The future of employment according to Nvidia's Jensen Huang and gentlemanly hackers

The month of May is coming to a blistering hot end. Across Europe, new high temperature records are being set and unpredictability is very literally in the air as people wonder how they will adapt to these rapid meteorological changes long term. In the world of IT, technology may be evolving equally rapidly, but are the takes from the AI faithful still that hot, or is the discourse increasingly predictable? Plus, a dapper ransomware outfit is gaining infamy as one of the most active and successful cyber gangs in the world. Welcome to the ITPro podcast. Hi, I'm Joan McAllion. And I'm Bobby Hallard. And today we're talking about some of the biggest news stories from the month gone by. Now normally we would have our news editor Ross joining us, but he's taking a well-earned vacation and so Bobby, you have been in the news editor's hot seat. Yes, coming up to two weeks of pure AI and hacking news. Well, let's take a look at one of these first stories. So NVIDIA CEO Jensen Huang, he has been letting his opinions on AI and the future of employment be known. So can you tell me a little bit more about that? Yes. So as we regularly cover AI, Jensen is very rarely away from the top part of our stories. He's always there. And this kind of story is kind of the bit where Ross left and I took over, I guess. It isn't a really hot take. It's kind of common knowledge, I would assume, that what we really need with AI is the infrastructure. So maintenance, basically. And he's kind of just spelled it out for everyone what that means or who that applies to. So we're not thinking coders and programmers and so forth. We're thinking electricians, plumbers, people that can manage basically a data center or keep it running or even build them. So Jensen Huang was at Carnegie Mellon University for the class of 2026, giving a commencement speech. And he spoke about economic growth through AI, but how it will be not so limited to the technology sector, but to trade workers. So like I said, electricians, plumbers, iron workers, technicians, basically construction. And so it's kind of a nice visualization of what they say that AI will create more jobs than it will take. So in this case, it's just we're going to have less coders and less people actually developing the software of AI and more people handling the physical bit of it. Yeah, it's a fairly, as you say, it's a fairly tepid take. Did he manage to avoid being booed by the audience? Unlike almost every other commencement speech where somebody has started talking about how brilliant AI is. I didn't see him get booed. I feel he's kind of un-booable just because they let it get to it. You just can't picture him being booed off stage. I think he's got a good connection with audiences and he presents himself quite well. I guess it depends on who he was actually speaking to, what kind of students they were, because I don't think it was a more doom and gloom, was it? It wasn't like, all your jobs are going and there'll be nothing for anybody. Yeah. Also, I suppose the ones who have been booed most recently did tend to start off with, so AI, brilliant, eh? But yeah, as you say, he does speak well. He is good. He has an excellent stage presence. There's a reason that he is sort of a bit of a celeb in tech circles. It's a fairly tepid take. The idea that the professions that will survive the AI revolutions are the ones that need to physically be done, trades, as we say in British English. Also, while Jenson Munger mentioned this, things like hairdressers, carers, nurses, doctors, whatnot, they're all supposed to be kind of fairly safe professions as well. I think there are two kind of problems, if you like, with this. I mean, it's like, excellent, we can all work for the infrastructure of the AI overlords, but not everybody wants to do those jobs. Some of us want to work in things that are a bit creative, and I think being a coder is a creative enough job as well. But equally, those are now fairly well-paying jobs. But if there's a glut in the market, then you're going to find that wages are going to be pushed down. And also, there's a finite amount of electrician and plumber roles, whether you're specialising in data centres or normal domestic or office buildings. I just think it's a sunny-ish take, but it's kind of shallow as well, in my opinion. I did think he did well with actually spelling out what jobs, but those jobs are not necessarily long-term jobs. So for example, you're a plumber, you build the piping for the data centre or you fix it, then the expectation is, what, is it going to keep breaking down, you're going to have to be on call? Because surely once you've done it, you then go on to the next job. Do a really shoddy job. Yeah. Do a really shoddy job. And therefore, you make sure you get repeat customers. There's also, part of the problem is a lack of infrastructure investment. So within that piece, there was research from McKinsey, and they predicted that data centres will need roughly 6.7 trillion investment by 2030 to keep pace. So while we're spending hard on AI and pushing out all this money for that specific technology, what it actually needs is the physical bit, which we're spending less on, by the looks of it. Yeah. And I mean, that's sort of a controversy in its own right, really, isn't it? I think it's probably unfair to call it nimbyism, actually, but a lot of ill feeling towards the construction of data centres, whether that be because people are worried about it being an eyesore or noisy or resources, especially in like water-scarce or water-stressed areas. If it's a water-cooled system, obviously, that's highly efficient. You need less energy to cool the data centre than you would if it was just air-cooled. And equally, now you get warm water cooling where you can achieve the cooling that you need, even with water that's maybe up to 40 degrees, which, in my opinion, is a little bit warmer than warm. They're closed-loop systems, but you have to get the water into the system in the first place, and it will need replacing every so often. So there are concerns. If somebody wanted to build a data centre next to my house, I'm not sure that I'd be overly impressed, and I know that they're not that noisy outside of the housing for it, inside they make a racket. And also, I am not a specialist enough in this area to be able to talk to it deeply because I don't have an economics or business background, but I know that there are genuine concerns about circular investment when it comes to AI data centre infrastructure. And NVIDIA is implicated in some of those concerns that everybody is funding each other to buy each other's kits. So whether or not there's actually a sustainable business model there and what happens with data centres and with demand for AI is another matter entirely. I mean, I was at Dell Technologies World last week, and one of the big messages there was that actually AI agents in particular are super expensive if you run them in the public cloud, as it's being called. So within the offerings from OpenAI or Anthropic, and that actually, once you've got past the training, you're into inference, you're into agents and stuff, it's more economical to bring everything back on-premises or into co-location. So argues Dell Technologies, who obviously have a stake in this argument. They want you to buy their desk-side AI infrastructure and whatnot. Naturally, that's also an NVIDIA play. But we're going to see how this plays out because there are several big data centre projects that we've written about in recent months that are yet to break ground and look like they possibly never will. Do you know who else was at Dell Technologies World? Who? Jensen Hurang. He was! He had probably a more, I won't say a spicier take, but a stranger take, which you covered yourself on the usefulness of AI. Yeah, it was an odd one. So he said AI agents are the first useful AI. So to quote him directly, he said, two years ago, when I was at Dell Technologies World, we had just started with generative AI as the product. Generative AI can, of course, generate content, but remember, it can also generate content to think with, generate thoughts, which led us to reasoning, which led us to planning, which led us to agentic systems. So now we have, for the very first time, useful AI. And I feel like it's quite the claim, and I am pretty sure that there are going to be a lot of people out there who think that this is very much not the case, who have been cheerfully using generative AI, from which agentic AI has sprouted. And I think they'll tell you it's pretty useful, actually. I feel like he's missed out some context, maybe, as if he's saying it's useful for this specific thing or in this guise, because it doesn't, like you say, it doesn't make sense. We've got useful AI everywhere. What do you mean, Jess? Yeah, I mean, as I said, there will be plenty of people listening to this podcast who use generative AI regularly, and not that I ever would, on the IT Pro podcast, get on a hobby horse of my own. But we're leaving aside things like computer vision, which is making huge differences in medicine. I've seen it up close. Somebody was able to talk to me about how a programme that they use has saved them way more time, so they can actually see more patients. And this programme, an AI programme that looks at x-rays and bone scans and is able to determine bone age and growth and whatnot, very, very accurately. The person I'm speaking to still does some sort of manual assessment to to double check it's human in the loop type stuff. But it's had great impact on her work. The thing is, it's not the new hotness. It's the kind of AI that I get excited about. It's not the kind of stuff that's desperately trying to draft my emails for me or write a document for me and all kinds of things that I find very unhelpful. It's speeding up assessments and increasing quality of care and the amount of physician time that is available. It's putting my carrots on the scales in Sainsbury's and having the computer look at it and go, yeah, that's a carrot, and it weighs this much. Here's your label. Or wondering what this plant is and using Google Lens or Apple's inbuilt image recognition stuff or image analysis in photos on iOS. This is all useful AI. Maybe I should write him a letter. Excuse me, Jensen. I think you're fine. So let's move away from Jensen one because he wasn't the only one talking about AI agents this month. Was he? Gartner had some research out this week, the final week of May, about AI agents and governance. Yes. And this one I found interesting because you can kind of visualize it as you would actual employees. So the idea is that they don't want one size fits all governance. A load of agents with all the same instruction or the lack of instruction. So you wouldn't take a section of your employees and give them access to everything or leave them to run consistently and hope that everything works out because it would just end up in failure or data privacy issues, all kinds of problems. But what Gartner's report kind of said was that if you fail to distinguish between an agent's ability to act and the scope of access it's granted, then it will come down to some kind of a failure. And these are often not noticed until the failure happens. The senior director analyst at Gartner, Shiva Verma, said that organizations are failing to treat agent governance because they use it as a binary. So it's either locked down or it's fully trusted and that is the cause of the failure. I think there's a good example of this is the Waymo cars that keep getting caught up in cul-de-sacs. They've got no riders and they've got no jobs and they just turn up at this one cul-de-sac and go round. It was on social media making noise and keeping all these residents up. And that's, I guess, because they haven't got an instruction for what they do in that scenario. Yes. And as I understand it, they're not allowed to park. So they're just kind of going around this little cul-de-sac. They tutor each other as well. So for, I don't know, for reasons, cuteness, I have no idea. So yeah, now our choice is which AI hell do you want? Do you want to live next to a giant mega data center which is running some kind of the latest generative or genetic AI systems? Or would you like to live on the cul-de-sac full of a herd of unoccupied Waymo cars with nothing better to do than hang around in the streets? I suppose part of this problem is also environment where it's like a bigger environment with more instruction is needed because it's not something that ever comes up in factories where each sort of robot or system has got a specific job and it's easy to get them to work in that small environment. Like at Waymo's, they're in the city in the world. They've got all sorts of probabilities. So I guess that is the problem with agents now as they're going into offices or, like you say, data centers. There's more things to go wrong, more things to think of. The one size fits all prevalence is a challenge, I guess. Yeah. I wonder, you mentioned sort of manufacturing there and I wonder if it is also like a maturity question. There's been automation in manufacturing for decades, decades and decades at this point. My first journalism assignment was to go and look at a bunch of factories in the north of Italy and I saw some very cool robot arms and got some brilliant photos and whatnot. That was longer ago than I'd really care to admit. Not quite, but getting on for 20 years ago. That stuff wasn't new when I turned up. If I were to go and visit the car manufacturer close to where I am, I know that a lot of it is actually just robots doing a bunch of stuff, which is from a health and safety point of view, perhaps no bad thing. From an employment point of view, perhaps not ideal, but it's that level of automation, that level of robotics has been embedded in that sphere for decades. Agents are new and it's almost like, it's almost like back to the kind of first days of the internet where you connect everything up and sort of hope for the best. I have seen people talking about zero trust with regards to agents and policies of sort of least privilege and that kind of thing. I think it's going to take a while for people to really pick up on why you need to do that. I think there's going to be a lot of hiccups, really bad, big hiccups that are going to make the news before people realise that actually agents aren't magical. You can't just let them do whatever you want. At Derweld I did also attend a press panel where the CTO John Rice was and I was able to ask him about his thoughts on sort of AI agent security in a general conversation about resilience regarding a recent story that also happened this month where a rogue AI agent deleted an AI startup's entire production code base and also all its backups, thinking in inverted commas that it's the right thing to do. They don't think, whatever. And yes, perhaps you should not give an automated programme access to both all your backups and your production code base. But some people are going to do it. People already have. I mean, we're talking about it, I'm talking about it. And I think we'll see more of that before people get a grip on how you actually work safely with agents. This could also be quite costly. I'm going to take it back to Del again because token economics, the agents can, if you don't set the parameters, burn through all your token allowance, which is essentially the currency of it or whatever. Yeah. And while we didn't cover it, I know that Uber's CTO has been talking about basically assessing from an economic point of view whether or not agents are actually worth the investment because we have covered this idea of token maxing and people just do as much AI as you can and just use up as much tokens as you can and using that as a measure of productivity. And that's just doesn't work that way. It's like measuring whether or not I have made a good meal by the amount of splatter there is on the back splatter of the hub. Who knows? Maybe it's brilliant. Maybe it's simmering nicely for a long time and it's full of flavour. Maybe I had the temperature turned all the way up and really it wasn't great at all and everything was cooked far too quickly. But if you just measure by counting the splatter on the back, you're not going to know. And it kind of, it is that kind of thing that your tokens are not an effective measure of anything other than the number of tokens that have been used. It's not measuring anything else. It is something that can easily get out of hand. So let's bring this background to Gartner though. So what is the answer to the problem of AI governance? So Gartner believes that the solution is in the form of proportional governance approach that classifies AI agents across different distinct autonomic levels. So each level represents a different trust boundary. It sounds more complex than they make out. The first covers sort of observe, as they put it, you observe the agents, limit read-only access, and you kind of define your data points, which kind of just seems like common sense. Like you're not going to just let the agent run wild with every single bit of your data, your organisation's data. I don't know, Bobby. Apparently some people do, but carry on. I've built this agent. It's just going to go everywhere and everywhere into our customers. It just sounds mad. But I guess this is the world we now live in. I expect every year you get this new buzzword of stories and trends, and I expect at some point we're going to be talking about rogue AI agents more than we once did. Kind of like when we used to talk a lot about poor cloud configuration. Yep. Leaky buckets. It'll be agents doing more than what we intended them to do, or businesses having hits to their finances because of agent trouble, or that kind of thing. Seems like that's the future for AI agents and people that report on that stuff. How exciting. So I'm going to move us away from AI agents now, and let's talk about the gentlemen who I referred to in the introduction. This is not the Guy Ritchie movie. No extraordinary gentlemen here. No leagues. No. No league of gentlemen. No extraordinary gentlemen. Just the gentlemen. Back in my day, the ransomware groups had much fearsome names. Or more rememberable names like Fancy Bear. You kind of got more of a feeling of what they did. This is just kind of confusing. See, I'm pro this name. Because Scattered Spider, and Fancy Bear, and APT47, or whatever. In the cybersecurity industry, there's a lot of FUD. There is an element of coolness to cybercrime. And if you've ever been to RSA, you will get that kind of FUD slash we're cool cybersecurity defenders vibe. So these names are given. So like Fancy Bear, it's Bear because they were out of Russia. Fancy, I don't know why. APT, whatever, was supposed to take away some of that glamour. Scattered Spider, similarly, it's kind of like, oh, they related to these guys, whatever. The gentlemen, in my mind, are sort of like LulzSec, if anybody is old enough to take that trip down memory lane with me. It seems like they're perhaps a little bit more, not that LulzSec were a bunch of good guys, but they are a bit more malicious than that. Because they do double extortion malware, if I'm correct. Yeah. Well, it's not necessarily a new group either. It's been around for a while. Kind of gained notoriety recently. So from an NTT study, they are the most active ransomware operators, second only to Quilin. And so what they're saying is 10% of all attacks are this, the gentlemen group. It's interesting, we were just speaking about other former groups. The NTT suggests that this is like a sophisticated group of people. And it suggests that they've come from all these other places. So it's a sort of amalgamation of other hackers and other groups into this, the gentlemen. Yeah, sure. Why not? It's a super group. It's a super group of hackers. You don't see many super groups anymore. No. Well, in the world of cyber security, apparently you do, or cyber crime, in fact. Something that did catch my eyes, actually. I mean, it's interesting to hear you say that, because I did try to, before coming on the podcast, I did try to look up a little bit more about this group. So Quilin are a Russian language threat group. So many of the others, I believe, Scattered Spider are Russian as well. So you can either trace them back to organised crime in Russia, or state-backed actors, typically in China and North Korea and stuff. Whereas this group, I couldn't find anything that really geographically located them, and who they're going after seems to have changed as well. Like some of the earliest reporting, they were looking, it was more sort of Asia and stuff. The most recent reporting which we covered, it seems to be more targeting Europe, I believe, like the UK and Germany. So they feel kind of somewhat nomadic, somewhat just able to turn their attention to wherever is going to be the most lucrative spot for them. I believe, so they are second only to Quilin, the Russian group, but they are now responsible for 10% of all cyber crime, that the gentlemen are. Yeah, 10% of all attacks. Is that globally or in Europe? Well, I think we should take it as around the world, because it says that, like you say, extorts organisations in Europe, with the UK and Germany largely. So that would suggest they do have a lot around the world. Interesting that notable victims include Synergy France, UK Electronics and Equity Life. I suppose if you're looking to get some money, then Equity Life or similar organisations are probably a fairly tasty target. I mean, obviously, there isn't really a way for anybody to sort of protect themselves from the gentleman any more than they would otherwise protect themselves. There's no kind of special formula. It's just do your cyber security, try to prevent a breach and make sure that your setup is such that if you are breached or you do suffer a ransomware attack, then you're able to restore from secure backups and get in a way that's not going to cost you more than paying the ransom. One of the specialisms the group seems to have is covert tunnelling, which I guess is how they go undetected for so long. So what does that mean? What's covert tunnelling and how do they do it? So covert tunnelling and rapid domain-wide deployment, like a shrinking window. So they come in and they come out. Once they're in, they kind of use lateral movements to go across your business. Not especially clued up on covert tunnelling, other than the fact it just sounds like they come in very secretly and leave even quieter. Yeah. Taking that phrase literally. Well, I'm sure that we will be hearing about them and indeed about these tactics in the future. That's also, unfortunately, all we have time for this week. But Bobby, it feels strange to thank you for coming on, seeing as how you're a host, but thank you for taking the hot seat and standing in for Ross for this month's news wrap up. Thank you for having me. It's been a blast. As always, you can find links to everything we've spoken about today in the show notes and even more at itpro.com. You can also follow us on LinkedIn and YouTube and why not sign up to our daily newsletter to receive all our latest news, analysis and reviews direct to your inbox. We'll be back next week with more from the world of IT, but until then, goodbye. Goodbye.