Mike Clayton interviews James and Yoshi on what to expect in 2026

[Music] Hi everyone and welcome back to another episode of the Project Plus podcast. Today we truly do have something a bit different because the guests are myself and Yoshi. We were recently lucky enough to be invited onto Dr. Mike Clayton's YouTube channel to talk about what happened last year and where we're going this year and it was such a brilliant conversation that we wanted to make sure it went out to all of our listeners. It was a great conversation wasn't Yoshi? Yeah and we spoke about some of the killer things from last year, you know, Continuity in 2025 also the next couple of years, "Elegance Benchmarks", "Lucinations", "For Structure", "Very Project Centric" and it was a great conversation on an all James. And I do encourage everyone to subscribe to Mike Clayton's YouTube channel. You can find them at onlinepmcourses.com and we'll put that in the show notes. But for now we're going to hand straight over to the man himself to interview yours truly. [Music] Okay, Yoshi, James, welcome. Thank you, thanks Mike, lovely to see you again. Okay, so, hi. So, Yoshi Sawnack and James Garner are the guys behind Project Flux, which is for me the best source of regular upstate information about what's going on in the world of AI as affects the project profession. They're both project professionals in the construction sector and the content they produce is fantastic, not just an index to what's going on but also some very informed, very carefully considered commentary. So, we'll make sure we link to Project Flux in the notes. But today I've got Yoshi and James on the call to find out firstly what their highlights of 2025 looking back with the perspective of early 2026. But also, what they see as the likely big picture changes or trends or possibly even events for 2026 in the world of AI and project management. So, can both of you. Thank you. Thank you. Have a nice. I'd like to kick off perhaps with Yoshi. What would you kind of pick out as your first highlight from 2025? Firstly, the money to be made in the stocks, right? In tech stock. Into the bubble, if you pull out just the right time and if many of us got that right with the internet bubble of the 2000s. Yeah, I mean, if you can ride the AI way right and then predict just before it burst, I think you're making a lot of money. So, that's I think that's the big highlight of last year. No, but more, I think a key thing for me last year, right? And I think it's having a massive systemic effect was this illusion of intelligence because of benchmarks, right? Right, up and all the way through from the back end of 2023 or the start of 2023 to 2025. Everyone was kind of caught in the solution of seeing these models smash intelligence benchmarks across the verbal numerical abstract reasoning, right? And people who were in the camp of AI and were thinking, okay, this thing is going to change the world were really brought into that. And I felt like we got carried away with it. And in the actual in actuality in real life, what we found was these models wouldn't hold up when applying them in real world context. So it might have PhD level reasoning, right? It might have state of the art model that can smash all of these benchmarks. But then it was struggling to just hold a conversation for three or four messages if it was in a subject that someone is a specialist inside. And I felt like that's created a huge problem last year. And it's one of the reasons why a lot of implementation projects have failed in businesses. And I think that needs a big correction going into the new year. And it's interesting because with the ability to create like custom gpts in tools like chat gpt, does that mean that even if you feed it a lot of contextual information, there are still those shortcomings? Yes, I mean, this is the most easy one you can say right. Every AI when it speaks to you will say about any topic, it will say something something isn't this, it's this. And it's to make you kind of frame a really complex subject in a really easy way to understand. So, for instance, you know, these AI models smashing intelligence benchmarks, I'm just the drop in the water. They're a massive transformation of how we're going to consume intelligence and use it going forward, right? And we keep saying that. Now, you go into the custom instructions and you say, don't do that. It will still do it. Despite how much you refine those systems and how much you tell it not to do it. You can even swear, I was like, don't do this, right? It's still going to do it. And that isn't the sign of an intelligence machine, right? I think that's where the shortcomings are. Is these things are hardwired to act in a way which can't mimic human intelligence. And the problem is we're focusing on building on emulating human intelligence, right? So, I think that's that's the core issue that some of these companies are trying to solve, right? So, for instance, how do we apply these things in the real world? Okay, openly, I now have this evaluation set, GDPEVAL, which says, how well can these models actually deliver the task we might see in businesses, creating spreadsheets, creating PowerPoint? How well can it? So, they've changed the benchmark. And the other thing is, ChatGPT, you know, and you recently have just incorporated this confidence score. Yeah, right. We'll tell you where you can adjust the confidence. So, if the confidence is quite low, you know to add more of your verification as a human into that output, than if it was a lot more of higher confidence. Yeah, I think that has actually been positive. And we've been, you know, those was an approach of profession. We've been arguing we should be doing that with our risk estimates for many, many years. So, I like that one. Thank you. James, what would be your first big thing from 2025? I think the realization, as Yoshe said, that what we've got is incredibly intelligent pieces of tools, if you like. But there are things that it's not very good at yet. And so, the things that Yoshe talks about. But the realization that that's actually an opportunity, because what we don't necessarily want, well, I don't think we're ready as a professional, or as a species for full AGI yet, so artificial general intelligence. So, when people say, oh, I use AGI, but it doesn't do this very well. I say, hang on a minute. If we actually turn it, I said, hey, I could do everything. You'd probably be moaning, because then we would have a serious job displacement for you. So, we're in this lovely sweet spot at the moment, where we've got a very powerful tool that can do a lot for us, but we still need the human layer to add on top of it, what I call the book ending approach, the human end of the beginning and the human at the end. And that gives us some breathing space, I think, to understand how we're going to use AGI responsibly. Because one of the big highlights, I think, of 2025, was seeing the profession start, starting to realize that they have a responsibility about how we govern AGI. So, the OICS, the Royal Institutional Child Service released their using AGI responsibly guidance. And I think that's, we're going to see a lot more of that. I hope that we see a lot more of that throughout 2026, rather than sort of handing off responsibility to the tech companies. The profession actually saying, right, we have this, we have technology now, and we have a responsibility as a profession to think about it. So, I'd say that's one of my highlights in actually seeing us trying to grapple with this problem. We're not there yet, we still have a long way to go. And of course, there's going to be other issues that come up throughout 2026. But actually seeing the profession almost head in the sand, maybe in 2024 or that, 2025, it was kind of the realisation. So, there was a lot of pilots going on, there was a lot of experimentation, which is all healthy, really healthy, good stuff. And that leads us nicely to what I think is going to happen in 2026 when we get to that section in terms of moving from pilots to actual proper adoption. I think the other big highlight for me, Mike, was the realisation that this isn't, and I say this a lot, that this isn't a technology roller. Even though it's a very powerful technology, this is actually a change management piece, like any other change management. And what we have, we could launch the best AI in the world. But if we're not preparing people to use it properly and responsibly, then it's not going to be adopted. So, I think that realisation has started to take on over 2025. Yeah. I think a lot of those themes I picked up on, particularly, this move from pilots' experimentation. So, certainly there has been a focus in the business press about moving towards adoption and generating value as an aspiration, and perhaps the first case is. For me, I think I'm going to pick actually three new terms that we, I mean, they may have been around before 25, but I was first, I first became aware of them. And one, thanks to you, James, which is a Gen TKI. It feels to me that this has been the year that Gen TKIs come across our desks in the way that large language models came across our desks a few years back. And that's going to be transformative, I think, because we're going to start to see people really delegating more of their authority to it. And I think you're absolutely right about some of the risks. It wasn't last year. It was just a very, very end of 2024, in 724, where the Nobel laureate Jeffrey Hinton made a speech. And he pointed out that these tools are created by companies that are motivated by short-term profits. And our safety is not their top priority. And I think that's very true. Just on that bit there, then. Yeah. I find this really interesting, because we've just seen meta acquire Manus for millions and millions of pounds. And this is the first company who haven't created their own AI model. But they've commercialized a Frontier Labs model to the point that they've been acquired. Well, they had their open source models. Yeah, yeah, yeah, but I still think a lot of it was, you know, a lot of the key power of Manus was coming from Claude initially, right? And now they're being acquired by one of anthropics competitors. So I feel like there's, in the Gen TKI space, there's opportunities for people to build a Gen TKI systems using a combination of these models to sweat the asset, to get more out of it and commercialize. Yeah, I think that's absolutely right. Yeah. James, you were there? Yeah, the other thing that's interesting is that even though you're right, they're motivated by short-term profit, I think the pressure is on now. And a big story that's just hit this week is that open AI are hiring for something called the head of preparedness. And they're offering a lot of money. So if anyone's interested to go and check out the open AI website, but this is about the potential impact of models on mental health and the impact of models on humanity. Even though they probably wouldn't do it unless they, the pressure was on, they are starting to realize that these things are important to the people, to their consumers, to their. Absolutely. In fact, the second term of the three that I was going to highlight is AI psychosis. Which isn't a mental health thing in the sense of, it's not a reflection of necessarily that term, just about the harm that AI can do to us. Although I think there really is a risk that, increasing, very rapidly increasing reliance on asking questions of AI is taking away not only our ability to think for ourselves, but risks taking away our critical faculties of challenging what our answers. But AI psychosis is, actually, it's about. It refers to the kind of uncritical belief that AI is firstly competent, but secondly and even more were really conscious. Of course, we don't know what the future holds and there are some people who say that it will cheat consciousness at some point, but. Well, there's something to say already is. I mean, it's fringe people, but. Well, I think that at the moment is a psychosis, I don't think that's actually true. Well, it's absolutely, but it's also, we're dealing with it. Something was that the definition isn't very clear. What is the definition of consciousness? Well, you can't define it. So if you can't define it, then anyone can say it depends what your threshold is. We are definitely entering very interesting times. And as these AI's become smarter and smarter, and as they inevitably become physical as well, in terms of not just being chat parts, but being robots, they'll start replicating and appearing more and more human. And at what point, we're getting to sci-fi kind of territory here, but at what point do we say, actually, this is as conscious as we are? Yeah. It's funny, you mentioned that as well, because one of the arguments of consciousness is attention. If you live you believe in emergent complexity, you would believe that attention is a hallmark of consciousness. Or one of them. Elements are founded on self-attention. They start you different, but again, that's where you might see some of those links coming through. Yeah. Any other big trends, Yoshi? Or any other arguments last year? Not so much big events, but a still-of-plaging issue that I think is impacting the implementation in businesses is hallucinations. I don't think we've solved hallucinations. And I think if we don't solve hallucinations, it will become harder and harder to implement into businesses. So for instance, you know, insurers are pulling out of AI or AI coverage because they can't guarantee that what the information that AI is retrieving is accurate and reliable. You're working in construction and projects. They're heavily complex systems. And you have to get the information correct. That's why it's built the way it is. And that's why there's a lot of bureaucratic processes in that profession. But I think hallucinations are still plaguing the industry at the moment. I don't think they will be solved by the current generation of large-language models because of how they're built in their architecture. So what do we do? I think what we do is try to scaffold the AI so we minimize hallucinations as much as possible. But now I don't think, when I say that, I think that the box has now been passed to businesses. Right? So if we know there are limitations in hallucinations, businesses need to put the right protocols and parameters in place so we can minimize the hallucinations. And that's why it's so important to upskill people in businesses not just those in their labs. Yeah. So I'm going to come back to something you mentioned earlier, which is the idea of confidence limits. Because obviously that should be one response to the threat of hallucinations. But to what extent can we rely on the estimates of confidence? And to what extent is it your view that those aren't potentially hallucinated confidence levels? Yeah, it becomes a like an infinity regression, right? Exactly. When you happen. And we've had this before when we were evaluating. We had an LLM as a judge, right? It's an example. You try using LLM to judge outputs. Well, hopefully you can keep going on and on and on and on and on and on and it will always try to re-evaluate the work that it's done. It doesn't truly know when it's finished, because that comes with human intuition in my opinion. And also the fact that we can rely on experts around us to say, yes, this is probably about right. But I think confidence levels might be slightly gimmicky. Yeah. But we'll have to see how it runs in the first few months. But if we believe that AI could attain human type consciousness, then it could have human type hallucinations, which is there are certain mental conditions where people are absolutely certain of their hallucinations. They have absolutely no doubt. Well, we just got to take the Nelson Mandela theory. I don't know if you know about that way. People are convinced that certain film came out. And they really are, because our memories are always kind of a gougress to a certain extent. It's never exactly what we saw. So we all will suffer with it to some extent. Yeah, absolutely. But any other big things from last year, James? Yeah, well, I think there were three big stories for the project professional people to be aware of. The first one was the one that, towards the end of the year, where ACON bought out, they spent $390 million on an AI company. I could not consider it. I thought that represented a big shift where we've got consultancies starting to buy tech companies. We've had it the other way round, but not that way round before. And that's going to be interesting to see what happens where the line starts blowing between what is a consultancy and what's a tech company. I thought that was a really big story. The other one was around, there was a study that came out, probably about summertime, the MIT study Yoshi, which showed that 95% of AI projects or pilots were failing. I mean, it's a bit more nuanced than that. But that was a really important study done by MIT, because what was important, what wasn't the headline, it was the reasons why it goes back to my other point. A lot of the reasons why it is around upskilling and it being rolled out in the wrong way. And then the other story that I thought was really interesting, which goes back to the whole, using things responsibly, was the Deloitte case, where they had to refund the Australian government back an awful lot of money, because nobody had checked the output of a report that had been AI generated. And again, that kind of links in with the other two stories, because what we're saying is, that wasn't a tech technology failure to my mind. That was a human failure, just because AI has been used to still need to check it. So those three stories are repeated time and time again. And I think all three of them will look back on as kind of milestones over the next few years. Yeah, and that kind of brings me to my third term, which is AI Slop, that has become ubiquitous now in all sorts of things. But potentially, picking up on Yoshi's point about hallucinations, plus the fact that it might be producing broadly correct, but low quality thinking. I mean, I think one of the things we've seen a lot of in the professional world, and let's face it, a lot of this for me has been on LinkedIn, is we see a lot of thought pieces that are clearly not, have clearly have no human thought in them. And if there's something I want for 2026, but I'm not predicting, it's for people to start actually thinking for themselves and going back to writing thought pieces for themselves rather than using AI to do their thinking. But you know, that's interesting, because it's not an either or. So I always say to people, yeah, AI starts bad. But what would you prefer? Carefully created AI piece where AI is being used to help someone understand their thoughts and get it out there, or an awful piece written by a human, where they haven't thought about it at all. So I think it's more nuance than just saying, AI stuff, rubbish, human stuff, good. Because if you use AI properly to write a piece, to assist you in writing the piece, and then it can help with it by writing it, but you're getting your thoughts out, and then it's maybe critiquing your piece as well. Then that's a productive use of AI. So AI Slop, yeah, there is an awful lot of stuff where people just press the button and they've read it, and they've just put it out there. But I would challenge anyone to really tell the difference between a human written piece and a piece written with AI, but consciously, and carefully using AI as a collaborator. Yeah, I agree with that. I'm not saying don't use AI to help you write stuff, but I think a lot of people are just pressing the button and saying, "It's on to this response to this post with an insightful comment." And you will know them. This is a marvelous insight into what we need to be discussing, how would you view this and those kind of things, they're just not adding any value to-- Yeah, I agree to agree to that. I think there's a huge issue. I think AI, when it was initially trained, these wide-language models were trained on the entire to the internet, which represented a diversity of human communication. A language is a massive thing. It's the most important thing that ever came out from our species, right? And more and more content, you know, in that internet theory, more and more content is being produced by AI. When we then retrain these models, you're going to get them retrained on content written by AI. And all what happened is you get this kind of collapse function whereby everything becomes convergent. And that is a direct cause of cultural diminishing. So we like to lose our diversity and our culture. So for instance, people say English is the best coding language. So everyone then has to start being a lot more proficient in English, for instance. But then what about all the other languages? What about the cultural nuances and how people sentence things together? And even if someone, I'd rather have something poorly written by a human that doesn't make sense, than something articulated in a very smart way by an AI that still doesn't make sense. Because the latter is a lot more believable and can lead you down the wrong way. Whereas if you know something's wrong, you can start to pick holes in it and see where it's gone wrong. Yeah, full sense of security gives you, doesn't it? Yeah, and everyone's doing, right? So AI is telling them, oh, yeah, that's really smart, whenever you're writing stuff. Someone's got the thing called the soggy cornflakes and business idea. And you go around asking every single AI, I'm going to start a business about soggy cornflakes. I'm going to sell cereal and it's going to be soggy. And some of the AI is telling them, "Yeah, that's a great idea." It's not a great idea, right? Yeah, believe it. So these things are trained-- I don't know what soggy cornflakes are, it's quite nice. It's for attention, and I think that's really damaging peace within language. Because language is a way which expresses a millennia of human culture, and I think that needs to be sustained. Yeah, and actually, I think I should spend hours on the topic of language. But yeah, fundamentally, human language changes. So I think a lot of people pass my age perception of language from a younger generation is that their language is in poverty and needs to be sited up. But actually, language evolves. The problem is that actually AI might stifle that evolution. Yeah. Because it's been trained on a particular iterations. But I think we are starting to stray into what might be in store for us as Project Professionals looking at the world of AI in 2026. Let's start with James this time. James, what would you pick out as the first thing to comment on for the possible future? I think the big overarching theme of 2026 in our profession, sort of leaving the tech advances to one side for a moment, is going to be that, as we said before, that removing now, we've had a couple years of experimentation and pilots to actually deploy in things properly and get and getting real value out of it, which means that we're going to see more and more of what I'd call the AI Project Manager. Rather than just a Project Manager who happens to use a bit of AI, it's going to become almost the norm where these agentic stuff and agents become part of the workflow. And you will be a massive disadvantage if you're not doing that. I think that's going to be by the end of 2026, will be in that situation where it's going to be almost impossible to ignore and people will be proving real ROI, real value from AI as well. So it will move out of that phase. I think the other thing that will happen is obviously a lot of the experimentation. Another thing, I just want to kind of come back and challenge you on that because to make that work, we're going to need a lot of knowledge and a lot of skills. And there are people like you who have spent, you know, the last couple of years really steeped in this and trying this stuff out and learning things. But you've also been talking to people. So what extent do you think that the knowledge has built up to a sufficient level to do that? I read in one of your pieces about APM not having the knowledge there. If you look at what PM I am doing, it's focused on Project Managers implementing AI projects rather than Project Managers using AI tools. So are we really ready for that? Do we have the infrastructure and infrastructure to really do that in 2026? So what I've noticed, I do a lot of talks as Yoshi does throughout. And we always do polls. If you went back to the beginning of 2025 or the latter part of 2024, you would ask people how many people are using AI on their projects. And it would be maybe 10, 15%. By the end of last year, there'll be only one or two people on audience who said they weren't using it. And that to me is the big asset test. I think people use it. The problem is there's a lot of shadow AI going on. So people aren't sure how to use it. So they're using their own version of the chat GPT and things like that, which is really dangerous. That's where I think the organizations are going to have to get houses in order to give people credible alternatives to the free ones out there. But I actually think now, people that I never thought would use it, I'm using it like members of my family, who never would have touched it. I think we've crossed that threshold last year, actually. I think it's not everyone. And I'm not going to say I'm not saying it's going to be easy. But I think we have crossed that threshold now. And that's why I'm quite confident that 2026 is going to really push us towards AI. And that's going to be the norm. An AI project manager is going to become the norm by the end of that point. I may be wrong, and I'll be really disappointed if I'm wrong, because I'll be thinking, wow, we're falling way behind other industries. But everything I can see from the talks I've had with people and the polls that I've done with people show there's been a massive shift over 2025. Perfect. And you were going to say something else before I interrupted you? I can't remember. Why do you find that memory? Josh, what would you start with in thinking about 2026? 2026, I think a really important part is strategy. Not strategy, it's we've been saying for years. AI will automate your repetitive task and it will allow you to do more strategic work. I think 2026 will be the year we'll be finally know what that strategic work is and for project managers. We don't quite know what it looks like. We don't quite know how much of that can be automated as models also improve. And that's why the uncertainty is what scares people, right, ultimately. And until they know what that strategic work looks like, they're still going to be scared and you're going to get poor ROI on your AI projects, right? Yeah. That's what I'm thinking. The second thing is context engineering, right? So I think that the one thing that will evolve AI, the current generation of AI models, or even the ones after, significantly, will be how well they can work with memory, right? How much context can they grasp? And how well can they operate within that context? If we have models which can still hold a lot of context, I mean, Google can hold up to a million tokens, right? A million contracts window. But what you find is performance degrades after about 250,000 tokens. Again, right? So there's this million-- For the benefit of people who don't quite know how token scale, can you put that into a kind of human context? What a quarter of a million? Yeah, so essentially, a token is like 3/4 of a word, right? And if you've got-- if you're trying to hold a million words in your head, hey, you might be able to remember it as a machine, but how do you navigate all of that to make sense within it? So as you can imagine, as you scale from a million to 2 million, 3 million, the harder it is to operate on a mover within that context. That suggests if I put a book off my shelf at random, that's probably about 100,000 tokens. It's about 60,000-- Up your word. Yeah, exactly. Now, if you get a prompt about what does that book say about this, and how does it relate to this concept? And what other causes might it relate to? What problems can I solve with it? I've got to reason through this information, remembering what's in that book. So you might remember what's in the book, but you've got to operate inside it and reason within it. So this is why reasoning becomes a big issue. Now, if these models are capable because they have higher memory or more increased memory, we still, as businesses, have to scaffold the context effectively. And I relate it to Ashby's law. Ashby's law is a system's principle thinking, which says, a systems thinking principle, which says, if you want to solve a variety of problems, you have to have equal to or more variety of solutions. And I look at that in analogous to complexity. Businesses are heavily complex. They solve complex problems. So you have to scaffold context in the right way. You have to have the right documents in place. You have to have manuals that the AI has come read. They have to be connected to certain data sources. You might have to give them calculators, because these things are reason with words, not with numbers. Well, the interesting thing, though, is context engineering. Project managers need to do better than that with context engineering, even without AI. This is the problem, is that we often, if we're trying to train a bunch of humans to do things, the reason it usually goes wrong is because we haven't given them proper context at all. So it's actually not a new problem. Yeah, but it's the inverse that there are too many processes or standard operating procedures that AI's can't quite follow. So you can see the uncertainty space at the moment. So I think memory will be issued. That will be something which might evolve technology-wise. But equally, I think in the meantime, businesses have to understand how to engineer context better as we use the current version of models. And once you can do that and you know what AI can then do, then it will help you understand what your strategic work is that humans can do. I have to say that plays very nice in something that I picked up from your newsletters throughout the last year, which is the idea that actually AI isn't as much creating new project management challenges as it is actually highlighting challenges that we never fixed. It is, the surface of the managers. And this context thing. Because part of-- I mean, if let's say the current iteration can only hold a million tokens worth of context, part of the emphasis then must be on to get rid of information, you know, suppress information that actually doesn't help the situation because it's getting in the place. It can even go flip. So it's about-- and that has always been a problem for project managers cussing through the noise to the real stuff. So I think that's a really interesting point. You raise a very good point because some of the problems with it, people see it as a problem with AI. But while AI surfaces problems that are always been there, you just haven't been able to see them because you haven't had anything powerful enough to find these weaknesses. A good example of that is when people deploy co-pilot and then the co-pilot will surface permissions issues within files. So say there's a confidential file on your SharePoint where it's going to be on salary, you know. You know, nobody would have a clue that that was an issue because no one was trying to get into it. But the AI, if you now type in, what is the CEO's salary? The AI will surface that issue. That's the big risk. That's interesting. That's interesting. Yeah, other hot issues for 2026 or trends. Well, I think one of the big words of last year, in fact, it was the Colin's dictionary word of the year, was vibe coding and we did a video on it, Mike. And I think we're just seeing the beginning of that. The vibe coding tools getting better and better. Even within Manus now, you can do vibe coding and things like Claude Codes. And what I think is interesting, I'm finding more and more project professionals experimenting with vibe coding and I'm seeing some outstanding examples of what can be done. And those people will really become the kind of champions, if you like, in some businesses. And you know, even when we did our video, which was a few months ago, it had some failed teas. Even in that time, since we've done that video, it's got so much better. I've spent most of Christmas creating some apps and bits and pieces. It's so easy to do it now. It's getting better and better. And the other one that we haven't really touched on. And I'm not sure if this is 2026. It might be 2027, but it's about around robotics. Because at the moment, we've been very much focused on AI as a chatbot. But of course, AI is embedded in loads of things. And the big thing that is going to happen over the next few years is the rise of humanoid robotics. In particular, Elon Musk has just awarded themselves a trillion pound, but a lot of that based on the fact that he's going to make the Tesla robots a huge success. And you wouldn't bet against Elon Musk, wherever you think about it. So I think probably I'm going to stretch it 2027. But between 2026 and 2027, something really important is going to happen to all of us, including all your listeners and viewers. Remember that first time when you were sitting in a coffee shop and you saw someone with the iPhone for the first time? And everyone was like, "That's an iPhone, wow." And then everyone was excited and you're playing with it. And then the second time you saw it, you're still like, "Wow, there's an iPhone." And then by the time you saw it for the tenth time, it kind of just fed into your consciousness and it just became part of the furniture. I think at some point in 26 or 27, and a lot of people are going to see their first humanoid robot just doing something and they are going to witness a humanoid robot doing something. And that's going to be a game changer for people. It doesn't mean that it's going to become mainstream in 26, but that'll be that point when people remember the first time they saw it. And then you'll start seeing it in different contexts within construction sites, et cetera, et cetera. So I'd be interested to come back to this at the end of the year to see whether that happens. But it's only a matter of time. I'd be interested because I have a few things that take the construction site. Yes, we kind of know how a human laborer would carry a hod full of bricks on a construction site. But if you were going to design a tool to carry bricks across a construction site from scratch and give it robotic and analytic capabilities, you certainly wouldn't design it on a human body plan. That would be no way be the most efficient way to do it. And yes, I think there will be gimmicky humanoid robots, but I'm not sure that they're a solution to any real problem. We are not designed around, I mean, there are people who believe it, but we are not designed ground up in the human body plan. We are well if we are able to evolve to solve a number of problems, most of which are not problems that AI is going to need to solve. The only candidate is, don't forget, we've built a world around us over thousands of years, which is made for human humans. Yeah. So maybe it's that way. You're right, that's probably not the most optimal kind of form for a robot, but maybe at least in the first generation of robots, they are going to be very similar to humans, simply because they need to navigate a human world. But that then creates the issues. This is the psychosis, the pseudo-social bonds and these things, because you get on Cany Valley and you can trust them more because they're a lot more human-like, even voice AI voices at the moment say Minar. And so I think they're also, that was just two things from me that I think will be two big things as well this year. The first thing is the obvious one that's hitting everyone in the face. 2026 is the year of data centers. I mean, so you can ask yourself, did AI fail last year? Well, no, because even if the current models don't save the world, they have created a narrative where companies are spending billions and billions and billions in infrastructure, which therefore increases the energy gradient and says, OK, now we have a lot more AI compute that can go around to build new models, right? So just the fact that you have the energy gradient will allow us to build better, bigger and better models going forward, which has massive implications for project deliveries, like, you know, how do we deliver these things efficiently, how do we call them effectively, what are the regulations, how do you distribute resource to people, etc, etc. I think that's the big one. And the second one is these models, I mentioned at the very start of this call, that we're so focused on intelligence benchmarks. Well, if you want to understand how effective an AI is for your industry, your profession, within you create your own relevant benchmark. So I think we'll start to see industry wide benchmarks, there'll be benchmarks for project delivery, there'll be benchmarks for finance and legals and whatever it is. So I think that's another thing we'll start to see. Yeah, I want to come back to the data sense, because I think we've got a little bit in the project world, we've got a little bit of a conflict, because I think one of you mentioned earlier, the statistic, you know, not very many AI projects succeed. But one of the one of the statistics I found very compelling in the project world was actually one of the things that seems to increase project success is when they explicitly aim to address sustainability and regeneration. And that has been another big trend in the project community over 2025 and hopefully will be in 2026. And of course, there is a real clash between the need to create sustainability in our energy supply and also in the way that our planet responds to our use of energy. And yet we're using more and more and more of it with AI. So that's going to be a challenge. But I want to, I mean, we need to come to an end. And I want to pick what I think is for me a really interesting thing for 2025. And actually my prediction isn't what will happen, but my prediction is that we will move closer hopefully to answering this question. But I'm going to place that as a question to both of you. So from a point of view of a project manager, I see there are two directions that the profession can go in, which I'm going to kind of call using this terminology, continuous improvement and re-engineering. Because in the not just we had enterprise resource planning tools came in, which moved us from continuous improvement of trying to make the world a bit better by using new tools to do the things we were doing in a slightly different and better way, which of course is something we can do every day with AI as well now. But we also had organisations like the big consultancies saying, well actually you shouldn't try to use these big software tools to just make incremental improvements, but to actually rethink the way we do things. And I think that choice is in front of us now as project managers. We could improve our project management by automating the things we do and focusing a little bit more on the things we've not had time to do, which is clearly going to happen. You've both telegraphed that. The other possibility is that there is going to be a massive shift in the whole way that we think about what projects are, how we do projects, which is going to mean that the project manager role is going to be vastly different, would even say gone. I don't think we know enough yet to know where the some of the speculation about that is going to lead us to a massive transformation or not. But I think 2026 will give us more information about the direction of travel. But I am interested in what each of you think is that the direction of travel, is it a radical change or is it an evolutionary change? Well, if I can go first, if that's OK, you'll see. Because I think it plays really nicely to some research done by a guy called Professor Richard Suskins. And actually, those two scenarios you played out, I don't think it's an either or. I think they can both be done at the same time. I think you need to do the kind of just improving, the low hanging fruit, as I call them, the 20% improvement, whilst at the same time also thinking about that vastly different future for the project profession, whether that's two years away, three years away, five years away, 10 years ago, it doesn't really matter. You've got to start thinking about that. And it brings to mind the quote from Professor Richard Suskins, and he applied it to the law profession, but it's equally applicable to the project profession, which is how in the future will we solve the problem to which the project manager is our current best answer. And that's a really fundamental question we've all got to ask ourselves. But that doesn't mean we stop doing anything. We stop taking the small wins and the 20% improvement whilst we're pondering that. So I think what we're seeing in 2026 is both for those things happening. We'll hopefully see a lot more like I said before in terms of scalable, ROI driven, successful projects whilst at the same time more and more companies and institutions will start thinking about that very, very broad fundamental question. Excellent. Yoshi, what's your perception? I've touched on this before. I think project delivery is a profession which is heavily bureaucratic. And I think it's at the detriment to the system. And therefore I think because we have AI, we can decouple delivering projects with processes. I think we could start to personalize AI and be able to break through and give deliverables fast and having to go through a heavily over bureaucratic process. So I think in terms of a change, I think it will be an evolution, but I think there'll be a lot more local changes in how people actually deliver projects better. And decoupling is the first thing. And secondly, I think that for projects to be run more effectively with AI, I think we need to start to consider systems in complexity thinking. I think we need to harmonize the three things to actually understand our products and deliver them a lot better because projects are complex adaptive systems. They're dissipative systems where you have people working with so many different stakeholders. It's almost impossible to reduce that down to statistics, because things are always changing a project. And what AI can't do is grasp emergent phenomena. When something happens in the supply chain or something's late or there's a black swan event, you're not going to pick that up. And I think we need to be able to couple those three things if we want to understand and deliver projects better. Because I'll tell you right now, I can tell you right now for a fact, no one on the construction site is going to start to gather data passively next year. They're just not going to do it, which means how do you then get site data and productivity data? You still, it's not going to have it. So the data quality issue is always going to be there unless we can evolve from just using AI to doing agent simulations using complexity thinking. So I think there's an opportunity to be had with those three things. That's interesting. And what both of you have said, I think, is actually consistent with my view. The way I'd express it is, I think the project manager role is less likely to change in the short term than the PMO role. I think, yes, yes. Fundamentally, we invented PMOs in the 1990s, or we've vetted the word PMO, the people were doing. There's jobs, the project control jobs, and the project support jobs. And the assurance jobs. But we created PMOs to solve a problem of how do you take the functional planning and monitoring and risk evaluation, the technical side of the project manager role, off the project manager's plate so that the project manager could spend more of their time on the strategic thinking and the human interaction of stakeholder engagement. And I think that's the low hanging fruit. For, first of all, automation and continuous improvement. But then gradually, people will rethink how that can be done. And we'll start to see effective. When I was at Deloitte, we had this concept of the PMO in a box, which was a basically a set of tools you could deploy. A team could deploy to create a PMO with a client. I think, fundamentally, of the PMO in the box of the future will be basically a matmini type. You know, there's still a box conceptually that you. Head of your time. Plug into your client and say, "This is my set of GPCs. This is my set of agents that I've created. I'm not bringing to you a set of word templates and Excel templates. I'm bringing to you a set of pre-configured agents with pre-configured choices of prompts. And that will go into your system. It will find your data and it will analyze it. And it will produce plans and things." And that's my view. And not playing normal. I guess we didn't retouch on this. You have all this data and the AI can do it. And the best thing about it is we don't even care how that information is then communicated. You're able to look at that internet three before and how we care about diminishing culture. Well, if you have agent-to-agent protocols, James is the project manager and you have an AI which can report to his AI. The content is then written for an AI. So it doesn't matter how well that information is written as long as it's there. So I think that will definitely change the run. Agent-to-agent protocols will also be a massive thing. And presumably at some point those agent-to-agent protocols will devolve back to the most efficient form of communication for an agent which is back to ones and zeros. And they'll have to translate it for us just as if working in Japan on a project. Someone will have to translate Japanese for me. Yeah. They're just another team member with another language set. And we actually saw that happen. There was a clip. I think it was this year or last year, so 2025, where they had two AI's talking to each other. And that, I don't know if this was simulated, but after a certain amount of time, one said to the other, well, let's not communicate in this language. And they mean there's a whole new language to talk to each other because English wasn't efficient. So I don't know if that might have had to be a little bit of AI about itself, but it's quite an interesting kind of prediction about where things might go. Yeah. Okay. Well, I think that's a good place to end. Thank you both very much. Would one of you like to tell me where people can learn more about Project Flux and how they can sign up for your newsletters and watch your podcast or listen to and watch your podcasts? Yeah, absolutely. Everything is on projectflux.ai, so very easy to remember. And if you go there, you'll have links to the podcast where you can sign up, whether that's with Apple or Spotify or YouTube. There's links to the newsletter where you can subscribe. And then there's all of our blogs there and other bits and pieces that we're getting up to. So that bit, that's kind of the main page to go to, you'll find everything you need there. So Project Flux. Well, Yoshi, James, thank you very much. This has been fantastic. Thank you, Mike. Thank you, Mike. Yeah, awesome. This was awesome. Excellent. ♪ Project Flux is here ♪ That's a wrap for today's episode. But before you go, exciting news. Project Flux is teaming up with Project Management Global to bring you the latest in AI-driven project delivery and essential management insights. Through this partnership, expect weekly AI insights, practical tools, and expert thought leadership, ensuring you stay ahead in the evolving project landscape. Stay tuned for exclusive content, doing research, and strategies to supercharge your project outcomes. You can learn more at www.pm-global.co.uk and be part of the future of Project Management. See you later. Thanks for joining us, you've been listening to Project Flux. Don't forget to like and subscribe on your favourite platform is the simplest way to help us grow. Until next time. Hi, everyone. Everything you've just heard is shared for discussion, not advice. Opinions are personal and don't represent any employer or organisation. Use your own judgement always as every project and situation is different. [Music]