September Tech News - Billion-Dollar Acquisitions, AI Shakeups, and Kubernetes Updates

[Stefan] (0:03 - 0:10) People actually like the human touch and it's easy to see in a chat when it's not a human. Sorry to say AI, but you're not there yet.

[Pinja] (0:14 - 0:23) Welcome to the DevOps Sauna, the podcast where we deep dive into the world of DevOps, platform engineering, security, and more as we explore the future of development.

[Stefan] (0:23 - 0:32) Join us as we dive into the heart of DevOps, one story at a time. Whether you're a seasoned practitioner or only starting your DevOps journey, we're happy to welcome you into the DevOps Sauna.

[Pinja] (0:37 - 0:45) Hello, and welcome back to the DevOps Sauna, where I'm joined with my colleague, Stefan. Hi, Stefan. How are you doing?

[Stefan] (0:45 - 0:49) Hi, Pinja. Actually doing well. We're closing in on Future of Software.

[Pinja] (0:49 - 0:50) Ooh, we are.

[Stefan] (0:50 - 0:53) That's going to be great, which means a busy week for both of us.

[Pinja] (0:53 - 1:44) It is coming. So this is an unofficial pitch. We're coming to Stockholm and Copenhagen at the end of October, the 21st and the 23rd of October.

Future of Software program, we have a lot of good stuff coming up, so check those out. We still have some tickets left. But in addition to this promo here, we wanted to talk to you about the news.

There's some interesting stuff that has happened during September, which of course is interesting from our point of view, but we also think it might be interesting from everybody else's point of view, obviously. So let's talk about a couple of acquisitions in the beginning now. So Atlassian bought the company DX, and this is a big one.

So Atlassian has had their own developer portal called Compass, and now they acquired DX, a developer productivity platform for $1 billion. Why do we think this is big, Stefan?

[Stefan] (1:44 - 2:49) It's fairly crazy to see somebody just pull $1 billion out of your pocket and buy another company. But I think it makes really good sense because we've seen Compass being implemented with customers, I've checked it out. But it's been this, we can see what you want to do here.

It's sort of been voiced as a manager's view and things. And when they buy DX, they surely want to really elevate the levels of insights they can gain. Because DX is well renowned, they're super good.

They do this, I think it was called DX4 that was announced and was really well received. Unfortunately, you can't see it unless you buy it. But of course, that's the product from their side, but it would be lovely to gain insights.

But the presentations and everything I saw where you sort of got a glimpse into it, looked super good, because it focuses on the actual things and not just metrics on top. And sometimes we do fall into the trap of just collecting metrics for metrics sake, even, which makes it super hard. And their focus has always been like, collect the metrics and look into them.

Don't just collect the metrics. So I think it's a super good acquisition for Atlassian, even though it's quite a lot of money.

[Pinja] (2:49 - 3:15) It is quite a lot of money. And DX itself, they basically came, I don't want to say out of nowhere, but they were a small player and they came out, was it 2022? And since then, they have tripled their customer base every single year.

So again, making a lot of sense here. And was it so that 90% of DX's customers were already using Atlassian's project management collaboration tools? So in that sense, again, a lot of sense in this purchase.

[Stefan] (3:16 - 3:23) Yeah, they've met each other in the market so many times. And I guess it's been on the radar for Atlassian for quite some time to boost their productivity space.

[Pinja] (3:23 - 4:04) And then another acquisition also made by Atlassian. So they acquired the browser company, which makes the ARC and DIA browsers. The deal size was $610 million in cash.

And when we talk about ARC and DIA, they are both AI powered browsers. So this is a bit of a question mark. When Stefan and I were doing the prep for this episode, we were kind of wondering, what does this game change mean for Atlassian to buy the Browser Company of New York?

In the press release, they said that they would like developers to have a very single flow, including the browser itself, though it would all be in Atlassian's environment. So this is a very interesting strategy change for them.

[Stefan] (4:04 - 4:40) But it's super interesting, because honestly, I've never heard about ARC and DIA before I saw the acquisition note. But how do you get people to switch their favorite browser? I see the idea here, and they want to super optimize the browser for all of the applications and so on, because like they state, the browsers we have today, they're built for generic web browsing.

They're not built for all of these applications. But would I switch my day-to-day browser? Would I have an extra browser running?

Most likely not. I always have one browser open, and if something fails in it, I'll open another browser. It's not like I'm running two or three different browsers because my machine would die from the memory load.

[Pinja] (4:41 - 5:00) I know that the idea, what Atlassian is after here might be that you would have all the support that the SaaS apps that they provide are also supported by the browser. So it's kind of built in. But I want to say, let's see.

I want to keep an open mind. But it was not in my bingo card for this year, to be honest.

[Stefan] (5:01 - 5:32) No, no. It's the big old discussion when people started talking about progressive web apps, like you should build everything as a progressive web app instead of an actual desktop application. A lot of people went that way.

Some people went back again. If you use Slack on a daily basis, I have the desktop app installed instead of having it inside of my browser because it's easier for me to use the desktop app because it's confined to that space. If I need to kill my browser, it crashes and burns with my 5,000 tabs open or whatever.

It doesn't affect Slack. I can still get in contact and reach out and say, I did something stupid. Please help me.

[Pinja] (5:33 - 5:41) Yeah. But as we're trying to keep an open mind, again, the acquisition deal was quite sizable with $610 million. But yeah, we shall see.

[Stefan] (5:42 - 5:49) It's just like an average day where you throw like $600 million here, $1 billion there. It's fine.

[Pinja] (5:49 - 6:10) It's fine. But let's talk about something we have discussed in this podcast before. So we covered the supply chain attack on NPM packages back in September.

And the attack story continued after we recorded the episode. So it was still ongoing. But what we thought was interesting was how GitHub has now taken action.

[Stefan] (6:11 - 7:09) Yeah. It's just going to be yet another week, yet another supply chain attack somewhere. But GitHub, they've seen the world burn now.

And now they're actually like, all right, we need to do something about it. Let's increase security. So they're killing off some legacy tokens.

They're switching you away from a one-time password. So you need to either have a passkey in your password manager or have a physical key to it. And the tokens you actually do create, you will actually have a shorter expiration on them to make sure they can only be used for this certain amount of time.

And they're removing some bypass options as well. So it's going to be more and more tricky to actually get these vulnerabilities in. It will still happen, but at least it's going to be harder to do, which is always this balance like, do you want to make it harder?

If you make it harder, are you going to get like the more complex, trickier adversaries? Or if it's too easy and open, they will most likely not go in. It's like, where are we going to end with all of this?

Welcome to the balance of the security world.

[Pinja] (7:09 - 7:49) Yeah. And we're talking about open source as well at the same time and how to not discourage or kind of take the whole base away from what is open source. Because that is the big base of how we build software nowadays.

So that's what I was wondering about. So I hope that we can come up with a solution because now, of course, there were other immediate actions like removing the compromised packages. But also, if we need to be extremely careful, how do we ensure that we are able to now still upload open source packages and open source software and code without actually getting these vulnerabilities?

As I say, it's a very, very tricky trade that you need to play with the security.

[Stefan] (7:50 - 8:54) And the open source community is looking into this as well because some projects get created, the maintainer is sort of running away from the project at some point because he doesn't use that technology or something like that anymore. And how do you hand that project over to the next people? It happened with an external secrets operator in Kubernetes world, where the maintainers got so burned out that they just couldn't maintain the project anymore.

So they sort of shut down the hatches, saying, we need more maintainers, else the project will die, which everybody would be super unhappy about. But then they actually took the decision to like, let's go into a minimal maintenance stage. Let's focus on getting it into a stable state where we can onboard new people.

We'll onboard some new people at our pace, make sure we trust them and so on. So like this is a central call that will touch all of your passwords and so on. So you need to be extra careful with the new people you're bringing.

So it's both like GitHub is trying to do better for NPM and other things at the moment, but the open source side is actually trying to improve as well to make it a more stable and nice world. So Pinja, what else do we have today? Something about AI?

[Pinja] (8:54 - 8:55) Obviously we have something.

[Stefan] (8:56 - 8:59) Like the world is AI all over the place, so we have to have something.

[Pinja] (8:59 - 9:00) Have to, we must.

[Stefan] (9:01 - 9:04) What about xAI? Should we still have some of the employees or?

[Pinja] (9:04 - 9:25) One of my favorite companies, but let's not go into politics too much, is xAI, so their product called Grok, xAI fired 500 AI developers, which is one third of their AI developers to be precise. And they are now moving to, from this generalist AI to more specific tasks on AI.

[Stefan] (9:26 - 9:58) Yeah. It's funny to see how AI changes at the moment because some of the big corporations are sort of slowing down. Some of them are speeding up.

Some are going into completely new business sectors. I think it was OpenAI that announced a social network this week with everything that was AI generated. So there will be no human content.

But when we look at xAI cutting 500 AI developers, I think we sort of see the industry changing a bit. Like there's been a rush to be the first front runners with AI all over the place. And now we sort of see who's going to be the successful ones.

[Pinja] (9:58 - 10:38) Yeah. This was the 1,500 people out of which one third was now let go. This was their data annotation team basically.

And they work to label and prepare the data that they use to train xAI's chatbot Grok. And they, according to the news, there is an internal email going out that says they announced an immediate strategic pivot, which also then was explained by the company now deciding to accelerate the expansion and prioritization. And again, the direction is the specialist AI tutors versus scaling back on the focus on the general AI tutor roles.

That was internal communication that went out.

[Stefan] (10:38 - 10:43) Yeah. And then you take that sentence and put it into AI and say, please make this human talk.

[Pinja] (10:43 - 10:45) Yeah. Make it sound better.

[Stefan] (10:45 - 11:19) It was like, so you're sort of like trying to expand and prioritize your specialist AI tutors, but you're removing or scaling back on your general AI tutors. Well, you need to tutor your AI to some degree, like labeling and tagging your data so you can actually train it well, but scaling back on that. So how are you going to solve this in the future?

Are you going to use AI to tag and label all of this data, or what are you going to do? There are so many bad histories here. The worst one, Microsoft threw their first chatbot on the internet.

It turned absolutely horrible within 24 hours, so they had to kill it off again. This is going to be interesting.

[Pinja] (11:19 - 11:28) Yeah. Was it Klarna who fired a bunch of people a year ago because they wanted to replace them in these roles with AI, and now they had to hire them back?

[Stefan] (11:28 - 11:30) Yeah. It was the support desk.

[Pinja] (11:30 - 11:31) Yeah, exactly.

[Stefan] (11:31 - 11:44) So they removed all of the support engineers, and a year later, it was like, oh, this didn't really work well because people actually like the human touch, and it's easy to see in a chat when it's not a human. Sorry to say, AI, but you're not there yet.

[Pinja] (11:44 - 12:25) Yeah. Speaking of which, another AI article that we saw was the study. There was research from Apollo Academy where they saw that the AI adoption rate in biggest corporations in the U.S. has now gone down from 13% to 12%. Of course, this is not a big drop, but it's kind of like a trend that everybody's now looking into, and it is in decline. They were asked, what is the adoption rate? That was one of the first questions Stefan and I had.

The measurement was that they were asked whether they use artificial intelligence in producing goods or services or in their business processes.

[Stefan] (12:26 - 12:28) Which is the open question, like, do you use AI, more or less?

[Pinja] (12:29 - 12:29) Yes, exactly.

[Stefan] (12:30 - 13:30) You can apply that to anything in your corporation, which is actually okay, but I think, of course, we'll see a drop to some point because people have bought into everything AI because everybody needed to have it, everybody wanted to be in that space. All right, so what's going on? How much are we utilizing it?

Maybe people have bought it, figured out it costs them a ton of money, and they started adding some measurements around it to figure out if it's actually adopted. Because I've been one of the slower adopters, and if people looked at my stats from the beginning, it's like, oh, dear God, he's not using AI at all. Let's cut the license costs from him and give him like a whiteboard and a pen back, or even worse, a pen and a paper.

But slowly ramping up, getting there, using it every day now. You'll see sometimes it just takes time, but sometimes you also need to look at your cost and figure out, is it utilized, and do you actually need to scrub it? Sometimes people had like three or four options for code generation, which was like, you don't need three or four options, you need one that is really good, that fits your needs.

[Pinja] (13:31 - 13:49) Yeah, we've been talking a lot about what is an AI strategy of a company? How do you use the tools? Do you feel that you know what you're actually using the AI tools for?

As you say, it is more important to focus on those. Do I need the three or four tools, or do I need to know my chosen tool?

[Stefan] (13:50 - 14:34) Sometimes we might see a drop in AI adoption, but it might not be the real story because we see a lot of people going into the agent space these days. If you start building agents and remove licenses from a lot of people, is that a drop or are you just using AI smarter in the end? The future will be run by AI agents that will remove a lot of crud from a lot of people, but we need to go into this balance of like, when do we apply AI, where does it fit?

Can it just remove all of the boring stuff from us? I'm from a site reliability background, but if I could have AI taking care of 90% of the alerts and issues that came up, I would be so happy because some of them were like, yeah, it's still an alert, it's still the same thing. All right, we recognize the pattern, just skip it.

We'll come back to you if it stays here.

[Pinja] (14:35 - 14:55) We might not want to jump into a specific conclusion from, well, hey, now it dropped from 13% to 12%, which is, again, percentage-wise, not a very big one. We're talking about big corporations, sure, but as I said, this is a number that is being quite heavily monitored at the moment and people are very interested in it because of the AI hype.

[Stefan] (14:56 - 15:53) People are still contacting us to get some sort of AI. We need to go into the same talks that we did with DevOps. I would like an AI.

No, you would like an AI strategy, how to apply it. We used to have people calling like, I want a DevOps. Well, you can’t have a DevOps.

You can have DevOps practices and we'll help you with that. We're going to see the same thing with AI here. Speaking of AI that is declining, it's definitely not declining for OpenAI.

They got a ton of backing from NVIDIA this month. They plan to deploy 10 gigawatts of NVIDIA systems by the second half of 2026 and it's an investment of $100 billion. Back to the Atlassian acquisition for a billion dollars, who cares?

This is $100 billion that NVIDIA is taking out of their pocket. Some of it is going to be in hardware and so on, but still getting an investment that is up to $100 billion into pure AI, that is a lot of money.

[Pinja] (15:53 - 16:25) Yeah. We're talking about supporting this deployment, including data centers and power capacity. Having the first gigawatt of NVIDIA systems to be deployed in the second half of next year, it might sound that it's one year away, but we're still talking about a huge, huge investment in this.

NVIDIA and OpenAI have been collaborating for the better part of a decade right now. This is not news that they're collaborating, but this is something that is right now.

[Stefan] (16:26 - 16:55) Yeah. If you look at the portfolio that invests into OpenAI, you have Microsoft investing heavily in NVIDIA. There's a list of 10 or 15 that have invested quite a lot of money.

I think everybody's betting on OpenAI. Will they make mistakes? For sure they will.

Everybody disliked the last model they came out with from day one. They've tweaked and tuned it. Now it's better.

It's going to be there for a while for sure. Well, now they have a social network as well, so that's going to be great. Everybody needs a social network.

[Pinja] (16:55 - 17:07) It does. When we're talking about OpenAI, it's got over 700 million weekly active users, and it might be one of the most adopted gen AI tools at the moment.

[Stefan] (17:07 - 17:07) For sure.

[Pinja] (17:08 - 17:18) Using one, let's say NVIDIA as a producer of chips, for example, we are a manufacturer, so we really are interested to see what happens out of this cooperation.

[Stefan] (17:19 - 18:08) Yeah. They own the whole prompt space at the moment. Everybody's like, oh, are you using ChatGPT?

I'm like, do you mean if I use AI? Do you mean if I actually use ChatGPT? Because people are defining ChatGPT as the AI these days.

That's going to be so many interesting things happening. I don't think we've seen the top yet. When we start having agents, building agents and so on, and models training to some insane level on every data in the world or something like that, I'm still like, have we yet figured out what data we're training on here?

Are we using Stack Overflow? Are we using Wikipedia? What are we using?

And on top of that, are we bringing in white papers from universities and everything? So much data, but unfortunately enough, many people are so open on the training data. Whatever they do, it seems to work and there's definitely going to be a need for AI for many, many years.

[Pinja] (18:09 - 18:51) Yeah, I agree. One more thing on AI. GitHub Copilot.

I'm going to preface this by talking about technical debt because this is about technical debt and GitHub Copilot's agents, but it is not the favourite term that business or technical people hear. Like, can you please fix our technical debt? Can we please use our funds to fix technical debt?

But it has been proven that it is actually a huge threat for innovation. Now, GitHub Copilot has updated their service with some specialised technical data agents and they aim this to be something that will actually help developers to tackle the technical debt, especially in Java and .NET. Yeah, like the initial releases is specialised

[Stefan] (18:51 - 20:40) for Java and .NET, so we're probably going to see it roam those spaces first, but I can imagine like GitHub Copilot, you have this huge code base on the internet to scavenge into. Some of it might be good, some of it might be bad.

So there's a lot of training data for it, for sure. But getting that tech dev agent in to help you on a day-to-day basis, I would have loved that for many years back. Like when I open some of my dusty old spare time projects, it's always like Visual Studio Code is screaming at me, like, yeah, you could optimize this, you can optimize that.

What if I could let the AI agent just make sure that I'm up to speed with, let's say, the latest language level of C# that is running on .NET? Like, that would be great. Even though I'm sort of a dinosaur in this industry, I've been working with .NET for 20 years, but I still try to keep up to date with the good practices and so on. So if it can do like the best practices and not jump into all of the newest, greatest features that sometimes make your code less readable, that's going to be good. So really, really high hopes for this one. But let's see how it works, because it can either go super good, or it can be like, yeah, it's okay.

Like we still need to review all of our code and rewrite it. But that's a day-to-day job for many people. But tech dev is a big issue, because when you stick with old frameworks or old structures, it will be costly for you.

I think it was .NET when they updated their core engine from one version to the next, you could actually see the CPU load drop by 25 to 50 percent. Imagine scaling that up to like a thousand machines running in production. Well, now you only have to pay for, let's say, maybe 700, 600, even down to 500.

That's a big, big savings you can do there. So I would imagine there's going to be some pricing lingering somewhere under the covers. We saw the shift of Copilot being for free, and all of a sudden, when you choose different models, it's going to be a bit more pricey.

That's true. Well, if the model is good, it's worth the money.

[Pinja] (20:40 - 21:11) Yeah, because now the Microsoft reports are saying that the organizations that are using these tools have already reduced modernization time by 60 to 70 percent compared to the manual processes. But there are, of course, some limitations. They cannot solve every modernization challenge, obviously.

It takes some iteration to get a little bit further, so it might be some complex business logic, some custom integrations there in the background. But reducing the time being used for modernization by 60 to 70 percent sounds quite a lot to me, at least.

[Stefan] (21:12 - 21:57) That's quite good. I used to run a developer team of eight people, and I had one guy allocated to fixing tech debt, because it was an old platform, and we needed to look into the future and how it should be. But I think for a year and a half, he was only doing tech debt, which he actually liked, but not every person is like that.

I would get so bored by doing tech debt every day, like, oh, come on, can I please do something that adds value to this? I know it's adding long-term value, but on a day-to-day basis, like, all right, back to the mines again. You're just like scraping. It's an interesting new world where we might be able to get some assistance for AI. Speaking of AI again, so I guess you're the expert on Kubernetes in this podcast, or am I wrong here?

[Pinja] (21:57 - 21:58) Yeah, obviously.

[Stefan] (21:58 - 22:02) All right, so you know everything about dynamic resource allocation.

[Pinja] (22:02 - 22:11) I guess not, no, but we do know the very, very basics of it. So the DRA, so the Dynamics Resource Allocation, is now generally available on Kubernetes.

[Stefan] (22:12 - 22:13) Yes, it is.

[Pinja] (22:13 - 22:19) And when we think about the AI workloads, I do understand the importance that this brings.

[Stefan] (22:20 - 22:22) See, you're becoming a Kubernetes expert all of a sudden.

[Pinja] (22:22 - 22:25) I know, exactly. Who would have thought?

[Stefan] (22:25 - 22:26) So you're taking over my position now.

[Pinja] (22:26 - 22:28) We wanted to save this bit for last.

[Stefan] (22:28 - 23:21) Yeah, but the nice bit is we can actually buy these insanely expensive GPUs or rent them in a cloud environment, and now we can actually spread the utilization of those GPUs. I think it was last year or this year I heard NVIDIA coming with the pitch, like, overall utilizations of the GPUs was 5% worldwide. So if we can squeeze that down and let's just get 50% to 80% utilization of the GPUs we have, then it's going to be good.

So with Dynamic Resource Allocation, we have a structured model for Kubernetes, how we can do that. There have been options with some extra installs into your clusters and so on, but now we have something that is standardized. It's good.

Everybody can understand it. There's been a ton of talks about it at the previous KubeCon conferences. So it's really nice.

You can get a small slice of your GPU instead of having to buy a full one.

[Pinja] (23:21 - 23:45) This was one thing that Kubernetes was listing. What is happening now? What is happening next?

So the core of the DRA is now generally available. There are a couple of things that are currently being promoted to beta, and there are some new alpha features. So those would be possibly part of the next release.

So there's admin access labeling and process lists. Those are the next things that Kubernetes is planning to add to the next thing.

[Stefan] (23:45 - 24:18) There's always this beta phase, and it will sooner or later go into general availability. In some cases, they will start tagging it deprecated, so they might remove some of the old things that you used to work for your GPUs, and it's going to be disappearing in four or five versions of Kubernetes down the line. They're actually quite good at cleaning up old stuff again.

So getting a simpler space where you don't have several ways to go where, like, should I use this one or that one? This one is new. This one is old.

Which one is the correct one to use? So you can sort of clean up the internet a bit again, make it easier for AI to train on.

[Pinja] (24:19 - 24:32) That's true. So we wanted to say, this was our last article, we wanted to save the Kubernetes for last. We know how much technical folks love Kubernetes, and I think Stefan's enthusiasm also proved the point with this statement.

[Stefan] (24:33 - 24:42) I've been in the fin-off space. I've seen how much money you can burn on your infrastructure. So being able to run smaller machines or less machines, I'm happy.

I'm all in for that.

[Pinja] (24:42 - 24:49) Good. Hey, I think that's all the time we have for news today. It was a very interesting round to go through what happened in September.

Stefan, thank you so much for joining me.

[Stefan] (24:49 - 24:49) Thank you.

[Pinja] (24:50 - 24:59) And thank you, everybody, for tuning in. We hope to see you next time. We'll now tell you a little bit about who we are.

[Stefan] (24:59 - 25:04) I'm Stefan Poulsen. I work as a solution architect with focus on DevOps, platform engineering, and AI.

[Pinja] (25:05 - 25:09) I'm Pinja Kujala. I specialize in agilel and portfolio management topics at Eficode.

[Stefan] (25:09 - 25:12) Thanks for tuning in. We'll catch you next time.

[Pinja] (25:12 - 25:20)