] Speaker A: But like, I was like, okay, well, maybe they put it in the API too, because maybe, you know, if you're a command line junkie like Ryan is, maybe you'd want to know this at the command line. And there's no mention of the cli, so I didn't. I didn't have a chance to test the cli. It might actually be there, too.
But the fact they didn't mention it makes me think they didn't do that. They just put it on the website.
[00:18:52] Speaker C: Yeah, and I don't even remember, like, they had the. They had the site name on the website, but I think in the cli, you could just get basically the zone name, not the.
Any other identifiable information.
[00:19:08] Speaker D: Okay, so the only thing I can think of here is the legacy regions. Thinking, like, you know, South America, right? It's just sa, South America. Like, there was no nothing else there. Like, so now, like, it says Brazil. Granted, like, you could find it in the console just by pressing on SA and said the word Brazil next to it. Those are the only things I could think of that, like, why they would add this.
Like, the new regions are like, Mexico, me, I think. Or no, that might be. All right, well, like, I think clearly I'm winging it here. But like, you know, mx, I can.
[00:19:48] Speaker C: Tell you why they added it. Some big customer got some fine active hues from GDPR and got mad at Amazon because we did. They didn't know that Jakarta was in Indonesia.
[00:20:01] Speaker A: I did just check the API call for AWS EC2, describe regions output. It just says region name, EUs3, endpoint, EC2, EUs3 and then opt in status if you have to be opted in or not. Nothing about geography there. They did actually mention a client in here. I don't necessarily know this client that was the one who needed it, but they said, for example, FanDuel, a leading sports gaming company based in the U.S. is scaling into new markets across the U.S. and Canada. They're taking advantage of the improved geographic transparency, make more informed decisions and ensure their meeting data residency requirements as they scale their business quickly. Now, they didn't give us a quote from the fanduel, but they did call them out. Someone who needed this.
[00:20:38] Speaker C: So I believe FanDuel allows you to do gambling. And so FanDuel ran afoul of some sort of local law and now we have this.
[00:20:48] Speaker D: Well, there's no local zones in here.
[00:20:50] Speaker A: I do say this is important for customers in highly regulated industries such as the financial industry or gaming. So yeah, Maybe.
[00:20:56] Speaker D: So maybe FanDuel didn't know that US East 1 is in Virginia and in Virginia they can't do gambling, so they got a fine there and they can do it in Ohio. So now they know that US East 2, it's in Ohio, that's acceptable. I'm making up what the rules are.
[00:21:13] Speaker C: But it's going to be where the client. Oh, this is going to be a mess. Yeah.
[00:21:19] Speaker A: Anyways, this is here. If you need this feature. We'd love to hear from you. Actually, if this is you and you know why this is important, I'd love to hear about it. But it seems kind of obvious to me that Virginia is the United States of America. And I knew it was in Virginia because that's where the region was called originally. So your idea that Ohio was unknown because it was called the Ohio region.
[00:21:37] Speaker C: So I don't know, I mean, I get it. I get confused with like, you know, EU3 or West3 or something like that.
[00:21:43] Speaker A: I get confused by the availability, you know, the. Yeah, the region monikers. Because I don't remember all the weird, like, is this the third US East Region or is it the fourth? I don't remember because I can't keep track of them.
[00:21:53] Speaker C: You know, I have no problem with them adding the information. But you're right, it's just like is it. Why is this a blog post?
[00:21:59] Speaker A: Why was it even mentioned at all? Like this is even worthy of a. What's a new blog post? This is a full Amazon Web News blog post.
[00:22:05] Speaker C: So there you go.
[00:22:07] Speaker A: Well, AWS is announcing that Amazon Q and Quicksight unlocks the ability for any employee to perform expert level data analysis using natural language. The new scenarios capability for the Amazon Q and QuickSight uses an AI agent to empower all employees to engage by a natural language to perform data analysis without specialized skills or expertise. And I have a quote here from Dilip Kumar, Vice President of Amazon Q Business aws. We are at the beginning of a workplace transformation driven by agents and Amazon Quicksight is pioneering how this technology can break down the technical barriers between employees and their data. The new scenarios capability. Everyone becomes their own aid analyst who can dive deep into the company data, helping them unlock insights, make better decisions and explore countless possibilities faster than ever before. Or you can misquery the data and accidentally connect two pieces of data and get a bunch of wrong data, make a bunch of actions and cost your company millions of dollars mistake from a bad strategy because you didn't know it true direct data or understand the data model. So there's that risk.
Yeah. Or the fact that you bankrupted your company by choosing Quicksight over, you know, minor things.
[00:23:09] Speaker C: Yeah, I mean using these tools like I love doing, you know, data projects and, and getting large metrics and Justin loves fancy graphics and, and charts. So you know, like this is match made in heaven. But the problem is always the devil's in the details with these things. When you ask it a natural language query and it's trying to compare two different types of data and it's like well, you have to build in mapping tables half the time. And so I don't know if this is replacing any in depth bi sort of teams yet.
[00:23:42] Speaker A: Well then also okay, you don't have access to some of the data, but you have access to this data. But if you had that other data, this would actually make that part more valuable. So then you really should be talking to the data analysts who can get you access to that data data or with the, or knows how to get you access to data. Like hey, you know if you talk to those guys over there and you got they give you permission to use our data, you could actually make your data better. Like there's all these things that data analysts can do that are important because they understand the data model at an even higher level than you do. And you're not limited to just what you know in your systems.
[00:24:12] Speaker C: That said, constructing queries in the widgets in in quicksight is a, an abomination. And so like anything they can do to make that better is AOK by me.
[00:24:23] Speaker A: Yep, 100%.
[00:24:25] Speaker D: They ever integrate that more into the console. So it's not like this, like one off.
[00:24:29] Speaker A: No, it's still. I think it's still one off. I think it's my design that way too. Because people who want to run reports on Quicksight don't necessarily want to be AWS console users.
[00:24:38] Speaker D: Yeah, that makes sense.
[00:24:39] Speaker C: Yeah.
[00:24:41] Speaker A: Aws is announcing OR2 and OM2 instances for the Amazon OpenSearch service. These expanding OpenSearch optimized instance family. The OR2 delivers up to 26% higher indexing throughput compared to the previous OR1 instance instance and 70% over the R7G instances. The new OM2 instances deliver up to 50% higher indexing throughput compared to THE OR1 instances and 66 over M7G instances in their internal benchmarks. So if you're using the Amazon flavor of elasticsearch, you should definitely take a look at these. These again are the search OpenSearch optimized instances which use best in class cloud technologies like Amazon S3 to provide high durability and improve price performance for higher indexing throughput and better for indexing heavy workloads. So it uses a combination hybrid architecture of S3 and instances to make your life better.
[00:25:26] Speaker C: It's funny to see these announcements years after, you know, running a giant elasticsearch project for a while because like, these are all the struggles and they're, they're getting addressed through open search and Amazon running a giant, you know, farm of these things. So it's cool because these things really do solve a lot of problems. I remember Graviton came out, you know, we're con. We were constantly yelling at that we're getting yelled at for cost and Graviton came out and we made the changes there to adopt Graviton and we all of a sudden we're saving 24% of our bill. So I imagine this is the same type of change for the people that are using Amazon OpenSearch. They'll get a bunch of capacity back and maybe can reduce the size of their instances or send it more data. Nice.
[00:26:11] Speaker A: Well, I'm pleased to announce that Amazon Corretto 24 is now generally available.
I'm still shocked that there is a Credo 24 because I still run Java 8.
Apparently this is the open Amazon's version of the OpenJDK24 future release. Apparently the next LTS version will get released in September, which would be Java SE25, which will then relate to OpenJDK25 and then Coretta25 at some point by the next year. The current long term support one is version 21, which I think is funny because everyone thinks long term support one is still Java even though Oracle does continue to support for it in 2019. But there's a bunch of companies including Amazon who offer a coretto version of Java 8 that's valid all the way through 2030.
[00:26:53] Speaker C: Oh, interesting. I didn't know that the open source tools had a longer support model than.
[00:26:59] Speaker A: The and I think it was one of the big drivers for Amazon was that they did not a they didn't want to pay Oracle gobs of money for Java 8 and number two they have a lot of apps on Java 8 that they probably need to upgrade. So got to get to it.
I'm just shocked there's. I mean I did. It was interesting. Just look at the Java versions. They did skip some numbers which helped me feel better about Just seems like Wikipedia are cool because they so 8 was good 9, then 10, 11 and then there's like a bunch of like small versions 12, 13, 14, 15. They only lasted like six months to a year and then they went away and then Java 17 was a long term support Java 21 and then I guess they're all. I guess all 25 exist but there's the lesser ones are all like one year support Windows.
[00:27:44] Speaker D: So it's kind of like the Ubuntu model which is like every two years you get a long term support version.
[00:27:49] Speaker A: Basically yeah, I like that model though. Every four looks like roughly four years. Eight to Java 11 was four and then from Java 11 to Java 17 was three.
[00:28:01] Speaker D: So whenever they feel like it got.
[00:28:03] Speaker A: It and then two to Java 21 and now another two years. It looks like they've standardized on two years now, but yeah, it was all over the place before.
[00:28:12] Speaker C: It seems even faster. I'm surprised that it's two years. It just feels like it's going by really quick.
[00:28:16] Speaker A: But maybe that's just we're aging. Yeah, I think it might be the time inflection point of COVID has messed us all up so.
Well, if you are eternally disappointed in the AWS mobile app and its limited coverage of Amazon services, the latest update might make you happier with 24 additional services now available, including service quotas, CloudFront, SES, Cloud9, AWS batch, all by integrated mobile web browser experience in the console mobile app. Now, I appreciate that they're giving you a console mobile app because it was very annoying when you were trying to use the mobile app before and you had to hit something that wasn't in there, it would basically dump you to the website, where you'd have to re authenticate, which was annoying.
And also it would be a desktop version, which you couldn't actually use on a mobile phone without lots of scrolling or turning or like, you know, squinting at it in the right way to figure out the button you need to push and hope that your large finger could poke a little button on a teeny tiny screen. So I appreciate that they are trying to give you a mobile web experience, but anyone who's on serious mobile development knows that mobile web is terrible compared to mobile native. So, like, if you're using this as a way to figure out which mobile web things are really popular to then build them into mobile native, I'm kind of okay with that as like an mvp. But I hope they're not abandoning mobile native completely for the console app, because if you're going to build an app, at least do it native, even though it costs more money.
I also don't know why they're doing Cloud 9.
[00:29:38] Speaker C: That's.
[00:29:38] Speaker A: That's a strange one, right? Are you gonna.
[00:29:41] Speaker C: You're gonna use your IDE on your phone?
[00:29:44] Speaker A: Oh, my.
[00:29:45] Speaker C: Like, I guess if you're, like, trying to, like, help someone debug a thing remotely. Like, I can kind of see it, but.
[00:29:51] Speaker D: Yeah, but they also eoled the tool. I thought they were ending.
[00:29:54] Speaker A: Yeah, that's my point. The Cloud 9 is being deprecated. I don't understand. And also the. The Amazon last. My iPad's right here. Let me check. Last time I tried to use the AWS app on my iPad, it was still not an iPad native app.
[00:30:08] Speaker D: That's because you had to have a fire phone in order for it to have Smart Jackson.
[00:30:11] Speaker A: That's right. A fire phone. That's right. How could I possibly forget the mystical fire phone?
[00:30:18] Speaker C: All the dozens of people that had.
[00:30:19] Speaker D: One that worked in the fire phone division.
[00:30:22] Speaker A: Yeah. So they all worked at Amazon.
[00:30:25] Speaker C: Let's see.
[00:30:26] Speaker A: I'm trying to log into the mobile thing on my iPad live. It makes great audio. You guys use the mobile app lately of Azure?
[00:30:35] Speaker D: I've used the console. Just, honestly, a couple times I've had to do stuff. I've gone into the. Just on normal Chrome I found and that seemed to work surprisingly well. Or at least get me the basic information I need.
[00:30:52] Speaker A: Okay, thank you for filling that space for me while I forgotten that I appreciate it. I have no idea what you said, but I'm sure it was great. Riveting.
Yeah. So actually they did make the iPad app now iPad native. So it's maybe Amazon Q on the iPad I could see maybe working out just fine. But still.
Sorry, Cloud 9 on the iPad, not Q. I don't know how I think we set up Cloud nine. I forgot how to do it. You had to set up a server or something.
[00:31:20] Speaker C: Did they ever fix the like the account authentication? Because that was what killed it for me was being able to switch from account to account to account was just nightmare.
[00:31:27] Speaker A: You can switch. Yeah, the multi session support is in there so you can switch between accounts pretty quickly, which is nice but you still have to like, you know, it still takes you to the website web browser to basically do the authentication part. But that was other nice thing they did to it, which I've actually really dug is they've created Cloud Shell inside of the mobile app and with cloud Shell I can pretty much do anything I want to. So yeah, that's cool. That's pretty nice because I've done that like.
[00:31:51] Speaker C: Like a local kind of version of browser based bash for that reason. So having cloud Shell would replace that entirely.
[00:31:58] Speaker A: I will say I've actually used it because I. My buddy called with his issue with the site and he need to spin up some servers for him because we were dealing with a bot issue and AI, nasty AI bots.
So I literally called me and I was in the middle of nowhere and I was like hold on a second, I'll do it on my phone. And I hopped on and spun up five more servers and was like, you should be good to go and hung up. And it was good until I could get home and actually look at the problem later. But yeah, it was. It does come in handy. I do admit that I use it occasionally, although I hate it every time. It's like a hate usage. Like I'm going to use the mobile app.
[00:32:29] Speaker C: Yeah, the value is there, but the delivery is lacking.
[00:32:34] Speaker A: Yeah.
[00:32:37] 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? Achero will click the link in the Show Notes to check them out on the AWS Marketplace.
[00:33:17] Speaker A: All right. AWS is giving you new flow management features for the AWS Network firewall. This enables customers to identify and control active network flows. This feature introduces two key functions. Flow capture, which allows point in time snapshots of active flows, and flow flush, which enables selective termination of specific connections in your flows.
[00:33:36] Speaker C: So flow capture is just, you know, the networking team is sick of providing packet captures, I imagine. So like now it's self service. It makes perfect sense.
[00:33:47] Speaker A: Yeah.
[00:33:48] Speaker C: Flush is interesting though that one I don't quite get. Like when we're talking about a firewall specifically. Right. Like if we talking about a route.
[00:33:57] Speaker A: It'S basically you terminating the session. Like, oh, I don't like the session I want to terminate it is why I assumed you would need that for.
[00:34:03] Speaker C: Yeah, yeah, I guess it is a specific connection. Yeah, no, you're right. That's probably exactly what it is.
[00:34:10] Speaker A: Yeah. If you do the flow flush or the flow capture, it basically gives you.
With these new cables, the customer can now view and manage active flows based on criteria such as source destination, IP ports and protocols. So, yeah, so if you have someone connecting on a protocol that you don't want them to be, and you identified them through your threat intelligence, you want to kill that. And so you can do that without having to turn off the rule or whatever to sever that connection. There's some handiness to that. I. I don't hate it.
[00:34:36] Speaker C: Yeah, it'd be kind of cool. You know, just string that together with your security tool. Right. You can automatically block. Yep.
[00:34:43] Speaker A: All right, well, by the time this episode is published, it'll. It'll be time for Amazon or sorry, for Google Next.
And that means that you should be packing your bags, getting ready to Google Next and come to two critical sessions, one on Wednesday called BRK2024 Workload Optimized Data Protection for mission critical enterprise apps. And you're coming there to see me talk briefly and then to come see me afterwards get stickers. That's your purpose, that's why you're there. It's also a great talk. There's a lot of good stuff they're going to cover. I've seen the presentation, they're gonna do a great job, but I don't really care about them. But they're gonna do a great job. I'll talk about my day job for a brief couple minutes and how we've journeyed through our process of running Metric Google Apps on Google Cloud. But I will have stickers. And then if you miss Wednesday because you're busy on a session or chalk talk or whatever you're doing, you have a second chance to see me at BRK 1 028, which is unlocked value for your workloads at Microsoft, Oracle, OpenShift and more, where I'll be doing a similar talk to the day before, but different. And I also still have stickers. So come see me at one of those two sessions. It's one place I can guarantee you will find me to get stickers. You'll see Ryan there most likely too. You can join him in the audience. Cat calling to me on stage messed me up. So we look forward to seeing all of our listeners at Google Next. Those of you who are Google users who decided to go to the conference, it should be a great time at Mandalay Bay. Looking forward to it. I can't believe it's only two weeks away, but it's right around the corner, which is great. Next week, guys, we are recording predictions, so just put you guys on notice.
[00:36:21] Speaker D: Oh, you would think that would be enough news. It's not going to be.
[00:36:30] Speaker A: And I unfortunately can't talk about any of the things that are in my two sessions because I already know about them and I can't talk about them.
[00:36:34] Speaker C: So.
[00:36:36] Speaker D: We. No, we're still screwed.
[00:36:38] Speaker C: That's good. I like the handicap.
[00:36:40] Speaker A: Yeah, yeah, I definitely need the handicap.
All right, well, moving on to other Google news, they're introducing Protection Summary, a new Google cloud backup and Dr. Creature. They're really into the protection thing this this month. AI protection now. Data protection. Data protection, of course, is critical to your cloud strategy and that includes backups and disaster recovery. And making sure your backups are set up correctly and align with your RPO and RTO requirements is critical to avoid things like ransomware. However, collecting the data in your complex cloud environment can be tricky and so Google is giving you a preview of the Protection Summary and the Data Protection tab, a new feature in Google cloud backup and Dr. That provides a centralized view of your backup configurations, helping you identify gaps in your data protection and empowers you to take action to improve your resiliency. Protection Summary will quickly help you identify resources with no backup configuration, and I put that in an asterisk. No backup configuration in Google Cloud Backup ndr.
But you might have other backups, those just don't count Here. Quickly configure backups for resources and then assess the backup configurations and vulnerabilities and eliminate your vulnerability to ransomware through these new capabilities. There's a couple good screenshots here in the article to show you how this looks in your system and it's pretty quick, easy UI from Google as usual, telling you when the last backup was, which this is also great for all your compliance people who want to make sure your backups are working. You can now prove it to them using this technology as well as show the vaulting configuration, et cetera, for your auditors.
[00:38:04] Speaker C: That was the first thing I was thinking about when I read through this was the terribleness that I did like you know, 12 years ago to plug in some sort of backup errors to a Slack channel so that we could pass an audit for notifications. It was ridiculous.
[00:38:23] Speaker D: Just emails solve that problem?
[00:38:25] Speaker C: Well, no, I mean it was just, it was, it was more about the signal itself. They're like, well how do, how would we know if it failed? You know, it's just like why do we need the evidence?
[00:38:35] Speaker A: Well, if you can't prove it failed, you can't prove it succeeded and so then you're screwed.
[00:38:40] Speaker C: Well if you can log a success. So this was in addition to the already captured success of a backup. So the idea of that if there wasn't a success in a time period that could be counted as a failure, that wasn't good enough. It had to actually notify a failure anyway.
[00:38:59] Speaker A: Google recently announced the public beta of AI Toolbox for Databases and today they're excited to expand its capabilities through a new partnership with Hyper Mode Generator. Toolbox for Databases is an open source server that empowers application owners to connect production grade agent based generative AI applications databases. Toolbox streams the creation, deployment and management associated tools capable of querying databases, secure access, robust observability, scalability and comprehensive management. And currently the toolbox supports Alloy DB spanner, Cloud SQL for Postgres, MySQL and SQL Server as well as self managed MySQL and Postgres boxes. I was really unclear about what Hyper Mode is providing to the story here, but you know, it basically it supports dgraph in partnership with Hyper Mode. So if you know what DGRAPH is, this is for you. I don't know what dgraph is and so this is where I lost this article but I did want to mention it because I had not heard of this toolbox before but just sounds like a really Fancy caching layer for databases to make AI, not crash database.
[00:39:52] Speaker C: Oh, I was thinking it was more along the lines of changing the query syntax so that, you know, you don't have to like, you can more like natural language processing of your prompt rather than a direct SQL query. But the funny thing about that is the first thing I thought of was like, well, people that know that SQL prompts write these terrible queries that take down services. I don't know if this is a good idea.
[00:40:15] Speaker A: I don't know if I addressed this either.
[00:40:20] Speaker C: But I don't know. I mean it's definitely something to play around with, right? Because it's. I'm never, I'm never going to learn SQL. Too old at this point. I don't want to. And now like I used to just Google it, but now I just have AI do it.
[00:40:32] Speaker A: I mean, we've secretly been teaching you SQL all this time by making you learn Athena BigQuery and like, so you technically know it better than you probably realize.
[00:40:41] Speaker C: I know it better than I'm willing to admit.
[00:40:43] Speaker A: I don't know, I realize, I think what you've. You fail to learn is how to use store procedures and that's actually a good thing. So you learned just enough databasing to be technology sound.
[00:40:52] Speaker D: So my, my day job, my team knows that if I have to log into SQL, we're having a really bad day because I should never ever be allowed to do that because just bad news bears across the board.
[00:41:04] Speaker A: Yeah, I mean, pretty much if I had to log into the Google cloud or get onto the vpn, I'm going to be cranky. So I don't do it very often and for good reason. Because they don't want me to do it because I will break stuff.
Announcing BigQuery repositories Git based collaboration in BigQuery studio Modern data teams use Git to collaborate effectively and adopt software engineering best practices for managing their data pipeline and analytical code. But most tools don't offer integration with Git version control systems, making Git workflow feel a bit out of reach or forced. This forces users to copy and paste code between UIs, which is not only time consuming but also error prone. To help, Google is releasing in Preview BigQuery repositories, a new experience in BigQuery Studio that helps data teams collaborate on code store and Git repositories. BigQuery Repos provides a comprehensive set of features to integrate Git workflows directly into your BigQuery environment. You can set up new repos in BigQuery Studio where you can develop SQL queries, notebooks, data preparation data canvases or text files with any file extension. Connect your repositories Remote git hosts like GitHub, GitLab and other popular git platforms edit the code in your repositories of the dedicated workspace or on your own. Copy the code or before publishing changes to different branches and you can perform most get operations with our user. Friendly interface lets you inspect differences, commit changes, push updates and create pull requests all within BigQuery Studio. You notice it didn't say anything about putting friendly commit messages, but I hope you do that when you make your git changes. Be a good git user I'm not.
[00:42:26] Speaker D: A good git user.
[00:42:28] Speaker A: Oh, I'm a hilarious git user user. Read my git commits.
[00:42:32] Speaker D: Mine are normally me just getting mad at the stems or it's the same message because I just do up enter my command line. So it's just more fixes or like fixes or like untested code. Like that's most of my commits.
[00:42:48] Speaker C: AI does that for me now too. There's just a button that just automatically.
[00:42:52] Speaker D: Ooh, I haven't thought about that.
[00:42:53] Speaker C: It generates the commit. I'm getting so lazy.
[00:42:56] Speaker D: But how do you do that commit? Wait, are you using a UI for your commits? Ryan?
[00:43:01] Speaker C: It's in natively in my ide.
[00:43:02] Speaker A: Yeah, in Visual Studio it's all GitHub.
[00:43:06] Speaker D: I'm still judging you for using the ui. Like I.
[00:43:09] Speaker C: It's not a ui. It's not a ui. It is command line. It's just command line. It's in my. It's my code editor and all I got to do is a command a key sequence to if I want to create a branch or if I want to push the branch to remote. It's just doing all that stuff. Native.
[00:43:24] Speaker A: Yeah, that's all nothing.
[00:43:25] Speaker C: It's not GitHub desktop or anything like that.
[00:43:27] Speaker A: So I, I do use the GitHub CLI for the pure purpose of logging into GitHub.
[00:43:32] Speaker D: So because I judge you a little.
[00:43:34] Speaker A: Bit, just say because the whole like going. Because it used to be easy get a key and you download the key and now you have to use HTTPs and you do. So like you just use the GH login and you just log in.
[00:43:45] Speaker D: Yeah, but they hide it now. Yeah, I know this because I went to go do something today where I wanted to download a repository and I was like wait, I can only see GitHub and HTTP. I was like, where's the SSH option. And then I was like, wait, I don't want to have to do the like start manually typing file. I'll just copy and paste this. And I did it over HTTP handing myself a little bit. But you know, it's not. They don't show you like the copy and paste of
[email protected] and now they.
[00:44:14] Speaker A: Were deprecating SSH support.
[00:44:16] Speaker C: Oh, I don't. Oh God, I hope not. That would suck.
[00:44:18] Speaker D: Oh, there's so many things about I got to break.
[00:44:20] Speaker C: Yeah, I mean, it's funny because I was thinking recently how nice the UI experience was from GitHub because I went to my personal repo and I noticed that it just defaults to SSH for on my ui. So I never. It's. It's not hidden for me. But I wonder if they are about to make changes.
[00:44:37] Speaker D: It is being deprecated. Oh, wait, sorry, that's a circle CI.
No, I don't see a few. Both that.
I think they're just making it be less in your face.
[00:44:51] Speaker A: I think, I think it's the SSH off of the key they got rid of. I think. I think once you auth, you can still do it, but you have to do it differently in the auth anyways, just easier. I didn't do anything. You hit the type of thing and it gives you a web URL. You go there, log in, you give it a little code and it logs it in and puts a token. I don't have to worry about anymore.
[00:45:09] Speaker D: That's one more piece of software I have to install my laptop.
[00:45:11] Speaker A: It's a CLI thing. I use brew install GitHub or whatever it is.
[00:45:16] Speaker D: Brew Install Brew bundle for me, I have my brew file.
[00:45:19] Speaker A: Yeah, yeah. Like, oh my God, that's. If I install Brew, I don't even care. I'm like, okay, whatever. Yeah. If I download a package and extract it, then yeah, sure, okay.
[00:45:28] Speaker C: Yeah, then I'd be complaining.
[00:45:30] Speaker A: Yep. All righty.
Well, Google is introducing Gemini 2.5, their most intelligent AI model ever, which they say every time they release a new Gemini model.
[00:45:40] Speaker C: I was going to say I'm glad it's getting smarter and not dumber.
[00:45:42] Speaker A: Yeah. The first 2.5 release is an experimental version of 2.5 Pro, which is the state of the art on a wide range of benchmarks and debuts at the number one on Elle Marina by a significant margin, which I love that. The benchmarking site is called Elamarina. Gemini 2.5 models are thinking models capable of reasoning through their thoughts before responding, resulting in enhanced performance and improved accuracy. The Gemini 2.5 Google has achieved a new level of performance by combining a significantly enhanced base model with improved post training. Going forward, they will build thinking capabilities directly into all models so they can handle more complex problems to support even more capable context aware agents. Google is proud that 205 Pro takes the top L on the rail leaderboard today. Gemini 2.5 without test time techniques like majority voting leads math and science by 2.5 times in the benchmarks. It also scores a state of the art 18.8% across models without tools used on humanity's latest exam. A data set designed by hundreds of SMEs capture the human frontier of knowledge and reasoning. 2.5 will have a big leap over 2.0 on coding performance as well as excels at creating visually compelling web apps and agentic code applications along with code transformation and editing.
So I don't know when this gets into Gemini code but I'm curious.
[00:46:52] Speaker C: Well I mean 2.0 was a big big fix over 1.5 so I'm hoping that it's as big of an impact.
But I am laughing because this is the second article just in this show with the sort of the measurement of large language performance. So there was Gaia in the Manus article and now there's Element Arena.
[00:47:15] Speaker A: Yeah I see most of the US companies all go to Ellen Marina as.
[00:47:19] Speaker C: Kind of their is it standardized? Because that's what I was going to.
[00:47:23] Speaker A: Laugh next was that I mean is it standardized? I don't know but definitely they've all sort of started mentioning Elle Marina as kind of their their go to. Okay but yeah the there are a couple others as well that are mentioned from time to time but Ellen arena I think ChatGPT use as well when they announced their new model a couple weeks ago so it's not the first time I've seen it.
[00:47:45] Speaker C: I mean it sounds like a really good business to get into. Right?
[00:47:49] Speaker A: Right.
[00:47:50] Speaker C: Yeah. So right now measure performance.
[00:47:52] Speaker A: So right now it's ranked number one Gemini 2.5 Pro XP03 to 25 followed by Grok 3 preview and then GPT 4.5 and then Gemini 2.0 flash thinking deep seek is down number seven.
[00:48:07] Speaker C: Is it largely do you know can you tell like with the quick glance Is it largely just response time?
[00:48:12] Speaker A: I I mean you're asking questions.
[00:48:14] Speaker C: I know John Ben come back right.
[00:48:19] Speaker A: But they have you know they have overall with style control, hard prompts, hard problems, Style Control coding, Math creative so like you click the different things, you can sort the ranks. They've got Price analysis, they got Web Dev Arena Vision, Text Image Copilot arena, and Arena Hard Auto. So lots of stuff there. I don't know what even means but people were really into this. I'm sure geek out on these things. I just not that person as much as I'd like to be.
All right, Azure. All right Matt, it's time for you to shine because they're announcing the public preview launch of Azure Functions, the Durable Task scheduler.
This new Azure managed backend is designed to provide high performance, improve reliability, reduce operational overhead, and simplify monitoring of your stateful orchestration. Durable Functions provides you a simplified way to develop complex staple and long running apps in the serverless environment. This allows developers to orchestrate multiple function calls without having to handle fault tolerance. It's great for scenarios like orchestrating multiple agents, distributed transactions, big data processing, batch processing like ETLS, async APIs, and essentially any scenario that requires chaining functions calls with state persistence. So tell me why this sounds like step functions but isn't step functions? Because Azure bastardized it go.
[00:49:29] Speaker D: It's step functions with a CloudWatch event that triggers it.
[00:49:34] Speaker A: The two use case for this.
[00:49:37] Speaker C: I'm pretty sure you could do that. I just don't know if you could do it directly. But I know you could do a CloudWatch event to trigger a lambda which could trigger step functions out there. Yeah.
[00:49:46] Speaker A: Can you just use EventBridge CloudWatch to EventBridge to Lambda?
[00:49:48] Speaker C: Yeah, I assume so. Yeah.
[00:49:50] Speaker D: Yeah, you could do that too.
[00:49:52] Speaker C: That's interesting. I mean I like these types of tools just because I find it fun to write programs in this way. Like instead of having these giant processes that sort of chug through, you can distribute into many serverless apps and have it all chained together and track state and then report back at the end. So because I like this model, I'm a big fan of these tools.
[00:50:14] Speaker D: Yeah, I mean it's a. It's function, it's step functions, you know, so it's gonna do everything Step Functions can do. I haven't played with it fully yet, but I know we're starting to move some stuff over to Durable Functions in our my day job just because it also gives you a little bit extra time with a function app and there's a few like overhead pieces that you get a little bit more optimized with.
[00:50:38] Speaker A: I would say though, you know, just looking at the screenshots of this, I. The thing I immediately miss, which is kind of sad, is the Visual workflow aspect of it. I just see lists of functions like oh, all these tasks and organization IDs and I'm like, yeah, it would be great to see for that in a workflow that I could actually understand.
[00:50:55] Speaker C: It is kind of funny.
[00:50:58] Speaker A: I've made fun of sub functions workflow view before, but when you don't have it, all of a sudden you're like, ooh, I kind of miss it.
[00:51:05] Speaker C: Well, the UI kind of looks like my first website, so I totally get that.
[00:51:08] Speaker A: Yeah, it looks terrible. Like it's not a great UI on amaz, but I definitely have appreciated the visualization many times.
[00:51:17] Speaker C: Yeah, I miss it working with GNA directly now for most of these tasks and yeah, no, it's. This is the visual side. Or being able to communicate or just talk about, you know, the flow and where things are isn't very easy to visualize, you know, using Kubernetes, tooling and cli. So it was kind of funny.
[00:51:40] Speaker A: I mean, GCP has workflows, right? But that doesn't touch knative. It only touches.
[00:51:44] Speaker C: Well, it's probably on the back end. I haven't played around with it.
[00:51:46] Speaker A: Right. Okay. I'm just curious.
[00:51:49] Speaker C: Yeah, no, yeah, I haven't used managed services and so. Because the right now the limiting thing is the environment.
[00:51:57] Speaker A: Yeah, yeah, because I think basically their version of step functions is workflows, which gives you kind of the same visual thing.
But I haven't used it either. So there you go.
So you're going to use this, huh? Let us know how it goes, Matt.
[00:52:11] Speaker D: Yeah, I'll show you guys soon.
[00:52:13] Speaker A: Well, for things you're not going to use that you won't tell us about so many Azure container app serverless GPUs are now generally available. This allows you to seamlessly run your AI workloads on demand with automatic scaling, optimized cold start per second billing and reduced operational overhead. Nvidia powers the serverless GPUs which allows you to seamlessly run billing. Sorry, run billing with scale down to zero when non Use this reducing operational overhead to support easy real time custom model inference and other GPU accelerated workloads. In addition, this supports Nvidia Nim, which I'll talk about more in a second. Which are part of Nvidia AI Enterprise. It's a set of easy to use microservices designed for secure, reliable deployments of high performance AI models. Inference at scale. Key Benefits for serverless GPUs scale to zero GPUs so only pay for what you need of the Nvidia A100 and T4 GPUs which are pretty darn pricey, only pay to the second. So you only use exactly as much as you need. There's a built in data governance using your data never leaves the container boundary. There's flexible computing options between the Nvidia A100 and the T4 GPUs. Again, if you have those two choices, that's all you get. That's flexible per Azure and a middle layer for AI development. Bringing your own model on a managed serverless compute platform and easily run your AI applications alongside your existing apps.
[00:53:25] Speaker C: I mean, I want to make fun of this, but I love the fact that it scales down to zero. No, I like it like if I was going to make an application like I would be bankrupt, you know, except for being able to put something like this in place. So it's like, I think this is kind of neat. Yeah, it's not full featured, but you're also trying to, you know, get a temporary sort of AI processing unit. Right? So it's sort of like, yeah, it's probably really good for most use cases, those larger use cases where you're going to need more, you're probably not going to run that in containers at all.
[00:53:59] Speaker A: Well, Nvidia Nim, let's jump into that. So Microsoft Nvidia had an. Nvidia had an event and Microsoft was there and so they announced several enhancements to help shape the future of AI. This includes integrations of the newest Blackwell platform on Azure AI incorporating Nvidia Nim microservices into Azure AI Foundry and empowering developers to accelerate their innovations and solve challenging problems. Nim provides optimized containers for more than two dozen popular foundation models, allowing developers to deploy generative AI applications and agents quickly. These new integrations can accelerate inference workloads for models available on Azure, providing significant performance improvements, greatly supporting the growing number of AI agents. Key features include optimized model throughput for Nvidia, accelerated computing platforms, pre built microservices, deployable anywhere and enhanced accuracy for specific use cases. Now I had to take the next level of deep dive on what NIM was because again, didn't really tell me. But all those tools like Pytorch and all those that are on the Image from basically AWS's container, those are all those. Let's put some Nvidia support. Nim things basically gives you all of the Pytorch type tools, all of the different capabilities you might want to use to use your GPU'S effectively to do training or inference, et cetera, all pre built into the NIM containers that are pre built for you. So that's what it is. But they made it sound like it was special, but it's not.
[00:55:14] Speaker C: Yeah, I mean there's been kind of curated Pytorch images forever and other ones with like a whole set of tools on there. So it is kind of funny like thank you.
[00:55:26] Speaker A: You're just giving us a container where it has a bunch of stuff built into it. Yeah, there's tensorrt, there's tensorrtlm built into this. You get a bunch of cloud native things, you get CUDA support, you get Dall E, NCCL all built pre built into the NIM architecture. So I mean it's helpful.
As I mentioned, you also get the general availability of the GB200 V6 virtual machines accelerated by the Nvidia GB200, NVL72 and Nvidia Quantum Infiniband networking.
And once you get that Nvidia NIM deployed On those fancy GV2 hundreds, you can run the Nvidia Agent IQ to take center stage with its open source toolkit designed to seamlessly connect, profile and optimize teams of AI agents, enabling your system to run at peak performance. And GenTech IQ delivers profiling and optimization, dynamic inference enhancements and integration with semantic kernels. So this is where I think Nvidia kind of loses me in their strategy because they're really good hardware manufacturers, they make great hardware. Now I've owned a lot of Nvidia GPUs in my computers over there, my gaming machines and I can tell you that how often they release new drivers for those dumb things, I'm not sure about their quality or capabilities that I would trust them to build any of this other higher value stuff like their Nvidia Agent iq. I mean I appreciate the effort and I appreciate that there's another model out there, but is this really where Nvidia is going to shine or are we going to expect them to make really badass GPUs? And I think the answer is badass GPUs with suspect drivers. I think that's the right balance for Nvidia and everything else is a bit of extra fluff. Will they be a dominant player? Not like CUDA is? Maybe we'll have to see.
[00:56:59] Speaker C: I will tell you the majority or at least statistically significant amount of my family IT work is around those immediate drivers and all my kids computers.
[00:57:11] Speaker A: Yep, every time they need an update, you're in the Hell again.
All right. Last year, Microsoft launched Security Copilot to empower defenders to detect, investigate and respond to security incidents swiftly and accurately. Now they're announcing that we're going to fire all those people with Security Copilot With AI agents designed to autonomously assist with critical areas such as phishing, data security and identity management, the relentless pace and complexity of cyber attacks have surpassed human capacity, and establishing AI agents is necessary for modern security practices. Microsoft Threat intelligence now processes 84 trillion signals per day, revealing the exponential growth in cyber attacks. Today, they're launching six Security Copilot agents built by Microsoft and five built by their partners. Available to you in preview in April, the five agents from Microsoft include amazing the Phishing Triage Agent, which Triage's phishing alerts accurately identify real cyber threats and false alarms, providing easy to understand explanations for its decisions and improve detection based on admin feedback. The Alert Triage Agent and Microsoft Purview to triage data loss prevention and insider risk alerts, prioritize critical incidents and continuously improve accuracy based on admin feedback. The Conditional Access Optimization Agent, which in Microsoft Entra monitors for new users or apps not covered by existing policies, identifies necessary updates to close security gaps, and recommends quick fixes for identity teams to apply with a single click and cause an outage. Vulnerability Remediation Agent in Microsoft Intune monitors and prioritizes vulnerable vulnerabilities and remediation tasks, address app and policy configuration issues, XYZ Windows OS patches with admin approval and Threat Intelligence Briefing Agent and Security Copilot which automatically curates relevant and timely threat intelligence based on the organization's unique attributes and cyber threat exposure. Those are ones from Microsoft. Any thoughts?
[00:58:50] Speaker C: So as the new security guy who's learning all these tools and going through and like going through all the things that are in Microsoft Defender like, I am very skeptical that this is going to actually solve any issues, but sweet Jesus, if it's an improvement on what Microsoft Defender or he does, be welcome because it's the patterns and stuff that are detected natively in those tools. Just by default is not good enough and so you have to spend a ton of time trolling through too much data to make these things work for anything other than forensic investigation after the fact.
So I hope so and I can't wait to play with this and I'm going to go see if I can get this in my day job.
[00:59:37] Speaker A: Well, the partner ones include Privacy Breach response agent by OneTrust, which analyzes data breaches, generate guidance for the Privacy team on how to meet regulatory requirements. So tell you what you need to disclose. The Network Supervisor Agent by Aviatrix, which performs root cause analysis and summarizes issues related to vpn, gateway or site to cloud connection outages and failures. The SecOps tooling agent by BlueVoyant, which assesses a security operations center, SOC and CF controls to make recommendations that help optimize security operations and improve controls compliance. The Alert Triage Agent by Tanium, which provides analysts with the necessary context to quickly and confidently make decisions on each alert. And the Task Optimizer agent by Fletch, which helps organizations forecast and prioritize the most critical cyber threat alerts, reduce alert fatigue and improve security. I like that last one. Like, okay, if you can predict the prioritization of the most critical cyber threat alerts, then why do I need to do anything? I'll just prioritize the most critical ones.
[01:00:32] Speaker C: That's kind of nuts. I mean looking at the sponsorship names on these things, I'm like each one of these is going to be just basically a firstborn child in price. I'm sure of it. Because security tooling is still just gouging.
[01:00:48] Speaker D: You'll have an option on the pricing is the problem. Like you have to pay, you have to have these tools in place, otherwise you can't pass your sock, you can't pass all your other certifications.
[01:00:58] Speaker A: Well, and if you, if you pass the first, you know, if it, let's say it cost you 300,000 but it avoided you losing data for a customer, you know, it could pay for itself. So like these things get spent a lot of money on for sure.
[01:01:10] Speaker C: Yeah. But in the sea of things like I'm, I'm, you know, my eyes are open to a new, a whole new world. Right. And so it is like it's a sea of, a sea of things of choices and so you, you can't buy them all. You're trying to find the ones that extract the most value and they all sort of promise this pangea of like you won't have to do anything. It replaces all your sock engineers and it really doesn't.
[01:01:31] Speaker D: No. But hopefully it like trims down some of the noise that you get because otherwise, you know, it's way too much noise signal to noise ratio that people have.
[01:01:43] Speaker C: Yeah, no, it definitely is. And so I look for any, anything to reduce that. And that was the first thing I noticed in the new job and getting on the distribution list was the amount of noise that comes out of these tools. Like I don't know how you're supposed to find anything.
[01:01:57] Speaker A: I mean, that's why they don't buy anything typically until it's too late.
All right, and final story for tonight is Oracle. They're introducing AI Agent Studio. Oracle has announced the a Oracle AI Agent Studio for Fusion Applications, a comprehensive platform for creating, extending, deploying and managing AI agents agent teams across your enterprise. This is part of the Oracle Fusion Cloud application suite. The new AI Agent Studio provides easy to use tools for customers and partners to create customized AI agents that address complex business needs and can help drive new levels of productivity. Oracle Agent Studio includes agent template libraries, agent team orchestration, agent accessibility, choice of LLMs, native fusion integration, third party system integration, trust and security frameworks, and validation and testing capabilities all out of the box.
[01:02:41] Speaker D: Oracle showed up to the AI Agent party.
[01:02:45] Speaker A: Yeah, I mean, everyone's getting into the agent party as they should.
[01:02:49] Speaker C: It's always interesting to see what managed services Oracle develops. Just because you think about the customer base that's on Oracle Cloud and what are they demanding the loudest. And so makes sense. This makes sense.
[01:03:02] Speaker A: Yeah. I don't have anything else to say about their agents. Glad to see them catch up as usual. I mean, I'm sure Larry Ellison will at Oracle whatever it is now, Open world, cloud, world, whatever, it's the best, better AI agent thing in the world and no one else can touch them with their amazingness. And they'll make unbreakable agents. I'm sure.
[01:03:29] Speaker C: That'S pretty good. I like that. I hope they figure that out.
[01:03:32] Speaker A: Yeah.
[01:03:32] Speaker C: Is that like a no hallucination agent?
[01:03:34] Speaker A: Yeah. Unbreakable. It's unbreakable. Doesn't say it doesn't hallucinate, but it doesn't break anything.
All right, gentlemen, that is it for another fantastic week here in the Cloud. Again next week we will do predictions for Google Next, see how our crystal balls are looking, and hopefully we'll win something, you know, bragging rights over each other.
[01:03:56] Speaker C: Yeah.
[01:03:57] Speaker A: Yep, yep. And again, remember, if you are checking out your Google Next schedule to book those two sessions that I am at, if you want to get your Cloud pod sticker, I will arm Ryan with them so he can deal with you people first. Just kidding. We will be sure answering questions about our talk, my talk, and then you. If I'm busy talking about my talk, Ryan can give you your stickers and if not, just make sure you say hello. Love to meet all of our active listeners at Google Cloud Next.
See you next week.
[01:04:23] Speaker D: Bye, everyone.
[01:04:24] Speaker C: Bye, everybody.
[01:04:28] 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.
