Digesting Dell Technologies World AI automation news

Transcript - Digesting Dell Technologies World AI automation news

Beth Pariseau: From Informa TechTarget, I'm Beth Pariseau, and this is IT Ops Query.

This podcast distills the signal from the noise about enterprise software development and platform engineering. Each week, we'll talk to expert guests about the latest tech industry news and trends that engineering and IT leaders need to know.

Don't forget to subscribe to IT Ops Query for more conversations on AI and the future of the enterprise digital workspace.

My guest today, Steven Dickens, has more than 25 years' experience in the tech industry as a business manager for CA, HPE, IBM, and an industry analyst with Futurum Group, and now his own company, HyperFrame Research.

Hi, Steven. Thanks for joining us.

Steven Dickens: Hey, Beth. You say all the right things -- 25 years makes me sound old. I started my career at age 10, for anybody who's wondering.

Pariseau: Same, same. I have 20, so, you know. Tell us where you are.

Dickens: So, behind me is Dell Tech World. Scrambling to try and find somewhere the audio wasn't crazy, so yeah, out at Dell Tech World this week in Vegas.

Pariseau: In fabulous Las Vegas. So, what has been the most interesting topic of discussions down there?

Dickens: Little thing that nobody's talking about. AI, you may have heard of it.

Pariseau: Right, something different.

Dickens: Yeah, just something to keep us all fresh and on our toes.

No, I mean, all joking aside, Beth, unsurprisingly, Dell are positioning themselves to be that kind of infrastructure provider for AI.

I obviously focus in the data center space, so kind of that's what I've been tuned into most. But also they're positioning, you know, very, very competent PCs -- almost what would have been servers five years ago, maybe even less than that -- for, sort of, they had the main guy behind the OpenClaw Foundation on stage today.

So, being able to position, sort of, I think local AI systems were at an inflection point of 'Do you bring the AI to the data, or the data to the AI?' Obviously, the client providers and the neocloud providers would be saying, 'Bring the AI to those public platforms.'

I think where Dell's sitting -- and they have Thomas Kurian, the CEO of Google, also joining the keynote today -- so it's this combination of they're a provider into the hyperscalers for systems, but they're also wanting to drive a lot of that AI workload to sort of be on-premises, either from a data perspective or just sovereignty in general.

Pariseau: Right. And, of course, they also launched the Dell Automation Platform for AI-driven IT ops and custom orchestration. So, everyone and their brother has an AI factory now and an agent orchestration platform. So, how can enterprise IT buyers make a choice? I mean, do they just go with whoever they already work the most with or what?

Dickens: Beth, you always ask the best questions. And that's just not me giving you flowers. I asked exactly the same question of Michael Dell in the -- so great minds think alike or fools often differ, whatever it is.

But no, I mean, all joking aside, we hear the AI factory story from Lenovo, we hear it from HPE, we hear it from all the players in the stack. And so I asked Michael Dell and his leadership team the exact same question in the closed-door. So I think, whilst the message is good from Dell, I think the differentiation of that message is the question. So, you're absolutely asking the right question.

I think what Michael came back with, which I agree with, his comment was, 'It helps being four times bigger than your next biggest competitor,' from a supply chain perspective.

Pariseau: OK, fair.

Dickens: So, I'm buying that. That makes sense. We are in a sort of supply chain constraint scenario, whether that's memory, whether that's GPU, you know, whatever it is, so being able to get access to that equipment.

I also see -- we talked about desktop and server. Being able to have those combined means they can go and sort of petition the downstream component suppliers. 'We've got bigger infrastructure needs, so … to the front of the queue.'

So, pretty much, only Lenovo is in that sort of devices and infrastructure space.

Pariseau: Yeah.

Dickens: So, then it's a supply chain and demand sort of management piece.

So, you know, Dell's always been really good at supply chain. You know, that's been their defining thing for the last sort of 30, 40 years. To have been able to do that at a time when it's most needed is not something new for them, it's not a new muscle. Obviously, other providers have had big supply chains. But, you know it's been a differentiating factor for Dell, I think.

And then just the speed of innovation. Their CTO was talking about just the speed of innovation. Now, I buy that by the feature and product velocity. So, I think, you know, Michael and his team did a good job, I think, of talking about how they're differentiated versus others.

Solution itself, and unpicking it and pulling it apart, is a rack with some storage and Nvidia GPUs. Could it be largely that different from anybody else? It's rack with some storage and some GPUs and some compute in it. That's a harder one to sell. But I think if you want it in a period of time that makes sense for your business, I'd probably back Dell to get it to you on time.

Pariseau: There is that reality.

Dickens: And that's the reality we're in, in the market at the moment. Will it be supported? Would it be, sort of, a trusted configuration and all the other things we need it to be? Yeah. But I think that's largely table stakes, whether I'm buying that from HPE or whether I'm buying that from Lenovo.

Pariseau: Right. I mean, and this is a little bit 'inside baseball,' but it was noted to me that Dell also added VMware Cloud Foundation 9.1 support for its private cloud platform this week, and there's a lot of water under the bridge with those two companies. So, any significance to that, do you think?

Dickens: Yeah, I mean, they've got some shared history, you know? A 'situationship' that was there for a while, to use my daughter's sort of vernacular.

But no, I mean, I think there's a pragmatism that's starting to emerge. We've heard a lot over the last sort of two-and-a-bit years, however long it is since Broadcom announced their intention that sort of closed on that VMware deal.

I think -- I've had this conversation with Lenovo and, you know, to a lesser extent with HPE -- but I think the announcement you saw from Dell, there's still a lot of customers that are gonna run VCF.

Cutting your nose off to spite your face and say, 'We're only gonna work with Nutanix,' or 'We're only gonna work with Red Hat or HPE' -- sort of saying that they've got their own in-house platform -- makes sense and absolutely they should.

But there's gonna be a lot of customers that are gonna have VCF and continue to have VCF, even after we wind the clock forward three years. I think having that support makes a lot of sense.

Would they go on the main stage and make that announcement, you know, coming out of Michael Dell with all the cameras rolling? Probably not, which is probably why it's not got out ahead of the show.

Pariseau: Right. And, you know, blurring the lines even further, right, there used to be a delineation -- and I guess for Broadcom, there still is. There used to be a delineation between public and private cloud for this sort of thing.

But now, lines are blurring all over the place with things like Google Distributed Cloud with Dell, and they're supporting Azure Local, and all the public cloud providers are trying to play nice with everybody else. So, now you have this across-the-board, multi-cloud hybrid cloud story coming from everyone. Does that make evaluation more difficult? Is it just a matter of, like, you know, who do you work with?

Dickens: Yeah. I mean, for me, sort of, to trot out the sort of trying phrase, if you will, of 'Cloud is not a place, it's an operating model.' I think we, you know, where the hyperscalers very, very definitely wouldn't ever entertain the use of the word 'hybrid' -- you just couldn't get that word to come out of anybody's mouth from AWS, for instance, up until recently. You know? And it was 'Move everything to the public cloud.'

I think sovereignty's changed that dynamic. I think people -- I mean, I think if I look at the hyperscalers who've got the best story on this, I love what Oracle's doing with Cloud@Customer, sort of the consistent pricing regardless of where you want to deploy it. They've got Oracle Cloud Infrastructure in the other hyperscalers. You know, they can stand up cloud in an MSP, they could stand it up in a data center.

I think that is the most, kind of, evolved model. I think others are still sort of getting there. Thomas Kurian was there on the main stage at Google, the CEO, today. So, I think you wouldn't have seen that three or four years ago. Google would have said, 'The right answer is to put it in our data center and let us run it.' I think we're a lot more pragmatic.

I think going back to your question about, kind of, how should we think about workload placement? A functional environment. I've been banging this drum for 20 years. It's availability, scalability, latency, security.

Look at your workload, see where it makes sense to put it. Don't put it there based on religion or a particular vendor. Just evaluate it from first principles, you know? There's still scenarios where putting workloads on a mainframe makes sense. There's scenarios where putting it on the public cloud makes sense.

The analogy I always use is: Workloads are like journeys. Not every journey should be taken in an Uber. Some journeys should be taken on a train, or a plane or a boat. You know? Ride a bike, rent a kayak, you know. Whatever it is, look at what you're trying to achieve with that workload, and then pick the right platform to run it on.

All the options make sense. I think we're starting to be in a lot more healthy place with custom silicon. You know, people are making workload choices, whether it's on a TPU, a GPU, a DPU, a CPU. I think we're making a lot more evolved choices. As you roll that forward, then, into where that infrastructure actually sits -- whose data center does it sit in? Yours, or do you rent it?

There's lots of factors in play, and enterprises are just going to make nonreligious decisions.

Pariseau: Right. And then, in the meantime, at a little bit of a higher level, so many other vendors with agent orchestration platforms. The Dell Automation Platform is, you know, there's lots of lofty vision about, you know, autonomous agent-driven AIOps. And, you know, I'm old enough to have been around for the first round of AIOps driven by machine learning, and it didn't quite end up being what it was made out to be.

And so, you know, also we hear the scary stories every week, it seems like, of an agent that went rogue and deleted somebody's production database and backups or, you know, just sort of didn't do what it was told to do. So, and of course, every vendor says, 'Our thing is governed, and there's guardrails.' But do enterprises trust AI agents -- generative AI-driven AI agents -- for autonomous infrastructure automation? Do you see that yet?

Dickens: So, I've been having this conversation with a number of the observability platform vendors over the last six months. The sort of way I've been phrasing it, especially, is: There's a human in the loop, but how long does that human stay in the loop?

So if I told you, 'I've done this 10 times and the answer's right,' and you trust it. Do you trust me on the 11th? If I do the same thing and give you a good answer 100 times, do you trust me to do it again on the 101st time? Is it the 1,000th time that you trust me, and it's 1,001?

I don't see it as a switch of trust; I see it as a dial of trust. I think where we are from these observability platforms, you're gonna overlay that trust sort of style that I talked about on top of whether the industry is highly regulated, what sector it's in, you know?

So, if you're a 10-person research firm, do you trust an agentic platform to do a task? If you're a highly regulated, HIPAA-regulated sort of local hospital, do the same task. If you're a nuclear missile silo, do you trust the task, you know? So, the task is the same in all three of those scenarios. The framework around it is very different.

So, say the task is 'Send an email with a daily summary of what's gone on.'

Pariseau: Right.

Dickens: In those, you know, research firms, I'd probably go, 'Yeah, just send that round the team Slack.' Healthcare, there's more to think about. Missiles, missile silo, absolutely not.

Pariseau: Right.

Dickens: Will that dial turn over time? Yes. Will we get more trusted over time? Yes. Are we gonna get 100% adoption? I think you have to keep human in the loop for missile silos from now until the end of eternity.

Pariseau: Well, that's the thing, too, is, I mean, there was interesting research that came out last week from Microsoft's research arm that found -- this was like a document-based delegated agentic workflow, it was not IT systems management -- but found that agents tend to corrupt these delegated workflows 25% of the time. And that, given tools, the kinds of MCP tools and skills and guardrails that people talk about, they got worse. They got 6% worse.

So, you know, like, you know, if I tell you nine times, do you trust me the 10th? If I tell you 100, do you trust me the 101st? With a probabilistic system, I don't know.

And yes, of course, if you're operating a missile silo, you're probably never gonna be hands-off. But, I mean, there's plenty of havoc that can be caused by a system, say, diagnosing a problem and automating a fix, as was demonstrated or discussed today at the Dell keynote with the automation platform, including database deletion and things like that.

So, I just wonder, you know, even if people do start to trust this more, should they? Is the technology inherently capable of handling things that need to be as deterministic and reliable as infrastructure management? Not so much documents and higher-level workloads, but …

Dickens: I mean, I always have this discussion when it comes to autonomous driving. So, people crash cars every day. Humans aren't infallible when they drive cars. But we hold autonomous vehicles to our highest standard.

If a Waymo runs somebody over in San Francisco, it's front-page news. If a local San Francisco resident runs somebody over in San Francisco, it's not front-page news.

Now, obviously, in both those scenarios, I don't want anybody to get run over. I'm not saying that. I'm just saying how we look at these tools -- how many times has a human deleted a production database in the backup? A bunch of times. Does it make front-page news? No, they probably just get a stern talking to from their boss. But when an AI agentic system does it, we hold them to a different bar -- that that should never happen.

So, the way I like to look at this, and I'll go back to the autonomous vehicle: If there's one crash in 1,000 miles of driving from humans, and it's one crash in every 100,000 miles for autonomous, it's 100 times better. Is it infallible? No, there's a crash.

We have people doing things in production that they shouldn't do. And they've, you know, I can think of a few scenarios in my time where we've had to -- like, even the most regulated, most secure, most highly available systems. You know, Barclays had an outage on their mainframe in 2009 from a workload scheduling upgrade. Took down a globally significant bank, and nobody could get access to that money.

This was 15 years before AI coming on. The front-page news for days was lines outside the bank, you know. I was on the periphery of that and saw what happened. The bank spent hundreds of millions trying to make sure it never happened again.

Will AI create an outage? Absolutely. But will it create less outages per unit of work than humans do? I probably think it will. So that's the lens we should be looking at.

Now, is it 100% infallible and holding it to that standard?

Pariseau: No, I suppose not.

Dickens: You want it to be you want it to be infallible and never fail, Beth, I suppose.

Pariseau: But I think if they're talking about 25% corruption in something as simple as a document-based workflow, there's reason to be cautious right now with the state of the technology, you know.

Dickens: Absolutely. I mean, are we early? Is everything perfect? No, we're absolutely early.

But this isn't a fixed problem yet. This isn't complete. We are, you know, the first innings of a seven-game World Series for this technology. You know, I've been in the U.S. 12 years now and I can use sports analogies.

But no, I mean, all joking aside, I think yes, we should be concerned. Yes, we should be worried, but in the same way that we'd be worried about putting a 23-year-old, two-years-out-of-college graduate into a production team. Should you hire them?

So if you're Citibank and you've got to do the overnight batch, and there's a 23-year-old on your team, and there's a bunch of guys and gals on that team who've been doing it 30 years, show the same level of concern and oversight and governance of that 23-year-old as you would from doing it agentically.

That's where -- these systems and these agentic processes are like 2 years old.

Pariseau: Right.

Dickens: They're not 25 years old. So that's how you've got to think of them, as two years out of college, not 25 years out of college.

Pariseau: And I think, you know, the reason we're even having the conversation and people aren't just like, 'Well, don't use it,' is because the dilemma is that we need better automation.

These systems are getting monumentally huge, and they are getting monumentally complex, and we need something to pick up the slack. I mean, I think that's the dilemma, right? It's like none of this is gonna work without some kind of better automation than we have, right?

Dickens: You can't go back and say, 'I'm gonna look at a bunch of screens, and I'm gonna filter the alerts as a human, and I'm gonna eyeball every alert,' and look at a traffic light and go, 'Yes, I need to fix that,' double-click on the alert and then type a fix.

That ship has sailed. It's gone. You're gonna have to use an agentic observability and an AIOps platform just to keep up. So, is that platform perfect? No. Is it the best we've currently got? Yes. Is it the worst it's ever gonna be? Yes.

It's only gonna get better from here. So, I think, you know, if you wanna be a naysayer and a doomer on an AI -- and that's not what I'm saying you are -- but if the listeners are sort of denying this technology, you're gonna be the farrier trying to replace shoes in a in a town that's moved on to automobile or something.

It's gone. You can't stop this. We can't stop building the data centers. We can't stop people using the tools.

My daughter's graduation was this weekend. Took a wonderful picture of her graduating with a cap and gown on, and there was a guy walking behind the picture. I did an AI autocorrect, took this guy out of the picture -- looked fabulous. Am I ever going back to just accepting that that photograph is just gonna have to live the way I took it? No. The same goes for AIOps and automation.

Pariseau: Yeah, you can't put the genie back in the bottle.

Dickens: No, no.

Pariseau: Alrighty, well, anything else you're hearing over there in Vegas that you wanted to mention or caught your eye?

Dickens: No, I mean, I think just the breadth and scale of what Dell's doing, whether that's storage, whether it's desktops, whether it's compute, you know, whether it's revving up into that automation line or whatever, I think I've come away impressed by the breadth and depth of the vision.

Now, there's some things where I kind of tilt my head and say, 'Would I buy that?' But I think overall, if people are looking for an infrastructure provider to kind of do what Dell does best, I think you'll be well served. So, I've come away impressed over the last couple of days.

You know, there's still some work to do. Am I buying automation from Dell rather than buying it from somebody else? That one is a bit of a head-scratcher for me, you know. But there's some other platforms out there, you know, Ansible, Hashi, you know, observability platforms -- the Dynatraces, the Datadogs, the Splunks of this world.

Am I buying, sort of, observability and automation from my infrastructure company? I'm not saying that I don't buy from Dell, but it works and that it's good. It's just, would I buy that sort of layer from my server storage, sort of, compute infrastructure provider? I don't know.

I mean, maybe in some of those SMB or other genius environments where I've got like a pretty solid all-Dell stack?

Pariseau: Yeah.

Dickens: OK, I'm buying that. But if I'm a big sort of heterogeneous bank, a Citi, a Wells Fargo, a Walmart, a Netflix -- you know, from one of those types of organizations with a heterogeneous -- am I gonna be sort of homing in on a Dell from an automation point of view?

So, you know, I've come away impressed. They've got some really good vision. I'm buying what they're selling when it comes to storage and compute, and they're saying all the right things. I think focus and stretch would be my only concern.

Pariseau: Yeah, OK. That's a fair point, and a good one.

Dickens: Trying to not be all rainbows and butterflies and be a bit more pragmatic, but it's been a good two days.

Pariseau: Great, great. Well, thank you so much for taking time out of the show and, you know, chatting about what's going on.

Dickens: Always a pleasure, Beth.

Pariseau: Thank you for tuning in to IT Ops Query. To learn more about enterprise software development and platform engineering, explore our content on Informa TechTarget sites. Find us on YouTube at our channel, Eye on Tech. Subscribe to our podcast to receive the latest episodes as they drop. And if you liked what you heard today, give us a rating and review on Apple, Spotify or wherever you're listening. Thank you for joining us.