[00:00:07] Speaker A: 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 B: We are your hosts, Justin, Jonathan, Ryan and Matthew.
[00:00:18] Speaker C: Episode 313 recorded for July 15, 2025.
[00:00:23] Speaker D: The Gartner Guide to Breaking Things on Purpose.
[00:00:26] Speaker B: I like it.
[00:00:27] Speaker E: Its own category now. Is it like, you know, we have different categories for each quadrant, like purposely breaking it? Sort of breaking it. Definitely didn't break it.
[00:00:36] Speaker B: And I meant to do that.
[00:00:36] Speaker E: Specialty breaking.
[00:00:37] Speaker D: I want the.
[00:00:37] Speaker B: I meant.
[00:00:38] Speaker E: I meant to do that.
[00:00:39] Speaker D: It should be one that deleted everything. That should be one of the categories.
[00:00:41] Speaker E: Rmisfr.
[00:00:42] Speaker D: Yeah, yeah, yeah.
Well, it's another fantastic week here in Cloud World.
[00:00:49] Speaker C: You know, we've got a bunch of.
[00:00:50] Speaker D: News for everyone but Azure. So for those of you who don't care about Azure, you're welcome. And for those of you who do. I'm sorry, I can't.
[00:00:58] Speaker E: I should argue with you, but I can't.
[00:01:01] Speaker D: You should probably. But summit's tomorrow. Technically, I was thinking it was today and I was like, man, it's gonna add a lot to the show. But then I saw the keynotes on Wednesday and I was like, missed the recording deadline.
[00:01:11] Speaker E: You're just excited that you're not gonna be here next week.
[00:01:14] Speaker B: That's true.
[00:01:15] Speaker D: I'm gonna have to listen to you to work through the whole list of whatever Swami announces on stage just tomorrow. So that'll be interesting to see.
[00:01:22] Speaker E: I feel like summits are hit or miss too.
[00:01:24] Speaker D: Yeah, yeah, they are, but. And if it's Swami who's doing the keynote, which it is, it's going to be heavy. Mlai. So you two are going to love next week. It's going to be great.
[00:01:33] Speaker E: We need to find somebody to help.
[00:01:35] Speaker B: Us next week or it might be one of those weeks where like the.
[00:01:40] Speaker D: Yeah, unfortunately, Covid has hit and no one has a voice.
[00:01:43] Speaker E: I do know a ton of people that have had Covid now in the past couple of weeks, so there's a.
[00:01:48] Speaker D: New shame going around. It's definitely. I mean, it's. Covid is the one weird disease that seems to be more popular in the.
[00:01:53] Speaker C: Summer than it is in the winter.
[00:01:55] Speaker D: So, you know, versus the rest of them all wait to winter. So you can get terrorized now year round. It's a follow.
[00:02:00] Speaker C: Follow the seasons.
[00:02:02] Speaker D: Diseases. Yeah, that could be a good show title.
All right, following up, we talked briefly about this in one of the discussions about Windsurf and that they were being bought by OpenAI and we were like, oh, that makes sense. You know, Cursor and all other competitors are going to buy this.
And then last week, it all fell apart.
Apparently OpenAI announced that they were not going to now proceed with purchasing Windsurf. They didn't really explain why, but all of the talent, well, not all the talent. The CEO basically and key leadership team are all hired away by Google in a $2.4 billion licensing and compensation package deal to secure the leadership team.
And basically then another company came in.
[00:02:48] Speaker C: And took the husk of what was.
[00:02:49] Speaker D: Left of Windsurf and is now buying it as well. And that's Cognition AI. And so they'll be acquiring Windsor's IP product, remaining talent after with whatever Google.
[00:02:59] Speaker C: Left behind, highlighting the intense competition for AI coding space among major tech companies.
[00:03:05] Speaker D: And both companies are of course, developing AI coding agents designed to accelerate software.
[00:03:08] Speaker C: Development with Cognitions, Devin Agent and Winster's tools, representing the growing market for AI powered developer productivity solutions.
[00:03:14] Speaker D: And the acquisition will ensure that all Windserf employees receive accelerated vesting and financial participation, which means pennies addressing the disruption caused by the leadership of Exodus to Google.
The consolidation of the AI coding space.
[00:03:25] Speaker C: Suggests smaller startups may struggle to retain talent and remain independent as tech giants aggressively pursue AI engineering capabilities.
[00:03:31] Speaker D: I mean, so you're the CEO of Windsurf. You basically were gonna get a big payday. OpenAI said, Nah, never mind. And Google came with a big paycheck. And he said, yeah, screw my company, I'm gonna go that way. Wow, that's fantastic leadership right there.
[00:03:45] Speaker B: I'm not sure that's the order of operations.
Like, we don't know if the, if Google came in and that's what tanked the deal.
[00:03:51] Speaker D: But, I mean, that's possible too.
[00:03:53] Speaker B: But I have no idea.
[00:03:57] Speaker D: I have no idea.
[00:03:58] Speaker B: It is crazy. And it does just show how competitive this space is right now. And it's, you know, it's not any real different than the tech industry has.
[00:04:06] Speaker E: Been a while, but we've just moved to a new area of ridiculousness.
[00:04:11] Speaker B: Yeah. And it's, you know, it's a lot harder for startups to compete in the AI space because of the, the hardware and infrastructure components. Right. So it's not as simple to have your little like, you know, three buddies on a couch create that competitive company.
[00:04:26] Speaker E: Or it is, because you use AI and just vibe code it all to help.
Depends which way you want to do it.
[00:04:34] Speaker D: All right, let's pour some out for Windsurf and move to other AI news.
[00:04:41] Speaker C: Grok 4 was announced last week.
[00:04:43] Speaker D: It's Xai's latest model and has a little interesting wrinkle if you watch the reasoning model as you ask it a question, potentially a controversial question like who's.
[00:04:57] Speaker C: To blame for the Palestinian, Israeli or Gaza or Hamas issues.
One of the reasoning steps it goes.
[00:05:02] Speaker D: To is check Elon Musk's tweets before formulating a response.
The behavior apparently is not consistent though, and so hard to repro. I wasn't able to repro it. While some seek rock searches for Musk's views, other reports the model searching for.
[00:05:17] Speaker C: Its own previous stances or providing different answers entirely.
And this discovery highlights potential challenges in.
[00:05:22] Speaker D: AI alignment and bias in Cloud hosted.
[00:05:24] Speaker C: LLMs where models may inadvertently incorporate owner preferences into their decision making processes with explicit programming.
[00:05:29] Speaker D: Which is the whole reason why I don't use Grok to begin with, because I just assumed it was not going to be using anything that I felt was viable unless it was about rocket science or electric cars. And then I felt pretty confident that it might give me good answers.
[00:05:40] Speaker C: But anything else beyond that I was.
[00:05:42] Speaker D: Just not super jazzed about Super Grok.
[00:05:45] Speaker C: Tier costs 22.50amonth and includes visible reasoning.
[00:05:48] Speaker D: Traces similar to OpenAI's O3 model and in the benchmarks, Grok 4 being latest, greatest and built after O3 and Claude 4 of course stole the crown as the best until next week when some other AI provider will launch their next model and take the crown right back.
[00:06:05] Speaker C: So we'll see what happens.
[00:06:08] Speaker B: Yeah, I mean I wasn't a Grok user before and this is part of it, you know, is why I'm not. And then this is just sort of. Yeah.
[00:06:18] Speaker E: Icing on the cake.
[00:06:20] Speaker B: Yeah. And just you know, it's obviously all my concerns about you know, the bro coders and the culture and must sort of cult of personality dictating these things and not being sort of a, you know, know something that's can be trusted like yeah, here we go. It's built right into the product and.
[00:06:42] Speaker E: It'S a feature of it depending on how you view it. Yeah, it's not a bug. Right. It's a feature.
[00:06:50] Speaker C: DigitalOcean is launching their gradient AI, a.
[00:06:53] Speaker D: Unified AI cloud platform that combines GPU.
[00:06:55] Speaker C: Infrastructure, agent development tools and pre built AI applications into a single integrated experience for full AI development lifecycle. The platform consists of three main components.
[00:07:04] Speaker D: Infrastructure, Platform and Applications. The new GPU options are being added.
[00:07:09] Speaker C: Included AMD Instinct, Mi 325X and the Nvidia H2 hundreds will be available next.
[00:07:14] Speaker D: Month providing more choice and performance options.
[00:07:16] Speaker C: For both training and inference workloads.
[00:07:18] Speaker D: The platform components will support MCP multimodal.
[00:07:21] Speaker C: Capabilities, agent memory and framework integration, simplify moving AI projects from prototype to production. And this will position DigitalOcean to compete more directly with major cloud providers in.
[00:07:30] Speaker D: The AI space by offering a simpler.
[00:07:32] Speaker C: Or integrated alternative for digital native enterprises building their AI apps.
[00:07:37] Speaker B: I mean I'm in support of any feature that Digital Ocean puts on their cloud just because I, you know, rooting for the underdog there.
And if you are a DigitalOcean customer, like how great is it to have, have this and not to go to one of the other cloud hyperscalers and and have maintain two separate infrastructures.
[00:07:56] Speaker E: And I kind of like they're building out their niche market, right? They kind of have for their current customer base who's asking for it and they're finding just the small pieces that these people need. Most people don't need 500 H2 hundreds to build their own model. They're just you know, doing calculations, inference, you know, little things like that. So like this will work for probably 80% of the people out there. So they don't need the massive thing. And if you're doing that massive of a setup, you're probably already on one the hyperscalers that does have this. So I think it's a really good play for them to kind of continue with that niche model that they're kind of, you know, that specific target audience that they're kind of going for.
[00:08:37] Speaker D: Well I have to go play around my digital OJO account. It's been a while since I've done anything over there because I run most of the cloud POD stuff in AWS but it's been a little bit since I've done anything digital osha I should go play because again they've gotten some good features recently that be interesting to see what they can do and are maybe cost effective. So kind of neat.
Well, apparently a report says companies are.
[00:09:00] Speaker C: Canceling their chat GPT subscriptions due to.
[00:09:03] Speaker D: Concerns about data security, cost benefit analysis and integration. Challenges with existing enterprise systems and organizations.
[00:09:09] Speaker C: Report Difficulty justifying the 20 to $30 per user monthly cost when employees use the tool sporadically or for non critical tasks.
[00:09:15] Speaker D: The trend highlights a growing enterprise preference.
[00:09:17] Speaker C: For self hosted or private cloud AI solutions that offer better data governance and compliance controls.
[00:09:22] Speaker D: And companies are exploring alternatives like Azure.
[00:09:24] Speaker C: OpenAI services or AWS Bedrock that integrate with existing cloud infrastructure and security policies.
[00:09:29] Speaker D: Technical teams cite API limitations, lack of.
[00:09:31] Speaker C: Fine tuning capabilities for domains of their tasks and inability to train on proprietary data as key factors.
[00:09:36] Speaker D: Driving cancellations and many organizational steam models.
[00:09:39] Speaker C: That can be customized for industry terminology and workflows.
[00:09:42] Speaker D: The shifts address Enterprises are moving from.
[00:09:43] Speaker C: Experimental AI adoption to more strategic implementation, focus on measured, measurable ROI and specific.
[00:09:49] Speaker D: Use cases and companies are consolidating around.
[00:09:51] Speaker C: Platforms that offer both general purpose and specialized models within their existing cloud environments.
[00:09:55] Speaker D: Development indicates a maturing AI market where.
[00:09:57] Speaker C: Businesses demand enterprise grade features like audit trails, role based access control and integration with existing identity management systems rather than standalone consumer orientated tools.
[00:10:06] Speaker D: I'm not surprised in this one, not.
[00:10:09] Speaker B: Even in the slightest.
[00:10:10] Speaker D: I mean I know I canceled my Chat GPT subscription months ago so I was trendsetter but you know, I think the I can see on the enterprise side in particular you have agent space, you've got AWS q, you've got Microsoft 365, they're integrating into my enterprise stack.
[00:10:27] Speaker C: They'Re getting into my data sources, they're.
[00:10:29] Speaker D: Trying to be able to do more.
[00:10:30] Speaker C: Grounding and rag type operations with my data.
[00:10:33] Speaker D: And ChatGPT just doesn't really have that story in any way. And so if you were just using ChatGPT to go ask questions or help you format emails and things like that.
[00:10:43] Speaker C: There'S a ton of tools that do.
[00:10:44] Speaker D: This, including deep seek for way less money or a local model can do those things without you needing to pay for ChatGPT.
[00:10:51] Speaker C: So I think it's really just a matter of innovation on this has slowed.
[00:10:56] Speaker D: Dramatically and that customers aren't seeing the value of going directly to them for this use case. I don't think their APIs are valuable.
[00:11:02] Speaker C: For embedding it into your product.
[00:11:04] Speaker D: I still think there's lots of other things that ChatGPT is good at, but for this particular area, paying for Enterprise Chatbot, I just don't think it's something that makes sense based on what competitors are doing. So maybe, maybe this will be the next announcement we talk about next week when OpenAI shocks us all with their new solution to this problem.
[00:11:19] Speaker B: Maybe I mean, because it really does feel like they fell behind the rest of the market after leading it for so long. Like in, like you said, like some of those innovations they're on top of just the, you know, the, the models themselves. And so it's because it really is a big difference between like generate me a pretty picture of a thing versus you know, something that has a business deliverable and can actually output and the business deliverable.
[00:11:47] Speaker E: Here's the key part. Most of the other services, like Justin said, integrate in with everything. So if you look at Copilot reading my email, it's, you know, I'm like, hey, find me this thing. And it can go through that whole thing and for the same $30 a month you get a lot more value out of it. Plus you'll have to make sure it's set up and it's another tool and everything else, you know. So to me this is really like people being like, okay, this was good, but there's other ones that are better now.
I mean it wouldn't surprise me if in a year or two ChatGPT kind of falls off the model kind of falls off the wagon of what we talk about as all the other models have kind of jumped up and done a lot of the same things and I, I don't know that they are that, you know, front runner anymore and everything.
[00:12:34] Speaker B: Yeah, it'll be interesting because it's the, the for profit model side of those things. Like it seems to be where they're sort of falling down or maybe it's just a distraction. If they go back to sort of the, you know, the, the think tank building models and, and really doing R and D on, on, you know, building out AI from looking under like not at a service level but at the pure technical level, it might find their way again. We'll see.
[00:13:03] Speaker D: I wonder how much of it also could be either limited capacity on Azure.
[00:13:07] Speaker C: So they can't launch some of these.
[00:13:09] Speaker D: Features, which is a possibility.
It also could be that again, like you said, their mission is more focused on AGI.
[00:13:16] Speaker C: I mean they have Stargate project they're.
[00:13:18] Speaker D: Doing, they're trying to get a lot more compute capacity to do things or you know, it's, it's just they, you know, they're not focused on consumer or on enterprise needs, they're focused on something else and we just don't know what that is yet.
Well, the 2025 DORA survey is now open until July 18th offering teams a.
[00:13:42] Speaker C: 10 to 15 minute self assessment tool to benchmark their software delivery and operational performance against industry standards. This, your survey focuses heavily on AI adoption. No.
[00:13:51] Speaker D: In daily work companies applying DORA principles have achieved dramatic improvements in their work. Basically SLB cut deployment deployments from 5 to days to 3 hours.
[00:14:02] Speaker C: GitLab reduces errors by 88% and these metrics demonstrate the tangible values of continuous improvement practices.
[00:14:06] Speaker D: Backed by Data Insights survey explores how organizations can maximize AI impact while maintaining developer well being. Find the transparent AI strategies and governance policies significantly increase adoption rates and also.
[00:14:17] Speaker C: Examines trust in AI systems and how teams can best support transition to AI.
[00:14:20] Speaker D: Enhanced workflows it's available in six languages. The survey welcomes input from all software.
[00:14:24] Speaker C: Delivery roles, engineers, product managers, CISOs, UX.
[00:14:27] Speaker D: Designers, and even executives like me to capture diverse perspectives on team performance. Participants gain immediate value through structured reflection on their workloads and workflows and bottlenecks. Do research continues to shape industry understanding of high performing teams while finding findings.
[00:14:40] Speaker C: Like the substantial impact of quality documentation on team performance.
[00:14:43] Speaker D: Anonymous data collected helped establish benchmarks and the best practices I don't think the July 15th date was correct.
[00:14:49] Speaker B: July 18th, I think it is. I was just clicking through because I was just looking at that going, I don't. It's going to be after this episode's published.
[00:14:57] Speaker D: They just dropped this article on July 11th.
[00:14:59] Speaker B: They really did. But yeah, to participate before July 18th.
[00:15:02] Speaker D: I suspect that this is.
Maybe it does say that, but I think maybe it goes till, I'm going to guess August.
I think it's a typo.
[00:15:11] Speaker B: I'm surprised because it. It's not usually such a short.
[00:15:14] Speaker D: Yeah. Normally it's like a month. So that's why I'm thinking it's. It's.
[00:15:20] Speaker C: A typo.
[00:15:20] Speaker B: Yeah. I was trying to run through the.
[00:15:24] Speaker E: It does say open till July 18th.
[00:15:26] Speaker D: Yeah.
[00:15:26] Speaker E: According to.
According to the article.
[00:15:29] Speaker D: That's why. Yeah.
[00:15:30] Speaker B: Yeah.
We're trying to, you know, have a different source, but yeah.
[00:15:35] Speaker D: Anyways, that's a.
[00:15:36] Speaker C: That's sort of strange.
[00:15:36] Speaker D: I would be surprised.
[00:15:38] Speaker B: Well, hopefully it's still open.
Yeah, hopefully. So when listeners hear this.
[00:15:45] Speaker D: Yeah. So the Dora Dev site, which is the official Dora website, doesn't say anything about when it closes. So like I said, I, I suspect it's incorrect. But yeah, problem with web stuff sometimes.
[00:15:59] Speaker B: Yep.
[00:15:59] Speaker D: So get your survey in as soon as possible. Especially if you listen to this episode and it's already closed. I apologize. But it's still open. Then we were right. And it's going to August, which is what I typically have seen in the past.
So we cover a lot of news here and sometimes things slip through or.
[00:16:16] Speaker C: We purposely kill them.
[00:16:18] Speaker D: And then sometimes Matt drops something. I think he was like, I missed this while I was out on leave with my new kid. And I was like, I don't remember that.
And he's right.
We missed our article back in April, which is kind of actually cool. So they introduced in April. And so if you see this in.
[00:16:34] Speaker C: The console, which is how Matt found.
[00:16:36] Speaker D: Out about it and you're wondering why we didn't talk about it, it's because we missed it. But AWS Systems manager now Offers just in time node access, enabling temporary policy based access to EC2 on premises and.
[00:16:47] Speaker C: Multi cloud nodes without maintaining long term credentials or SSH keys.
[00:16:51] Speaker D: This addresses the security versus operational efficiency.
[00:16:53] Speaker C: Trade offs many organizations face when managing thousands of nodes. The feature supports both manual approval workloads with multiple approvers and automated approval using Cedar policy language, allowing organizations to implement zero standing privileges while maintaining rapid instability incident response capabilities.
[00:17:07] Speaker D: Access automatically expires after a defined time window. Integration with Slack, Microsoft Teams and email.
[00:17:12] Speaker C: Notifications streamlines the approval process, while EventBridge events enable audit trails and custom workflows.
[00:17:18] Speaker D: Sessions can be logged for commands and RDP recordings for compliance requirements and AWS.
[00:17:23] Speaker C: Offers a free trial period covering the remainder of the current billing period plus the entire next billing period per account.
[00:17:28] Speaker D: Per region, after which pricing is usage based.
[00:17:31] Speaker C: This allows organizations to test configurations and policies before committing to the cost.
[00:17:36] Speaker D: Ellusian works seamlessly across AWS organizations, supporting consistent access controls whether managing single or.
[00:17:41] Speaker C: Multiple accounts, with administrators defining policies, operators requesting access, and approvers managing requests through a unified console experience.
[00:17:50] Speaker B: I mean how unlike AWS to launch something that's so fully featured, all the integrations and the org support. Like it's kind of crazy.
[00:18:01] Speaker E: Yeah, no, it seems like a cool feature. It runs on Jonathan's favorite method of security which is through tags. So you know a lot of the automation, you know be like okay dev person can automatically get access if tag equals dev is in there. So there are some features or setup design of it that might not be the what works for your company but and there's some like prep work if you want to use it. But it does seem like a really nice feature, you know, especially when you're dealing with multiple teams and multiple organization or accounts and getting all those notifications which to me is kind of the key part.
[00:18:38] Speaker B: A lot of the tagging concerns have really been done away with too because there's a lot more granularity in the IM permissions where you can associate a tag but you can't create a tag or you can't modify a tag. Now like there's there's different API granularities that you had or you didn't have before and I don't know in aws but in GCP you can set tags centrally and set the value centrally as well and then use policies based on that. So it's become a pretty powerful tool and it's a lot better than overloading the host names or trying to do it some other maintaining these static lists of resources.
But this is, you know like just in time access.
I'm super stoked on in general. Been working towards that for a while in my day job and I don't know why we continue to have machines with, you know, access or active directory. Like it just, it seems so old school now.
Love this.
[00:19:38] Speaker E: It will never die, especially if you're in a Windows environment.
[00:19:41] Speaker B: But it can and it should. This supports Windows machines. It'll do rdp.
[00:19:46] Speaker E: Yeah, but then how do your server to server communicate? You have to have the AD somewhere.
[00:19:51] Speaker B: No, you don't.
You can do it with ad. It's sort of like built into the. NET framework.
But you could also just not do chat tokens or services that requires a.
[00:20:04] Speaker E: Developer to do work.
[00:20:06] Speaker B: I'm aware.
[00:20:09] Speaker E: The pricing of this is pretty interesting. I don't know if you guys looked at that, but it's per server, per node, per hour, billable rate.
[00:20:21] Speaker D: It sounds just like every other just in time access tool out there.
[00:20:25] Speaker B: So the agent has been that way for a long time.
[00:20:28] Speaker E: But I thought the agent was free.
[00:20:33] Speaker D: The base agent is free.
[00:20:34] Speaker E: Yeah.
[00:20:35] Speaker B: But to get into the systems manager where you're managing thing, I think there's a cost. I thought.
[00:20:41] Speaker D: Yeah, yeah. If you're using like the patch thing or using, you know, the regular like you want automated checks and configuration drift stuff that costs money if I recall.
[00:20:50] Speaker E: Oh, I guess I've never dealt with that that much. Yeah, I thought the patching was free, but maybe I'm wrong.
[00:20:56] Speaker D: Or actually it may. It may actually cost money, Ryan. But I think it's free for the.
[00:21:00] Speaker C: First like 10,000 systems.
[00:21:02] Speaker B: There's DEF. There's definitely the free.
[00:21:04] Speaker D: It's a very healthy free tier.
[00:21:05] Speaker B: The free tier is definitely a thing. Sure.
[00:21:08] Speaker E: Maintenance windows are free on prem. Instances. Definitely cost money. I remember that.
[00:21:13] Speaker D: Yeah, that's definitely what it costs.
[00:21:17] Speaker E: Session manager was free. Pass manager is free. I think those are really only the features of this that I've used.
[00:21:23] Speaker D: Yeah. So just in time, node access is an upcharge.
[00:21:25] Speaker E: Yes. No, I was looking at it for sure on that.
[00:21:28] Speaker D: So yeah, 72,000 hours is basically a penny per node hour.
And then you have.
[00:21:35] Speaker B: And that's connected to the node.
[00:21:37] Speaker E: Yeah. So the system boots up. So if you have 20,000.
[00:21:41] Speaker B: Or is it. Yeah, runtime.
[00:21:43] Speaker E: Runtime.
[00:21:43] Speaker B: So I was trying to figure that out.
[00:21:45] Speaker E: So the way they Describe it is 20,000 node hours, which would be roughly 200 SSM managed nodes were connected each for 100 hours.
Which is weird because, you know, most time these systems then would be up 24,7 because you know, people don't turn stuff off ever. You know, then it's only. It's $274. So 100 systems for 100 hours a month. So working hours roughly.
No, yeah, a little bit less. Half. Half working hours a month.
But also you probably don't have 20,000 dev systems slash.
[00:22:23] Speaker B: I hope not.
[00:22:25] Speaker C: They also charge you for op center.
[00:22:27] Speaker D: Usage in SSM Incident Manager.
[00:22:31] Speaker C: They charge you for app config.
[00:22:33] Speaker D: They charge you for app config.
[00:22:35] Speaker E: Was fading away. You were wrong.
[00:22:38] Speaker B: Nah, App config is the configuration like on demand.
[00:22:42] Speaker D: Yeah, yeah, but yeah, basically it's very, very cheap though. It's like configuration we see 0.0008 so you know of a penny. Pretty cool. Pretty low.
[00:22:54] Speaker C: Parameter Store has no additional cost for standard tier, but the advanced tier does have a cost.
[00:23:00] Speaker D: And then there are some pricing for.
[00:23:01] Speaker C: API interactions on it. Change Manager has a cost, Automation has a cost.
No charge for App Manager maintenance, Windows compliance, Inventory run, Command State Manager, Fleet Manager.
[00:23:13] Speaker D: Those are all free.
[00:23:14] Speaker C: And then we do pay for the on premise instances. Session manager is free, Patch Manager is.
[00:23:19] Speaker D: Free and distributor is free as well.
[00:23:23] Speaker C: Other than storage for the non AWS.
[00:23:25] Speaker D: Packages you deploy through distributor, which is typical S3 pricing. So there you go. Lots of free goodies with ssm. But not everything is free.
[00:23:33] Speaker E: I think most of the features I use are free. The only one I think I've used is Automations that it costs money but it's very minimal I feel like.
And the first 125,000 steps are free in an account, so I feel like I don't even think I hit the, the tier that I would pay for it.
[00:23:54] Speaker B: But can you really put a cost on not having to manage Active Directory? Like it's just so worth it.
[00:24:00] Speaker E: Yeah, well, but you're probably managing Active Directory to, to manage SSO into your AWS organization.
[00:24:06] Speaker B: No, no, no, no, no, no.
[00:24:09] Speaker E: That's somebody else's problem.
[00:24:12] Speaker B: No, it's just not required. It is so it is the number one pattern you see everywhere. But it's. There's so many other ways to do it.
Don't do it.
[00:24:24] Speaker D: Well, I mean you can do it, it's fine.
[00:24:29] Speaker C: Just doesn't make security happy.
All right, let's move on to Kiro is the new AI powered IDE that introduces spec driven development, automatically generating requirements, technical designs and implementation tasks from simple prompts to help developers move from prototype to production ready applications.
[00:24:47] Speaker D: The platform's key innovation is its Specs.
[00:24:49] Speaker C: Feature which creates ears, notation, acceptance criteria, Data Flow diagrams TypeScript interfaces and database schemas as they synchronize with the evolving code base. Addressing the common problem of outdated documentation.
[00:25:00] Speaker D: CURA hooks provide automated quality checks by.
[00:25:03] Speaker C: Triggering AI agents on file events, for.
[00:25:05] Speaker D: Example automatically updating test files when react.
[00:25:06] Speaker C: Components change, or scanning for security before commits, enforcing consistent standards across development teams.
[00:25:13] Speaker D: It's built on VS code or Code OSS with VS code compatibility.
[00:25:17] Speaker C: Kira supports model context protocol for specialized tool integration and is currently free during preview with some limitations targeting developers who need more structure than the typical AI coding assistants provide. This represents a shift towards more structured.
[00:25:28] Speaker D: AI assisted development, moving beyond simple code.
[00:25:30] Speaker C: Generation to address production concerns like maintainability, documentation and team consistency. The traditional AI coding tools often overlook.
[00:25:38] Speaker D: And I downloaded it and I've been playing with it most of the day building a mobile app cross platform, which I've never done before and I have.
[00:25:47] Speaker C: No experience doing and I have no.
[00:25:48] Speaker D: Idea what it's doing but so it's working great. Yeah, it's working great.
But you know, there's a couple things about it. I've used cursor, I've used Big rub code fan as we talked about a bunch of times.
And so you know, it does have a much more interesting starting process. So you know, you start at that prompt place where you're saying look, I want to create an app that does these basic things and you kind of describe it, it asks you questions back, which is really quite nice. And then it basically after those questions and you're chatting with it, it then.
[00:26:19] Speaker C: Produces a requirements doc which is a set of user stories.
[00:26:23] Speaker D: So like user story number one, as a podcast listener, I want to browse.
[00:26:27] Speaker C: All available episodes from the Cloud pod.
[00:26:29] Speaker D: So that I can discover and select.
[00:26:30] Speaker C: Content to listen to.
[00:26:31] Speaker D: Amazing.
After I got through requirements and I agreed on these and I, you know, there's a couple things I said no, you know, I don't like like that feature exactly. I want to change it a little bit. And so I was able to tweak.
[00:26:41] Speaker C: It either directly in the document or by chatting with the bot.
[00:26:44] Speaker D: Then created a design spec and the design spec basically goes over the entire details. It provides a mermaid architecture diagram of how the app will work, basically talks about the technology stack, the components, service interfaces, audio service interface, storage, data models.
[00:27:00] Speaker C: Playback, et cetera, and all the error handling and testing strategy for my new Android and iOS app app.
[00:27:06] Speaker D: And then it broke all of that down into a 11 step task list.
[00:27:13] Speaker C: That'S more detailed than the task list.
[00:27:15] Speaker D: I've seen on the other chat things. It's like, you know, first one is.
[00:27:17] Speaker C: Set up a project structure and develop environment.
[00:27:19] Speaker D: Then it has subtasks, initialize, react, native figure, typescript, set up, project directory structure, install and configure. And it says this is a requirement for 3.1 and 3.2. So unlike a lot of other ones of these, they don't always tell you what the order of sequence needs to be. And so if you want to jump.
[00:27:34] Speaker C: Around, it'll tell you why you can't.
[00:27:35] Speaker D: Because you haven't fulfilled one of its requirements to even start the task, which is great. And then gives you basically when you're ready to start, there's a button that says start and gives you the typical interface on the right hand side, similar.
[00:27:48] Speaker C: To what you see in root code.
[00:27:49] Speaker D: On the left hand side, which is basically your chat prompt where you can chat with it and say let's do. And you can literally in the task list you can check the task you want to do first and it kicks off the process and it goes to the same process of asking do you want me to run this command? Do you want me to allow me.
[00:28:01] Speaker C: To always run those commands for you.
[00:28:03] Speaker D: You can also set up a bunch of things on the left hand side. It has a spec section, it has agent hooks for all your MCP stuff, agent steering, you can generate that and then MCP servers is connecting to for its purposes for what you want to do. Relatively simple, relatively intuitive. I like the UI quite a bit compared to Cursor which are both of these are built on.
[00:28:26] Speaker C: BS code.
[00:28:27] Speaker D: Basically after each task completes, it gives you a full task implementation summary. These are all saved so you have all the summaries versus some of the.
[00:28:35] Speaker C: Things a root code, you sometimes lose.
[00:28:36] Speaker D: That if you don't save it off or don't have a process to basically.
[00:28:40] Speaker C: Take those things and put them in documentation.
[00:28:42] Speaker D: And so overall I'm pretty impressed For a first cut of this, they do have pricing on the website or what they're recommending is going to cost when it gets out of preview. Right now it's free, so I'm just burning tokens, building this mobile app, going to burn them live. They're free and it uses right now clots on it 3.7 and 4.0. Those are two options.
[00:29:02] Speaker C: I was a little surprised Nova wasn't.
[00:29:04] Speaker D: Here, but that just shows you maybe Nova's not ready for coding exercises right now. They say there will be a free tier which will be $0 per month with agentic capabilities for the Hero IDE, up to 50 interactions per month specs, agent hooks, MCP and agent steering, you'll be able to get, for $19 a month, you'll be able to increase your interactions to a thousand interactions per month. And for $39 per month, the cell, you'll be able to get 3,000 interactions per month. Now, the one challenge is I don't know what an interaction is.
[00:29:31] Speaker E: That's what I was just wondering.
[00:29:33] Speaker D: I barely know what a token is, so it's definitely something. I think it's an issue where they need to figure out how to make that clear to people so they know what they're doing.
[00:29:42] Speaker B: Especially if the part of the process is the bot asking you questions, some of that is going to be outside of your control.
But overall, I think this is just how coding is going to look in the future. Like, you know, I think gone are the days of like SCRUM meetings where you're all sitting around putting tasks on a JIRA board and trying to organize it that way.
I think it'll be much more automated, much more built into the system.
And I, you know, like, I think eventually these types of things will. There'll be groups of, you know, ides and a team working together. So it's, it's going to be a huge change to how we develop. But I'm, I'm pretty excited for it because I. More and more I use it the more I'm just not mired into a lot of these details. And as long it does still require a lot of adult supervision.
[00:30:35] Speaker E: Oh yeah.
[00:30:36] Speaker B: But, but it is a. You know, it's really magnified my work output a ton. So things like this are just going to make it better, formalize the projects more, you know, structure it better. So I look forward to trying this.
[00:30:51] Speaker E: Yeah, I played with it a little bit and I like, I like the pattern that I kind of worked you through. I know with Roux, you can kind of make it do that. You can change the type that's in there, you know, like what you're, how you're interacting with it and what you're doing. But I liked how it kind of built out those documents and I've always disliked putting stuff. While I love Confluence and the idea of it, as soon as you have to integrate source code or any logic flows or anything like that into it, Confluence immediately out of date. Because no developer, no one actually ever remembers to update Confluence when you're already in git. And having everything in a markdown is better. So I like the concept of having it all in those markdown files. And I did like a simple bash script that I was playing with. And you know, it is far more overkill the way I designed it because it's, you know, handling multiple pieces and iterations and test logic and all that stuff in there.
But I like the way it kind of broke it down from like, is this what you want? Okay, let's design it and then let's break it up into tasks, which is, you know, what I naturally do and what I've taught many different people that have worked for me to do, which is take a project, break it down, you know, and kind of design and break it down. And this just formalizes that structure and doing it in such a way that you can then really get the. The bots and to. To develop for you and do the like, the individual chunks of work that are single items that it can handle. Because if you said, I'm sure just develop this from scratch, it would not give you what you want. So once you broke it down and iterated through it and iterated over each of those steps from, you know, design, require or requirement design to tasks, you've done that breakdown for an AI bot to do it or to do, you know, for somebody in your team to work on it with a bot and then to develop it and to have the output. So I like the user workflow that.
[00:32:52] Speaker B: They did with that.
[00:32:54] Speaker D: Yeah, for a V1. Very happy. This is also very full featured. What's going on? Amazon, right?
It is interesting they didn't decide to brand this with Amazon or AWS Kiro, partially because I think they're trying to avoid confusion with Q IDE tools. But also it didn't work out for honeycode, so I hope it works better for Kiro. And then my only other comment is.
[00:33:19] Speaker C: They made a really cute little ghost logo as the logo for this thing.
[00:33:24] Speaker D: And all I can think of is that must be the ghost of all.
[00:33:26] Speaker C: The engineering jobs they've killed.
[00:33:27] Speaker D: Or product managers.
[00:33:31] Speaker B: Nice.
[00:33:32] Speaker E: It does say in the FAQ here during the preview, curo is free for everyone to use, including developers. Q Developer Pro subscribers can log into their existing via their iam.
[00:33:44] Speaker C: Interesting.
[00:33:44] Speaker E: They kind of have it, but not really.
[00:33:47] Speaker D: Yeah, well, it'll be interesting to see how it evolves. I already see there's an update for mine today that I should probably install at some point.
[00:33:53] Speaker C: So they're already getting bug fixes out.
[00:33:55] Speaker D: Probably from feedback today is they launched. Well, it launched yesterday morning actually, so it's been a full 24 hours. I assume you'll hear a lot about this at the summit tomorrow as well.
[00:34:05] Speaker E: So wait for the security person. Dlu, you're two days out of date.
[00:34:10] Speaker C: No, he's right here on the call.
[00:34:11] Speaker D: So he doesn't yell too much.
[00:34:14] Speaker C: All right.
[00:34:15] Speaker D: AWS is launching the P6E GB200 Ultra.
[00:34:19] Speaker C: Servers with Nvidia Grace Blackfell GPUs offering up to 72 GPUs in a single NV link domain with 360 petaflops of.
[00:34:26] Speaker D: FP compute and 13.4 terabytes of HBM3e.
[00:34:30] Speaker C: Memory for training trillion parameter AI models.
[00:34:33] Speaker D: The new instance uses Nvidia superchip architecture.
[00:34:36] Speaker C: That combines Blackwell's GPUs with Grace ARM CPUs on the same module, providing significantly.
[00:34:41] Speaker D: Higher GPU to CPU bandwidth compared to current P5EN instances while delivering 28.8 terabytes per second of EFA networking. P6E GB200 ultra servers are only available.
[00:34:52] Speaker C: Through EC2 capacity blocks for machine learning.
[00:34:54] Speaker D: In the Dallas local zone, requiring upfront.
[00:34:56] Speaker C: Payment for reserve capacity blocks of either 36 or 72 GPUs with pricing determined at purchase time.
[00:35:02] Speaker D: Integration with aws services includes SageMaker Hyperpod for managed infrastructure with automatic fault replacement.
[00:35:06] Speaker C: Within the same nvlink domain.
[00:35:08] Speaker D: EKS with topology where routing for distributed.
[00:35:10] Speaker C: Workloads and FSX for Lustre providing hundreds of gigabytes per second throughput for large scale AI training.
[00:35:15] Speaker D: This is Target Frontier AI workloads including.
[00:35:17] Speaker C: A mixture of experts models, reasoning models and generative AI applications like video generation and co generation positioning AWS to compete with the high end AI infrastructure market.
[00:35:27] Speaker B: So if you're a big enough Amazon customer, you can get Amazon to run your Amazon Outpost with custom hardware. Cool.
[00:35:34] Speaker D: Yeah, that's what it sounds like, right?
I was like, what big companies are in Dallas that might have wanted this? I was trying to work that out earlier.
[00:35:41] Speaker B: Yeah, I couldn't figure that out either, but it's kind of a crazy thing to be available in a single local zone. Never heard such a thing.
[00:35:51] Speaker D: Well, maybe this has a lot of.
[00:35:52] Speaker C: Power requirements and so this is the.
[00:35:54] Speaker D: Only place that has a power plant.
[00:35:55] Speaker C: Large enough to handle this need?
[00:35:57] Speaker D: I don't know.
[00:35:58] Speaker B: Yeah, just since the nuclear deal went fell through right now this is the only thing it works out.
[00:36:04] Speaker E: AT&T, Texas Instruments and Southwest Airlines are all headquarters in the Dallas Fort Worth area.
[00:36:10] Speaker D: Well, I mean, Southwest can't even spell AI based on their infrastructure issues, much less.
[00:36:15] Speaker B: Yeah, they can't even patch generally.
[00:36:17] Speaker E: Aren't they on like XP or 31 or something? Was that they're like we didn't learn.
[00:36:22] Speaker B: Anything good about their setup, that's for sure.
[00:36:24] Speaker D: Yeah, so it's not that allegedly on the Windows 3.
[00:36:28] Speaker B: Oh right now I highly suspect this is AT&T somehow, but yeah, that makes sense.
But yeah, no, it's kind of crazy.
[00:36:41] Speaker A: 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.
[00:37:21] Speaker D: AWS Builder center has gotten a major remodel. It now consolidates developer resources from AWS.
[00:37:27] Speaker C: Builder center and the Community AWS into.
[00:37:29] Speaker D: A single
[email protected], providing you a unified.
[00:37:33] Speaker C: Hub for accessing tutorials, workshops, and community engagement tools. The new Wishlist feature allows developers to submit and vote on feature requests for AWS services, giving the community direct input into product roadmaps and enabling AWS aims to prioritize development based on actual user needs.
Built in localization supports 16 languages with.
[00:37:50] Speaker D: On demand machine translation for user generated.
[00:37:52] Speaker C: Content, removing language barriers for global collaboration.
[00:37:54] Speaker D: Among AWS builders and expanding accessibility to.
[00:37:57] Speaker C: Non English speaking developers.
[00:37:59] Speaker D: The platform integrates AWS Builder ID for.
[00:38:01] Speaker C: Consistent profile management across all AWS services, offering personalized Profiles with custom URLs or Q& QR codes for networking and events and conferences.
[00:38:09] Speaker D: Connect features highlight AWS heroes, community builders, user groups and cloud clubs made it easier to find local meetups and connect with experts in specific idios services Service.
[00:38:17] Speaker C: Areas of technology so this is nice.
[00:38:21] Speaker D: It's a little bit of a bunch of things that they kind of had hodgepodge all over the place kind of.
[00:38:25] Speaker C: Pulled together into one single home, which.
[00:38:26] Speaker B: I kind of like, yeah, I'm gonna have to like dredge up my Builder ID because I don't.
[00:38:31] Speaker D: I'm pretty sure it's tied to your Amazon id and somehow it is.
[00:38:34] Speaker B: Yeah, but I don't remember exactly how, but yeah, no, it's kind of. It's a crazy thing. They had like various different things that always made it feel like separate systems.
So I think this is good.
I haven't looked at it yet. So it's like I'm a little nervous because it's Amazon's unified ui.
Consoles have been a little bit weak lacking, so we'll see. But I do really like the, you know, the feature request sort of option now you don't have to have an account team that can go chase your feature through the product teams. It's much more community based so that's cool.
I like it.
[00:39:16] Speaker D: AWS Pricing List API now supports four new query filters.
These enable extract exact attribute matching substring.
[00:39:23] Speaker C: Searches and include exclude lists for more targeted product searches across AWS services.
[00:39:27] Speaker D: The update simplifies finding specific product groups.
[00:39:29] Speaker C: Like all M5EC2 instances types with a single filter instead of multiple complex queries, reducing your API calls and improving efficiency. Is that good for us or for Amazon?
This enhancement addresses a common pain point for cost optimization tools and FinOps teams who need to programmatically analyze AWS pricing data across thousands of SKUs.
[00:39:47] Speaker D: The new filters are available in all.
[00:39:49] Speaker C: Regions where the price list API is.
[00:39:50] Speaker D: Supported, making it easier for organizations to build automated pricing analysis and comparison tools.
[00:39:55] Speaker C: And real world applications include building custom cost calculators, automated pricing alerts, and multi region price comparison tools for reserved instance planning.
[00:40:04] Speaker D: I do find it funny. This is a sidebar here. AWS CLI filtering, which is also just basically a representation of the API filtering is one of those things that just drives me crazy because I never really remember it properly.
[00:40:18] Speaker C: And it brings me such joy watching.
[00:40:20] Speaker D: The AI bots also screw it up. I'm like, okay, if the AI bot who has the documentation in its brain memorized can't get this right, I don't feel so bad.
[00:40:30] Speaker B: Yeah.
[00:40:31] Speaker D: So just a little side Note there on APIs and filtering.
[00:40:34] Speaker B: What do you mean? You just can't remember the schema model of 19 layers deep to know that that's.
[00:40:42] Speaker D: Even looking at the documentation.
[00:40:44] Speaker C: I typically get it wrong for the.
[00:40:45] Speaker D: First two or three tries and I've literally watched when I'm running ECS debugging and I'm like, hey, go grab the logs to figure out what happened there. And it's like trying to figure out the current log group, just can't get the syntax quite right and the time is wrong. And then it's like, oh yeah, the Mac has a weird time format that.
[00:41:02] Speaker C: I need to convert to the weird.
[00:41:03] Speaker D: Format that Amazon uses in UNIX time. It's just like, just like, ah yes, this is amazing.
[00:41:09] Speaker B: Yeah, it is kind of funny because I have, I've, I've often felt that AI is just making the same mistakes that I go through just faster.
[00:41:17] Speaker E: So it is.
[00:41:17] Speaker B: Which is pretty great.
[00:41:18] Speaker E: It's 100% is.
[00:41:19] Speaker B: But it's such a big confidence boost because, you know, I always think I'm just a bumbling idiot and. Because I am. But.
But so is AI. So, you know, we're good.
[00:41:28] Speaker D: Yeah. I mean, I've seen AI make the same mistake multiple times in a file. So like it's not always the smartest in the world, but it does bring me a lot of joy just every time. Like, yes, thank you for that. I appreciate that you messed that up too.
And then normally their code is more complicated than I would even attempted. So I'm like, I mean you're a bot that can take in in context Windows. I would have just done the I would have taken the whole list of EC2 instances. I wouldn't even try to filter that. Just found it that way.
[00:41:52] Speaker B: Y Pipe Grip.
[00:41:54] Speaker C: Yep.
[00:41:58] Speaker D: Well, if you were like, man, I.
[00:41:59] Speaker C: Don'T want to use the API for priceless. Well, Amazon also has you this week was releasing an open source model Context.
[00:42:06] Speaker D: Protocol server that gives AI assistance like.
[00:42:08] Speaker C: Amazon Q Developer, CLI and Cloud Desktop.
[00:42:10] Speaker D: Direct access to AWS pricing data, including.
[00:42:13] Speaker C: On demand, reserved and saving plan options across all regions.
[00:42:16] Speaker D: MCP Server enables natural language queries but.
[00:42:18] Speaker C: AWS pricing and product availability allowing developers.
[00:42:20] Speaker D: To ask questions like list the cheapest.
[00:42:22] Speaker C: EC2 instance for machine learning in US.
[00:42:24] Speaker D: East1 and get real time responses from the AWS pricing list API. This addresses a common pain point where.
[00:42:29] Speaker C: Engineers manually navigate complex pricing pages or write custom scripts to compare costs across services and regions.
[00:42:34] Speaker D: And now just make the AI do it for you. Server uses standard AWS credentials and minimal.
[00:42:39] Speaker C: Configuration, making it straightforward to integrate into existing workflows where teams already use AI assistance for development tasks. Available now to you via the AWS Labs GitHub repository.
[00:42:48] Speaker E: Okay, so I have a question.
[00:42:50] Speaker B: Yes?
[00:42:50] Speaker E: When was the last time you ever had an engineer go in to figure out what EC2 instance type?
[00:43:00] Speaker D: Do I count as the engineer?
[00:43:01] Speaker E: No, no, no, sorry, a developer that is not a cloud engineer. Because everyone I've met just says, ooh, this one's big and shiny and we'll put more power behind it. That makes my code run faster.
[00:43:16] Speaker D: Yeah, it was slow. I just bumped up it up and now it's not slow and it's like, yeah, okay, thanks.
[00:43:21] Speaker B: But think about it this way. You're making it finally easy for them to interface with, right? So all they have to do is ask a natural language question.
[00:43:29] Speaker E: Why is My software then they can ignore increase the size here. Here's the MCP that will tell me, oh, this server is only $0.40 more an hour. Don't worry about your, your CFO's brain exploding on the other side of it.
[00:43:44] Speaker B: Although on the plus side, I was thinking about, you know, a feature that Matt and I worked on where we were defining spot fleets and this would have you know, like hours and hours of defining mappings and groups of various features. Like now it'd just be replaced by prompts. Fantastic.
[00:43:59] Speaker E: Or just the API above. As long as it's not the cli. Though Boto is not much better at times how many lists and it's typically the API structure.
[00:44:09] Speaker B: No, not cli.
[00:44:12] Speaker D: Yeah, the nice thing is that because this is not MCP and you know, between AI's ability, you can point AI at an Amazon account and say, hey.
[00:44:22] Speaker C: Give me recommendations for savings.
[00:44:25] Speaker D: And then you can, you can ask it like, it's like be a finops person and tell me all the ways I can I can save. Awesome. It has pretty good recommendations. I've done it a couple times now on my account just to see what it would come back with.
[00:44:36] Speaker C: And it's.
[00:44:36] Speaker D: It's had sound recommendations in a couple different cases. One, one recommendation saved me a couple hundred bucks a month.
[00:44:43] Speaker E: So.
[00:44:43] Speaker B: No, that's cool. I would, I. It would make me nervous on a larger workload because of just the amount of interpretation and decisions that have to.
[00:44:51] Speaker D: Be done and tell it to do the savings. I just told it to tell me what the savings could be, he said.
[00:44:57] Speaker E: But not that it seemed right.
[00:44:59] Speaker B: Got it, got it.
[00:45:00] Speaker D: Yeah.
Amazon documentdb global clusters now support up.
[00:45:05] Speaker C: To 10 secondary regions.
[00:45:08] Speaker B: Yeah, yeah, I don't think you read that right.
[00:45:10] Speaker E: Hold on, reread that correctly.
[00:45:13] Speaker D: Sorry. Apologies, Ghost of Peter.
[00:45:17] Speaker C: Amazon DocumentDB with MongoDB compatibility introduces support for up to 10 secondary region clusters.
[00:45:23] Speaker D: Amazon DocumentDB global clusters now support 10.
[00:45:26] Speaker C: Secondary regions, doubling the previous limit of.
[00:45:29] Speaker D: Five, enabling broader geographic distribution for applications.
[00:45:32] Speaker C: Requiring low latency reads across multiple continents.
[00:45:35] Speaker D: This expansion addresses disaster recovery needs by.
[00:45:37] Speaker C: Allowing organizations to replicate their MongoDB compiled workloads across more AWS regions, reducing the.
[00:45:42] Speaker D: Plast radius or as I see it.
[00:45:43] Speaker C: Increasing the chances of needing the doctor.
[00:45:45] Speaker D: Plan of regional outages while maintaining local read performance. This increased region support particularly benefits global.
[00:45:51] Speaker C: Enterprises running customer facing apps that need to comply with data residency requirements across multiple jurisdictions while maintaining consistent while the feature enhances ability and global reach. Customers should consider the cost implications of running clusters across 10 regions, including Cross region data transfer charges, which will burn you pretty hard.
[00:46:08] Speaker D: This position stockw more competitive against MongoDB.
[00:46:10] Speaker C: Atlas, which supports similar multi region deployments.
[00:46:13] Speaker D: Giving AWS customers a fully managed alternative without leaving the AWS ecosystem.
[00:46:18] Speaker B: I meant to click through and read on this, but I was trying to figure out if this was like a specific 10 regions where you could set up these replicas or if it or is it more like you can configure 10 globally. Just pick your 10 of whichever Amazon region supports documentdb.
[00:46:37] Speaker E: I think you you specify I didn't click through it, but you specify the.
[00:46:41] Speaker D: Regions any place where DynamoDB is supported I imagine, and then as long as you only combine 10 of them together, you'll be okay.
[00:46:48] Speaker B: That's a lot.
That's a lot.
[00:46:50] Speaker D: It's. I mean it's a lot of sharding.
[00:46:52] Speaker C: To do.
[00:46:56] Speaker D: Amazon SageMaker Studio now supports.
[00:46:58] Speaker C: Remote connections from Visual Studio code, which.
[00:47:01] Speaker D: Allows developers to connect to their local.
[00:47:02] Speaker C: VS code installation directly to SageMaker's Managed Compute Resources, reducing setup time from hours to minutes while maintaining existing security boundaries.
[00:47:10] Speaker D: Developers can authenticate through the AWS toolkit.
[00:47:12] Speaker C: Extension or SageMaker Studio web interface, then access their SageMaker development environment with a.
[00:47:16] Speaker D: Few clicks with while keeping their preferred VS code extensions and AI assisted development tools. This addresses a common friction point where.
[00:47:22] Speaker C: Data scientists want their familiar local IDE.
[00:47:24] Speaker D: Setup but need access to scalable cloud.
[00:47:26] Speaker C: Compute and datasets stored in AWS without complex SSH tunneling or VPN configurations.
[00:47:30] Speaker D: This feature compliments SageMaker Studio's existing Jupyter lab and code editor options, giving teams.
[00:47:36] Speaker C: Flexibility to choose between web based or local development experiences while leveraging the same underlying infrastructure.
[00:47:41] Speaker D: This is currently available to you only.
[00:47:42] Speaker C: In U.S. east Ohio region, suggesting this.
[00:47:44] Speaker D: Is an early rollout that will expand.
[00:47:45] Speaker C: To other regions based on customer adoption.
[00:47:47] Speaker D: And feedback or if this actually finally solves those pesky data scientists deeds. As Ryan can attest, data scientists like their local tools.
[00:47:57] Speaker B: I mean it's, it's definitely kept me from adopting SageMaker. A larger thing being sort of forced into their their interface and their notebook interface because it's just it. Yeah, I, I do like it locally. It wasn't terrible. I could use it before but that's a lot easier if I don't have to do that. So I like that this pattern is becoming more prevalent where you're, you're keeping your context focused directly in that IDE and the IDE is going and reaching out to the different services.
Yeah, very cool.
Little the single region's a little strange because I Don't know if your SageMaker deployments in a different region. Can you just not use this? Which is kind of lame, but I.
[00:48:40] Speaker D: Assume that this will be coming out to all of them eventually. But some mvp.
It's just impressive to me how much of VS code is becoming the center.
[00:48:50] Speaker C: Of all development things.
[00:48:52] Speaker D: We just talked about these other tools like sorry, Kiro and we talked about.
[00:49:00] Speaker B: Kiro and the MCP Windser and all.
[00:49:03] Speaker D: These things, and they're all built on top of Visual Studio code.
So it's just, it's crazy how much people want to live inside that tool, which is great. I love VS code. I've been a big fan for a long time, but.
[00:49:14] Speaker B: Well, that's why I think it's just gonna, that's gonna be. It's gonna get even worse, right? Like you're, you're gonna see every interface go through that for, for all kinds of things. So it's, it'll be interesting to see how that melds with, you know, the rest of a production deployment in an application. But you know, like I'm finding myself using GitHub Actions directly via my VS Code extension and all these things, so it's can do a lot.
[00:49:39] Speaker C: Yeah, for sure.
[00:49:41] Speaker D: Let's move on to GCP Backup for.
[00:49:45] Speaker C: GKE is their new cross project backup.
[00:49:48] Speaker D: And restore solution in preview, allowing users.
[00:49:51] Speaker C: To backup workloads from one Google Cloud project, store them in a second project and restore to a third project.
[00:49:57] Speaker D: This addresses a key challenge on multi.
[00:49:58] Speaker C: Project GKE deployments for teams seeing centralized backup management across project boundaries. The feature enables critical disaster recovery capabilities by storing backups in separate projects and regions, protecting against regional outages or compromised primary projects. Organizations can meet RTO RPO objectives while simplifying regulatory compliance through proper backup isolation.
[00:50:17] Speaker D: Cross project functionality streamlines development workflows by.
[00:50:20] Speaker C: Enabling easy environment seeding and cloning, and.
[00:50:22] Speaker D: Teams can populate staging environments with production.
[00:50:24] Speaker C: Backup data or create isolated sandboxes without complex manual processes.
Developers can be granted delegated restore admin.
[00:50:31] Speaker D: Roles to restore specific backups without accessing live production environments. And this preview requires completing a form with documentation available for cross project backup.
[00:50:40] Speaker C: And restores with no specific pricing changes.
[00:50:43] Speaker D: Mentioned, which, if you have very complicated.
[00:50:46] Speaker C: GKE configurations, this can be very handy.
[00:50:49] Speaker B: Yeah, I mean, I haven't heard of GKE configurations being a huge target of ransomware, but this definitely seems like what this is targeting in terms of the delegated roles and being able to store the configs in a central project and then delegate them out to yet another project. Which is an interesting sort of scenario that I hadn't thought of, but that's.
[00:51:12] Speaker D: Pretty cool indeed, right?
[00:51:15] Speaker C: Introducing Cloud Storage Bucket Relocation the first feature among major cloud providers allows moving storage buckets to different regions without changing bucket names or disrupting your application.
This preserves all metadata, including storage classes, timestamps and permissions, while maintaining object lifecycle management rules.
[00:51:32] Speaker D: The feature uses asynchronous data copying to.
[00:51:34] Speaker C: Minimize downtime during migration, with only a brief write lock period during final synchronization.
[00:51:39] Speaker D: Organizations can perform dry runs to identify.
[00:51:41] Speaker C: Potential issues like CMEC incompatibilities before initiating the actual move.
[00:51:45] Speaker D: Key use cases include improving data locality.
[00:51:47] Speaker C: For performance, meeting regional compliance requirements, and optimizing costs by moving between storage tiers.
Spotify and Groupon reported successful migrations of petabytes of data with minimal manual effort compared to the traditional approach.
[00:51:58] Speaker D: Bucket relocation is part of Google Storage.
[00:52:00] Speaker C: Intelligence suite and supports moves between regional.
[00:52:02] Speaker D: Dual region and multi region configurations and.
[00:52:05] Speaker C: The three step process dry run, initiate relocation and finalize can be completed through simple GCloud commands.
[00:52:10] Speaker D: This addresses a significant pain point in.
[00:52:12] Speaker C: Cloud storage management where previously organizations had to use storage transfer service to copy data to new buckets with different names, requiring application updates and risking extended downtime.
[00:52:22] Speaker E: This is a really cool feature that would have saved me much time in the past life of hey, we set up this thing oh you know, years before we actually started using the cloud and it was for this one thing and now we've launched everything in this other region and every time we have to access this one specific bucket it is somewhere else and how do we fix that? And their process is actually pretty cool too where it sets it up, does kind of that sync and then sits at 99% and you kind of do that last one. There's a sounds like there's a few gotchas with like that final step which is like you should be at 99%, you'll get some HTTP 412s which I've never heard of that one before isn't in my memory of what the HTTP codes are. But this is a great feature, just quality of life, especially for customers that have moved or have decided hey, this one region, which from talking to you guys enough sounds like capacity planning is more of a thing. So if you're out karua region and you need a setup somewhere else, feels like that's kind of what this was built for.
[00:53:26] Speaker B: Yeah, no for sure. Like you know, certain machine types being only available in certain places, certain services.
So this is, you know, all kinds of reasons could get you into that situation where you're now trying to do Cross Region S3 Write your reads and so this. Or I. Sorry, Google Storage.
Right to read. So this is pretty great. Like I. It's really impressive that they're, you know, maintaining the bucket name and able to sort of do all that mapping on behalf of the customer. So that's great.
Yeah.
[00:53:58] Speaker D: I'd say definitely to see this come from other cloud providers in the future. It'd be hugely beneficial.
[00:54:05] Speaker B: Something tells me this would be very difficult for AWS.
The underlying infrastructure of S3 has got to be just not.
[00:54:13] Speaker D: I mean Amazon will just do it with a make sts do it in the background. You'll never know what's happening. That's just how they'll do it.
[00:54:20] Speaker C: But you're reusing the name is where Amazon.
[00:54:22] Speaker B: That's the tricky part.
[00:54:23] Speaker D: Yeah, yeah.
[00:54:24] Speaker E: But Azure is on the same problem. I assume Google would too. So they must have done something too with like a pointer or somewhere to kind of redirect under the hood.
[00:54:34] Speaker D: I assume. Yeah, I mean that would be what I would assume is done.
[00:54:40] Speaker E: Well.
[00:54:40] Speaker C: Cloud Run now supports direct deployment of Docker Compose files through the new GCloud run compose up command, eliminating manual infrastructure translation between local development and cloud deployments.
[00:54:51] Speaker D: This private preview feature automatically builds containers.
[00:54:53] Speaker C: From source and leverages cloud runs volume mounts for data persistence.
[00:54:56] Speaker D: The integration supports Docker's new model attributes.
[00:54:59] Speaker C: In the Compose specification, enabling developers to deploy AI applications with self hosted LLMs and MCP servers using a single computer configuration file.
[00:55:06] Speaker D: This position's Cloud Run as a cost.
[00:55:08] Speaker C: Effective option for AI workloads with pay per second billing and scale to zero capabilities.
[00:55:12] Speaker D: Cloud run GPUs which are now generally.
[00:55:14] Speaker C: Available combined with Compose support creates a streamlined path for AI development with approximately 19 second time to first token for.
[00:55:20] Speaker D: Models with Gemma 3.4B. This competes directly with AWS App Runner.
[00:55:25] Speaker C: And Azure Container apps with native GPU support.
[00:55:27] Speaker D: The collaboration addresses the growing complexity of a Gentek AI application by supporting Docker's MZP Gateway and Model Runner line.
[00:55:33] Speaker C: Developers maintain consistent configurations across local and.
[00:55:36] Speaker D: Cloud environments and you can sign up.
[00:55:37] Speaker C: For the preview at the form in the notes.
[00:55:40] Speaker D: This position GP strategically strategically in the.
[00:55:43] Speaker C: AI infrastructure market by adopting open standards while leveraging Cloud Run's existing strengths and serverless compute.
[00:55:48] Speaker D: I mean I don't care what the AI nonsense in this, but like man being able to run Docker Compose inside of Cloud Run.
[00:55:53] Speaker C: Amazing.
[00:55:54] Speaker D: Yeah.
Yeah.
[00:55:55] Speaker B: All of my projects and I have instructions now that I inform AI of like okay if, if I if I type this command or run this type of task, then do it this way and then if I for production do it this other way and this does away with all of that, which is fantastic.
I'm curious to see the rough edges on this, because you've been able to do sort of continuous integration delivery with cloud run for a while, but it had to be like a publicly available GitHub repo.
So this is.
I'm hoping that this is as transparent as it's made to be in the sense that I'm just running Docker compose, say in my ide and it's just making magic happen behind the scenes because that's fantastic. And I will shut up and take my money. I'll be signing up for this Google's.
[00:56:40] Speaker C: Announcing Seoul, a new transatlantic subsea cable.
[00:56:43] Speaker D: Connecting the US Palm Coast, Florida, Bermuda.
[00:56:46] Speaker C: The Azure Islands and Spain in the Santander region, marking the first operational fiber optic cable between Florida and Europe. This complements their existing nuvim cable to create redundant transatlantic paths with terrestrial interconnections at multiple points. The Cable strengthens Google Cloud's global infrastructure across 42 regions by providing increased capacity, improved reliability, and reduced latency for AI and cloud services between the Americas and Europe.
[00:57:08] Speaker D: Seoul features 16 fiber optic cable pairs.
[00:57:11] Speaker C: And will be manufactured in the US.
[00:57:13] Speaker D: Google is partnering with DC Blocks for.
[00:57:15] Speaker C: The Florida landing station, developing a terrestrial route to their South Carolina cloud region.
[00:57:19] Speaker D: While telsius provides infrastructure to Spain integrated with Madrid cloud regions, Florida and Spain.
[00:57:25] Speaker C: As new connectivity hubs for Google's network. Sol joins Google's growing subsea cable portfolio.
[00:57:30] Speaker D: Including nuvim, Fermina Equiano and Grace Hopper.
[00:57:33] Speaker C: Demonstrating their continued investment in owning network infrastructure rather than relying solely on consortium cables. This gives Google more control over capacity, routing and performance for their cloud customers.
[00:57:42] Speaker D: The cable addresses growing demand for transactivity.
[00:57:44] Speaker C: Driven by AI workloads and cloud adoption.
[00:57:47] Speaker E: The image annoys me a lot in this article. I can't get over that. I don't know how else to state that.
It very much bothers me.
[00:57:55] Speaker D: Yeah, yeah. I don't know who drew this diagram, but it is not the way the Earth looks.
[00:58:00] Speaker E: Not to scale.
[00:58:01] Speaker D: Yeah, I think that Waslands are not there. Bermuda is not there.
[00:58:04] Speaker E: It's Bermuda's the size of like the whole east coast.
[00:58:08] Speaker D: Yeah, it's massively. I mean, I was in Bermuda last year. It's not that big and it's not that far off the coast.
[00:58:14] Speaker B: My biggest gripe is that the land masses are in blue and the sea is in is gray. I was like, wait. It's very confusing. Yeah.
I'm pretty sure AI generated this. This dynamic for sure.
[00:58:27] Speaker D: This is one of those times where you should have paid the graphic artist. Yeah.
[00:58:33] Speaker C: What's up with the little waves in.
[00:58:34] Speaker D: The middle of the diagram? Like this, like totally tribal. It's like a tribal tattoo.
[00:58:39] Speaker B: Yeah.
[00:58:40] Speaker D: Of waves.
[00:58:41] Speaker B: So they're on color. So that's the only way you can tell it's water, man.
[00:58:45] Speaker D: Yeah. That's a bad diagram.
[00:58:46] Speaker C: I don't know what they're doing.
[00:58:47] Speaker E: Does say not in the actual blog post, but one of the other ones that I looked it UP begins Operation 2027. So that's not that far off.
[00:58:55] Speaker B: It's really not. That's kind of crazy.
I'm also surprised this is the first one in Florida. I would have figured Florida would be.
[00:59:02] Speaker C: Huge landing zone, but I guess it is sinking.
[00:59:04] Speaker D: I don't know.
[00:59:05] Speaker B: Yeah, that's true.
[00:59:07] Speaker E: Hold on. Where's my submariners count? No, there's a. There's a lot in Florida, but they mainly go down to the islands.
[00:59:17] Speaker B: Island or South America. Yeah.
[00:59:19] Speaker E: Yeah. South America is the island central. There's actually a lot to go to Central America.
[00:59:24] Speaker D: Yeah. It makes sense because from. I mean, due south of Florida. Isn't that crazy to get to South America.
[00:59:30] Speaker E: Yeah. So it looks like a lot of them are like Miami. There's only one in West Palm. Where's that one go?
That one goes to British Virgin Islands.
[00:59:40] Speaker C: Nice.
Apparently Google is finally gaining ground in cloud market share by focusing on data.
[00:59:46] Speaker D: Analytics and AI workloads, areas where they.
[00:59:49] Speaker C: Have technical advantage over AWS through services like BigQuery and Vertex AI. The company has shifted strategy from trying to match AWS feature for feature boo to emphasize their strengths in machine learning infrastructure and data processing capabilities that leverage their search and AI expertise.
[01:00:02] Speaker D: Google Cloud's growth rate now exceeds both.
[01:00:04] Speaker C: AWS and Azure, though a smaller base to start with.
[01:00:07] Speaker D: Product success in industries like retail and.
[01:00:09] Speaker C: Financial services that need advanced analytics. Really it's retail doesn't want to be on Amazon.
[01:00:13] Speaker D: Key differences include BigQuery serverless architecture that.
[01:00:16] Speaker C: Eliminates capacity planning and Vertex AI's integration with Google's pre trained models, making enterprise.
[01:00:20] Speaker D: AI adoption more accessible. Strategy appears to be working with notable.
[01:00:24] Speaker C: Customer wins including major retailers and banks who cite Google superior data analytics performance and lower total cost of ownership for specific workloads.
[01:00:30] Speaker D: Again, because they can't say out loud that they don't want to be helping their competitor.
[01:00:37] Speaker B: Yep, pretty much.
It is interesting because I will say that you know this is you know, focusing on Google strengths and I agree that they're you know, containers have been a strength for a long time and you start adding bigQuery and vertex AI you've got a pretty powerful platform to build off of the feature to feature is, you know, it's going to miss all those like enablements that make it really easy to stand up a full application on the cloud. So like it's kind of a bummer but we'll see what actually you know, is lacking.
[01:01:11] Speaker E: But if you're full blown containers inside of Kubernetes and leveraging all that, you're managing a lot of that at that level, not as much at the infrastructure level.
So you're not going to use a lot of those other features in those same ways.
So and you know they, they kind of say it here. You know Justin just said it's like there, this is targeted towards you know, their key differentiators. So obviously they're going to be better at their key different differentiators.
Otherwise they're probably not the key differentiators.
How many times do you think I.
[01:01:45] Speaker B: Could say key differentiators before we cut your mic? I don't know. Probably not very much.
[01:01:54] Speaker D: All right, moving on to Azure Microsoft.
[01:01:57] Speaker C: Is interesting Phi for mini Flash reasoning.
[01:02:00] Speaker D: Say that three times fast.
[01:02:01] Speaker C: A 3.8 billion parameter model using a new decoder hybrid decoder architecture called Sombay that combines Mamba state space models with sliding window attention and gated memory units to achieve 10x higher throughput and 2 to 3x latency reduction compared to standard transformer models.
[01:02:16] Speaker D: The model targets edge computing and resource constrained environments where compute memory and latency are critical factors making it deployable on.
[01:02:23] Speaker C: A single GPU while maintaining advanced math reasoning capabilities with 64K token context length.
[01:02:28] Speaker D: Key innovation is the gated memory unit.
[01:02:30] Speaker C: Mechanism that enables efficient layer representing sharing preserving linear pre filling time complexity while improving long context retrieval performance performance.
[01:02:37] Speaker D: Real time applications Primary use cases include on device reasoning assistance, adaptive learning models.
[01:02:42] Speaker C: And interactive tutoring systems that can require fast logic inference. The model available on Azure AI Foundry, Nvidia API Catalog and Hugging Face represents a practical approach deploying AI reasoning capabilities at the edge without cloud dependency addressing the growing need for low latency AI interference in mobile and IoT applications.
[01:03:01] Speaker B: I think about half of that I'm going to need Jonathan to explain to me but it is neat. I mean I do, I do think it's neat that they're innovating at that low level of a space with at the memory unit. So to, to offer these, you know, smaller models at the edge. I think it is a neat, neat sort of system and I look forward to, you know, when we can like dynamically route queries and prompts between, you know, smaller models and larger models more dynamically a little easier. And so this is pretty cool to. Cool to see and be fun to use.
[01:03:39] Speaker E: I think it'll be interesting when you're on your mobile device and you say, hey, run me this thing. It tries to run it on a model like this and then if it can't get you a good result because it's not enough, you know, data points and the parameters, then it kind of goes off. So that's kind of where I see this going, which is your edge based computing kind of coming back alive where your phone and your laptop, everything else has enough that it could run these small models, give you, you know, just quick feedback and do it offline also, versus everything always have to having to be online.
[01:04:14] Speaker B: Yeah, it's interesting because, you know, the prompt speed is a big area of interest and everyone's trying to sort of optimize on that and that failure mode when this can't process your request and it has to ship it off to the mothership like, oh, it's gonna be painfully slow.
[01:04:30] Speaker E: Yeah, but if you're on, you know, if you're not, if you're in a third world country that doesn't have the best Internet or you pay, you know, more per gigabyte than that, then these types of things could become, you know, better. In those types of environments where, you know, hey, let me run this locally. You know, we're used to 5G and other technologies that you're online and if it takes 3 seconds to load a page, I'm sure you're like me where I might.
Why haven't you launched my LT AZ EC2 or sorry, RDS cluster with 7 nodes right now I want to write the second, but other people are not as impatient as you and I, Ryan.
[01:05:08] Speaker B: I think you mean entitled.
[01:05:10] Speaker E: Yeah, that works too.
But if you can get into this not on the grid type of communications, so get those features, military or not. First world countries or even rural America where you don't have these massive Internet pipes, you do have these places where these things will become more useful.
[01:05:35] Speaker B: I'm going to go sip my champagne and spam the refresh key.
[01:05:41] Speaker D: Oracle, I have a story for you guys. You might be excited to know that.
[01:05:44] Speaker C: They are cutting cloud costs for the government.
[01:05:47] Speaker D: They're apparently partnering with the GSA to offer federal agencies a 75% discount for.
[01:05:53] Speaker C: Six months on licensed technologies plus migration.
[01:05:55] Speaker D: Services to Oracle Cloud, targeting a significant.
[01:05:58] Speaker C: Number of government systems still running older.
[01:06:00] Speaker D: Oracle versions on premise, Oracle claims their.
[01:06:02] Speaker C: Second generation cloud offers 50% lower compute costs, 70% lower storage costs and 80% lower networking costs compared to competitors, though.
[01:06:09] Speaker D: These comparisons lack specific benchmarks or competitor names. But it's Amazon that's what they always pick on. The partnership removes data egress fees when.
[01:06:15] Speaker C: Moving workloads between FedRamp and DOD IL4IL5.
[01:06:18] Speaker D: Certified clouds addressing a common pain point.
[01:06:21] Speaker C: For government agencies considering multi cloud strategies.
[01:06:24] Speaker D: Oracle is positioning their integrated AI capabilities into database 23 AI and application suites.
[01:06:28] Speaker C: As differentiators, though the announcement provides no.
[01:06:30] Speaker D: Technical details about actual AI features or.
[01:06:33] Speaker C: How they compare to aws, Azure or GCP offerings.
[01:06:36] Speaker D: While Oracle emphasizes cost savings and monitoring.
[01:06:38] Speaker C: Benefits, the real impact depends on how many federal agencies actually migrate from the.
[01:06:41] Speaker D: Legacy Oracle site systems, which have persisted precisely because Oracle doesn't force upgrades. There's also there's in the fine print on this one, the pricing discount doesn't last forever. It's about six months.
[01:06:54] Speaker B: Yeah. Details what it got you?
I I'm trying to figure out if this is like an opportunistic trying to take advantage of like new DOGE policies where it's like hey, look at all these savings you can get. Or if it's like somehow staying in front of it where they're they're worried DOGE is going to cut all the ERPs, I guess, or other stuff running on Oracle.
It's kind of interesting.
[01:07:20] Speaker D: All right, and then we have a.
[01:07:22] Speaker C: Cloud journey this week.
Chaos Engineering recommendations from Gartner, which is always exciting.
[01:07:28] Speaker D: Gartner's 2025 hype cycle for infrastructure platform.
[01:07:30] Speaker C: Highlights Chaos Engineering as essential for testing AI resilience, particularly for applications using generative AI API calls that need validated fallback patterns when services fail or experience latency.
[01:07:41] Speaker D: Game days are becoming critical for enterprise.
[01:07:43] Speaker C: Preparedness against catastrophic failures like crowdstrike or.
[01:07:45] Speaker D: Cloud provider outages, with financial institutions using.
[01:07:47] Speaker C: Them to verify disaster recovery plans.
[01:07:49] Speaker D: For operational resilience and compliance, organizations should.
[01:07:52] Speaker C: Prioritize chaos engineering on business critical systems.
[01:07:54] Speaker D: First focusing on payment services, single points of failure, and elevated security privilege components.
[01:07:58] Speaker C: Where downtime costs average $14,000 per minute.
Reliability scoring platforms provide measurable metrics beyond.
[01:08:05] Speaker D: Simple uptime downtime tracking, enabling teams to.
[01:08:08] Speaker C: Identify performance degradation and latency issues before they impact users.
[01:08:12] Speaker D: The increasing complexity of modern systems combined with AI adoption makes proactive reliability testing through Chaos Engineering a necessity rather than optional as AUC's cost global companies an.
[01:08:22] Speaker C: Average of $200 million annually.
[01:08:25] Speaker B: Yeah, no, it's kind of interesting and it's, it's funny to me that Chaos Engineering has made it all the way up to Gardner where they've got sort of, they're streamlined for it. It is kind of cool. But I mean, I do think that these kinds of exercises are key and I love the ability now with so much of the digital twin technology being able to do full end to end replication of these things and not just do tabletop exercises.
So this is, this is kind of neat and it's definitely something I want to see more of.
[01:09:01] Speaker E: It just depends on how much you trust your actual environment or most people won't want to actually let this run loose in their production environment.
But if you do trust, and you do trust that you have it, or if you do have a good automated QA framework where you let loose on your QA environment or your staging environment and just perpetually run your QA automation test to find those failures, it's a different, you know, a way to kind of leverage Chaos Engineering. But I do love Chaos Engineering because there's no better way to figure out what's going to break than, you know, go pull the plug and see what happens.
[01:09:39] Speaker D: Wait, have you any guys use these tools like Gremlin at all?
[01:09:44] Speaker B: I've played around. I've never been actually able to do it at a real scale just because I've never gotten enough trust built into things and you know, like it's never had a system that's really sophisticated enough to where you're trying to find additional failures. You can usually take a look at your architecture and go like, that's where it's going to break right there.
[01:10:05] Speaker E: Yeah, yeah. I'm in Ryan's, but I've kind of looked at it. I'm like, okay, cool, we know where it is, you know, or we know where our pain points are. We're already solving for those.
We don't need to add more things onto the list of places where it fails.
So, you know, I feel like the police, the places I've worked full time internally, I kind of know the places when I was a consultant, I've told people to do it and you know, and, or everyone's like, no, no, we're good, we know where our failures are. We got everything good.
It would be fun to go, you know, deploy this and see what happens. But people don't really, like, when you accidentally break production, they get a little bit upset. With you.
[01:10:45] Speaker B: Well, at $14,000 per minute, it does sort of make sense.
[01:10:49] Speaker E: I mean, even the AWS service that I've kind of played within, like my dev environment watched it take down a zone and then it, you know, the EC2 detects it and launches another one. So I've done it in those ways. But hopefully in my small little environments that run, you know, a webpage or two and a few other things, you know, it doesn't really care if a zone goes down or no one really cares if my website goes down because most of the time it's not even hit.
[01:11:14] Speaker B: So that's the difference.
[01:11:17] Speaker D: And that's another fantastic week here in the cloud, guys.
[01:11:20] Speaker B: We made it. Yay.
[01:11:22] Speaker D: We made it.
[01:11:23] Speaker B: Bye everybody.
[01:11:24] Speaker E: Bye everyone.
[01:11:28] Speaker A: 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.
[01:11:59] Speaker D: I do have an after show for us, which is, you know, everyone's talking about amazing AI tools and how they're helping developer productivity. And now we're on the other side.
[01:12:08] Speaker C: Of that pendulum with this article. Stop forcing AI tools on your engineers Engineering managers are facing pressure to force AI tool adoption on their teams, but mandating specific tools like Cursor or acquiring token usage metrics can backfire and slow productivity.
[01:12:22] Speaker D: Rather than improve it, companies should give engineers dedicated time 20% workload reduction or.
[01:12:26] Speaker C: Full expiration weeks to experiment with AI tools and their actual code bases.
[01:12:29] Speaker D: Rather than expecting zero cost adoption, the.
[01:12:32] Speaker C: Focus should shift from measuring AI tool usage to measuring actual outcomes of engineers. Engineers using AI tools deliver better results.
[01:12:39] Speaker D: Share those specific workflows internally rather than generic success stories. Monday.com's approach of a five week AI.
[01:12:45] Speaker C: Expiration with 127 internal demo submissions shows how large organizations can enable organic adoption through peer LED workshops and real use case sharing. And AI tools excel in greenfield projects and simple code bases, but adapting them to complex existing systems requires careful evaluation of what actually works versus following industry hype.
[01:13:01] Speaker D: Yeah, there's definitely a big forcing moment here where everyone's like, well, everyone needs AI tools and all that, but I think every engineer has to kind of figure out their use case and patterns.
[01:13:11] Speaker C: And how it adds value to them.
[01:13:12] Speaker D: And I can Briefly talk about some stuff we've done in the day job.
[01:13:17] Speaker C: Around this where we did cursor pocs.
[01:13:19] Speaker D: And we did GitHub copilot POCs and we've done Claude pocs and all that. And the feedback I use it, I'm like, this is amazing. It's going to be great. And then the feedback we get is kind of mixed where in my use case, the way I was using it, it actually wrote bad code or it didn't help me the way I thought it would. And so it is kind of a.
[01:13:40] Speaker C: Feast or famine.
[01:13:43] Speaker D: On some of those things, but it's definitely a challenge.
[01:13:47] Speaker B: I don't know, I don't think I agree.
I feel like it's another tool that we can use just like, and it's got metrics around it, you know, in terms of like adoption and, and AI. But you know, I, I also feel like this is one of those use cases where it's like, you know, how, how diva and princessy are developers where it's like trying to figure, you know, you have to give them 20% workload reduction so that they can play around with AI. No, you know, like you, everything's focused on business outcomes already.
And this is just another metric. Much like your commits and your contributions over time and your PR requests and things that we were already tracking, this is just yet another metric in that set to track productivity. And I think the reason for that is because there's a lot of productivity misses in the industry where there's people that aren't producing as much and they're using excuses to get around that, which isn't great.
[01:14:50] Speaker E: I think that it's the way I've seen people start to use it is more letting people, you know, different people play with it. But then, you know, the five week POC and then everyone talking about it is, you know, even talking with you guys or talking with my team and other people on my company is, you know, you hear somebody say, oh, I used to end this one. You're like, I never even thought about it because it's such a new tool out there that I think you have to talk about it in order to figure out what to do with it. You know, I played with it in some ways and I've said, hey, just write me a script that does this. And you know, when I was playing with Kiro earlier, it like completely over engineered what I actually need in real life, you know, because it's like 700 lines of a bash script that I don't even think can Actually get past line 30 when I try to run it.
But, you know, talking about it and, you know, talking with you guys, talking with other peers, talking with teams at work helps me find new ways to leverage it. And new things I think about, you know, even like, things like, hey, don't use, you know, Kira, but use a general AI bot to, you know, say, hey, I have a project to do, a migration of DNS. Break down that down into tasks. And so, and then I can create the Jira from it. You know, things that like, I just didn't think of, but talking with some. Somebody on my team and they were like, oh, I tried this, I was curious. It got like 80% of what we needed. So I think the communication around it is key. I think the only way people are going to learn how to use it is to play with it, but play with it in real world situations. Because the first thing I did was I opened up here. I was like, what do I want to do with it?
It's like, first, like, what are your requirements? I was like, I don't know what the project is. So I always have a problem designing without like a thing. Like, I need a thing to go play with.
[01:16:38] Speaker D: Yeah, I think that's why, that's why I chose earlier the, you know, I was thinking about it cool to have.
[01:16:43] Speaker C: A mobile app for the podcast.
[01:16:45] Speaker D: Not that I'm going to ship this thing, but maybe I will. But I was like, this is something I don't know a lot about and something that vibe coding can help me.
[01:16:52] Speaker C: Figure out details on. And so that was at work.
[01:16:54] Speaker D: But yeah, you got to kind of make up a project or scratch an itch that you have personally, like, you know, you know, a couple of people we know are building sprinkler systems, you know, on Raspberry PI. Like, okay, that's, that's their itch.
[01:17:06] Speaker C: That makes sense.
[01:17:07] Speaker D: But, you know, it really depends on what, what you're working on and what.
[01:17:10] Speaker C: Makes sense to you.
[01:17:12] Speaker B: And it doesn't have to be major. Like, you know, watching Justin go through the show notes AI bots, like, that was inspirational for me because I kind of didn't. I didn't have a project like that and I'd play it around. I had dabbled, but it wasn't until I saw that did I, you know, did I start writing sort of instructions and sort of doing a little bit more than just sort of expecting AI to write, you know, 80% of the boilerplate code. And then I was going to have to go through and finish it off. Like this is much more collaborative back and forth. I wouldn't say I'm vibe coding exactly because I am sort of pointing it too much in a specific direction so that it doesn't get too lost. But, but I do think that it's, you know, a powerful tool that I think if you, it's one of those things, if you don't adopt the technology as it's introduced, you're going to get left behind.
And if engineering managers are having to force their teams to do that, I think that that's telling and I think that's going to be. I think the problem is somewhere else. And I don't. I think blaming it on AI is probably misplaced.
[01:18:18] Speaker E: Yeah, I mean, some people just aren't gonna want to. And you have to kind of accept that, you know, but the people that want to are gonna be the ones that play with it. And you're just gonna see productivity, massive productivity enhancements with that. Like, you're gonna see it just because you're gonna get those. And if people don't wanna do it, then look, they're not gonna do it, which is fine. But at one point they're going to fall behind or they're going to figure out how to make themselves in their current process more efficient and. Or just they're going to write better code than any AI bot and they're going to say, look, I'm just better. And look, if they're that good, they're that good.
[01:18:59] Speaker B: If they're that good, I mean, that's, it's exactly, it's going to be, I'm not that good.
[01:19:03] Speaker E: Yeah, I'm a mediocre developer at best. Right. See, my code, Justin has probably pretended to look at it once or twice.
[01:19:10] Speaker C: I've seen it.
[01:19:11] Speaker D: It's okay.
[01:19:14] Speaker E: I'm okay with, okay, it's a c. I'm okay with that.
[01:19:17] Speaker B: It's just going to be another tool like an ide.
[01:19:19] Speaker D: Right.
[01:19:20] Speaker B: We used to refactor by changing function names by search and replace and going and changing one by one as you get more sophisticated use ide, you're changing all of them at once and refactoring on the fly. And now you're just making the bot go do it for you.
Things change over time and that's.
Finding those productivity enhancements is always going to be part of the job because you're, you're never going to be given enough time to complete projects ever.
[01:19:48] Speaker D: Well, we're continuing to keep an eye on this pace space around this and you know, there's definitely been. Everyone's firing all their engineers a couple weeks ago to. For AI. Then this week I saw like, no.
[01:19:58] Speaker C: You shouldn't do that.
[01:19:59] Speaker D: So, you know, it's the ebb and.
[01:20:01] Speaker C: Flow of the press in some ways.
[01:20:02] Speaker D: In some of these conversations.
But we'll keep an eye on it.
[01:20:05] Speaker C: And keep you posted here at the cloud pod.
[01:20:07] Speaker D: So see you guys next week, right?
[01:20:09] Speaker E: See ya.