Ep 65: Christmas Special, 2025 Wrap-up

We put our reputation on the line and asked an AI to grade our 2025 predictions. The result? A solid A.

But we aren't resting on our laurels. In this festive special of The Innovation Room, the Geeks senior leadership team looks ahead to what 2026 holds for business leaders. We explore why modular neural networks will change how we update systems, why AI agents are becoming the new digital operations layer, and why human-centric AI literacy is the investment you can't afford to skip.

This year, we also introduced a new segment to our annual wrap-up. We debate which business behaviours earn a spot on the ‘nice’ list, like treating tech partners strategically rather than transactionally, and which land you on the ‘naughty’ list (we’re looking at you, restrictive quotas).

Join Matt, Dehan and Lindsay as they unpack these insights while attempting to construct a ‘minimalist’ gingerbread house. 

Transcript

0:03

Welcome to the Innovation Room, the space we explore the application of technology and AI in business, the practical and sometimes philosophical ideas shaping our future. Well, hello everyone. Welcome to the Innovation Room. Um for those who have watched the last few years, we have a tradition of getting our senior some of, well, some of our senior leadership team in a room and reflecting on highlights and reflections of 2025 and looking ahead at 2026. Um and given where I guess, creative people. Our goal is to do something fun and Christmassy at the same time. You can see Matt here took a head start on our gingerbread house because last year we were told we were too distracted while we were doing this. So I have here Matt, Exec Director and CTO Geeks, Dionne, Ops Director at Geeks, who you may know. So today then, if you guys are okay, we'll get started by reflecting on, and this is something new that we've introduced, we're going to reflect on our 2025 predictions, but I don't know if you guys are ready, but we're going to get graded from AI on our 2025 predictions. So bear with me for a second while I get our AI to grade us. Hi there. Can you please grade us on our 2025 predictions and say what we hit spot on, what was partially correct, and maybe what hasn't come to fruition yet? Absolutely. Let's do a bit of a grading rundown. I'd give you an overall grade of an A for your 2025 predictions. Here's how it breaks down. What you hit spot on, you nailed the idea that AI would shift from being something experimental to something that everyone just expects. Also, your prediction that small productivity gains would stack up into big returns was spot on, and you were absolutely correct that AI would become a huge factor in enhancing customer experiences. Those were definite bullseyes. What was partially correct, some of the more advanced predictions, like the idea of multi-agent AI systems behaving like teams, are starting to emerge, but they're not fully mainstream yet. So you were right on the direction, but it's still a bit of a work in progress across the board. What hasn't quite come to fruition yet, one area that might not have fully materialised is the complete democratisation of AI, to the point where absolutely every small business is on board. While a lot of progress has been made, some smaller companies are still catching up. In summary, you were mostly right on target, and the few things that haven't fully unfolded yet are simply stillable. Overall, it's a solid A. Not bad, guys. Ace. Ace. Cool. Sorry, if you heard anything funny there, we knocked over our microphone. Full disclosure. So, is there anything... what are your reflections then, guys, on our 2025 predictions? I'd probably start with, I think... Yeah, we had a prediction around the multimodal models that used to come out, or what we thought would come out a lot sooner. We really only saw a lot of that coming out towards the back end of this year. So I really thought that would have happened a lot faster last year. And what do you guys think, what do you think slowed down our, you know, and I think it's a partially correct, right? So just for clarity, what we're talking about is AI agents and the fact that there would be teams would be augmented with multiple AI agents working together as part of the team. So what do you think are the main reasons why that hasn't it's a work in progress and it hasn't come to fruition as much as we might have anticipated in this year? I think the biggest challenge in that domain is controlling and coming up with a protocol for AI agents to work with each other because you want to give them freedom to be able to talk to each other and even create new AI agents dynamically but with power comes responsibility and we don't know yet how to properly control AI agents in a way that we don't take away their flexibility, but also we keep them on leash in terms of if one of your AI agents starts creating a million different agents, that's going to cost a lot of money and also a process that should in theory take a couple of minutes, might take hours. It might cost businesses a lot. So governing the AI AI communication with each other, AI models is a little bit challenging and I think this is something that a lot of research is being done on and I think in 2026 we will see some really good progress in that domain. I think building on that, my observation is cultures are working really hard to get ready but they still, a lot of them weren't ready for what needed to properly, when you have multiple agents, as you said, the accountability, the data access, data readiness, I think there's a lot of work to be done there more foundationally for businesses. So I think that's what slowed things down, is a lot of businesses are still working to get that in order before they can really start using more scaling and adopting more effectively. I would even go so far to say there's an underinvestment in the adoption phase of these projects. So you see a lot of experimenting with it. But for it to actually scale, it means there has to be an investment in adopting effectively and putting the process or, I think Dionne, you mentioned earlier, rethinking process to make it work. There's one thing to add also there is also just kind of the human behavior that goes with it. I think when companies introduce AI agents into the business, almost the user expectation is it for it to immediately save me this massive amount of time, but they don't... that adoption process of being aware that there is gonna be a stage that you have to supervise it, and your immediate productivity gain won't necessarily be as much, but if you spend enough time going through that adoption phase properly, you've probably seen companies give up on it too quickly, or individuals give up on it too quickly to say, Oh, I'm not getting, I'm not saving 20 hours a week introducing this agent. by only trialing it for a couple of weeks and then they give up on it. I think that leads into perhaps the education piece. So one of the predictions that we made that was accurate was a lot of investment and increased AI education in 2025. And I think what you're describing, a lot of that stems from people aren't quite there in their AI education and understanding what is actually possible with AI and where maybe there's some over-inflated expectations with what's possible and therefore designing too big of projects. So that was another place where we were accurate with our predictions, which is that little quick wins and generating ROI up of those were going to be much more successful than those designing the big bang kind of AI projects or plans. And that's definitely been one of, I would say, our most accurate predictions this year was we're seeing lots of smaller, quicker wins, implementation, and then therefore it makes the adoption phase easier and the education phase easier. So let's move on to a new addition we're trying this year, which is our Naughty and Nice list. And in the meantime, let me give you some decorating candies and Dion. Why didn't it didn't matter to me? Sometimes it doesn't work. Well, it's still teamwork. We're talking while you're building, but yeah. Last year our house was kind of falling apart and then our office dog Chief Mischief Officer managed to get some gingerbread out before we were done, so that was fun. So this year we're going to introduce a section of this called the Geek's Naughty Nice List. So I wanna hear, and I'm gonna start with Matt, what type of businesses are you seeing either on the naughty or nice list? What would you do? And you can pick any type of business in our space. We could maybe be cheeky and talk about a few of our clients, things like that. I think businesses that It's related to one of our predictions for 2025. Businesses that don't focus on incremental productivity are on the naughty list for me because there are a few challenges with that approach. One is to wider understand how tempting it is to focus on big gains. At the same time, AI models are getting better and better on a daily basis. We need to take actions, but at the same time, we don't want to spend a lot of time and resources on implementing something in the system that tomorrow a new AI model is more intelligent than the current models can do that easily. You don't want to wait forever to get to the smartest AI model. So what is this? The naughty or nice list? And what is it? It's the naughty list because a lot of clients focus on huge gains, so 30, 40, 50%. And that means you might spend a lot of time and energy on implementing something that you can achieve using a smarter AI model. So doing incremental productivity gains ensures that you're not wasting any efforts. I would say the flip side of that, the clients that I would put on our nice list this year, both from strengthening our partnership but also seeing the impact that they're bringing to their business and their industry, have been the ones that treat their relationships as a strategic partnership. And They plan to do continuous evolution, so there we have a roadmap together, we have initiatives that we're working towards, and we break down. productivity gains and to win. So they're not thinking this is a one-off project, they're thinking that we're going to evolve and augment our team with AI, evolve our client's hyper-personalized experience, make better decisions, all of those things with AI. So I'll do a few shout outs to a few of our clients, well I'll pick one in particular because they've been with us for 14 years now, if I'm not mistaken, Search Acumen. They've had this mindset really from the start, even before AI, but we've been working with them with AI for, what, five to seven years, something like that. And they're such a great example of their culture was more ready, we've worked together with continuous roadmaps for years, and our engagement and our team is really one team. So they really get how to make these things a success. So they go on my nice list this year. Anyone else that you guys would put on our naughty or nice list? I don't want to name, but... Just do it. There are, again on my naughty list are the... Matt loves the naughty list. Focus on using AI in production. So there are different phases that you can use AI for. You can use AI for when you're designing a product, or you can use AI when you're implementing a product, or you can use AI when you're using that product in production. And a lot of people think that AI can only be used in production, and that's the only form that means you're using AI for your application. Give an example, please, for those that don't know what it means to use AI in production. Imagine you want to design a system, You can brainstorm with AI and get AI to help you design the system in a better way. You can use AI to generate code for you, and that means that AI is doing the development for you. And then you can use AI in your application, in production, so sentiment analysis or generating content or tasks like that. They are the ways that AI can be used in a product in production, so using AI in a business doesn't necessarily only mean using AI in production it can be in different formats and in different stages of delivering your project. I think one of our ones that's definitely on the nice list is clients that also look at it not just from a tool perspective, but from a workflow perspective. So not just thinking of what tools can I get or what process can I just automate. It's around thinking more around having maybe less applications, but the workflow works better and redesigning those workflows. for the tools that are available. And on the converse side, the naughty list is probably more people that think of AI as just a tool. We use AI, we use ChatGPT or Claude or whatever the case may be, but it's not embedded in workflows and there's nobody that takes ownership and accountability of it. I think the workflow is a really good point because AI is a black box and if you want and we are not still comfortable with the idea of not being perfect. And with AI, if you want to get a very high degree of accuracy, you have to put a lot of effort. Whereas with a human-centric workflow, you can verify the output of AI in different stages of your workflow. So if you have the different stages that you're using AI for, if at the first step, AI makes a tiny mistake, all the way through the workflow, you're going to have problems and the problem gets bigger and bigger. But if you pair AI with humans and get humans to verify the output of AI and then use that, because with AI we get the scalability as well, so you cannot use humans in every single step. But what we can do is to use the feedback that humans provide as a training for AI to get better and better so that our Employees don't have to give feedback for the exact same situation. And at every step you're verifying how AI is doing and the input of the next step is going to be a valid input so that AI is not going to make mistakes because of the previous steps. I think that's a good point on, you know, how we said when we were looking at our reflections from 2025, there's a big expectation of people often from AI to fully automate full workflows. And I think that's when we say, and it's probably a good way to bring context to what we mean by breaking things down into smaller pieces, because like on the naughty list would be the clients that expect full automation straight away and don't get that it's a journey. But also you hear a lot of things like, shouldn't AI just learn? Which shows that education piece is so important. So you're explaining very well how you break it down so you can teach AI, just like when you teach skills to people, or you need to master skills, you don't just keep it into one broad role and expect that they understand. and all the skills involved in that role, you break it down into contextual pieces. So I think that's really helpful. We have the last naughty one, which is, I know a couple of clients that will maybe share the sentiment, is the swear word of the day is probably quotas. And I know we try and push a lot in terms of what we can do and how our clients can use the system, but I think over the last couple of weeks we've had some quite stressful conversations around especially quotas around tokens per minute and so forth being a big restriction on some of the tools. So those that implement those quotas definitely on our naughty list this year. Yeah, they're being so polite. Basically what they're saying is Azure and OpenAI will both go on our naughty and nice lists this year because the great thing is is we've seen the models advance so much and what's possible is now increasing what you can do with AI and the different use cases and the way different industries can benefit from those models, as Matt was talking about earlier. But then the dark side is there's a lot of limitations and restrictions on how AI works for those of you that haven't been working with AI yet, is they give you amount of tokens and quotas to do that. And based on, depending on what the business case is, They might only be able to, for example, start, like you kind of have to hope that those quotas will increase, and you're only able to start with, you know, so much of a transaction volume. So it kind of slows down the scaling aspect of AI. So that is sort of a beware if when you engage with AI. That's one of the things that we explore in our discovery and design process to work things out. But also, all of us are kind of at the mercy of these models because depends on things like how they're operating, but there's also macro factors that we're not gonna get in today, but like energy and things like that, that there's some big challenges and opportunities that will determine how quickly we're able to scale AI. But it's here and it's happening, and it's happening quickly, and then it'll just be how quickly will these other more infrastructure pieces support that growth. Anything else for a naughty or nice list? Are we ready to look ahead at 2026? Nothing problem. Okay, so let's move on to 2026 AI predictions. So I think we got an A, or when I saw AI, it gave us an 8.5 out of 10 last year. So let's go for 10 out of 10 this year. And technically, one of the things, if I start us off, that we're supposed to see in 2026 is One of my predictions is that the decision making will be better and should be significantly better, both at executive level but all through the business. So I'm hoping that us using AI will make our insight and decision making better for our predictions for 2026. So what do you guys think? I'll probably start with one. I think AI agents is going to really become the norm for almost a digital operations layer, where a lot of your SMEs now are going to have this whole layer of agents just always running. And I think that's going to start becoming the norm through 2026, where you might go home for an evening, but you might still have an AI triaging inboxes, or sending out invoices, or reconciling blank statements, whatever the case may be, I think that's going to be a lot more embedded in specifically smaller, medium type businesses through 2026. What about you, Matt? I think in 2026 we might see modular neural networks in a less nerdy way. It's So neural networks have been built based on the human brain, but they don't function like the human brain. So in your brain, you have got different parts that do different things. And in theory, if a neural network is built like that, you should be able to replace the module. So the thinking part of the neural network, you can replace it with a more advanced one. And that means that you don't need to retrain the whole model. So let's bring it even so I love when this is the fun with working with geeks so let's break it down another level so let's take it into business context think about how our clients might experience that. Our clients will it's so models will get better and better much quicker Because, as I said, because you don't have to retrain the whole thing, you don't have to wait for the entire cycle of a model to get the new version, but also it means that we might be able to have a... You might be able to create a module specific to a company and then plug that into a neural network. Cool. So if I were to, and I'm going to validate my understanding in real time here, so I might be way off, but if I were to break that down, right now, for example, some of our use cases and things we're doing with clients would involve, for example, processing documentation and then you know, making decisions based on that documentation, maybe looking at, and there might be some complexity with things like formatting, perhaps, or other issues. What you're saying, if I understand correctly, is while a model is handling all of that, you could retrain the element where formatting is causing issues. Is that what you're saying? In terms of, so it's more focused, whereas what happens now when you're debugging or trying to work with a model is it could throw off other parts of the model that were previously working, and it makes it a more, challenging R&D process. So you can do it as a human so you might have a developer that is very creative but it's not very risk aware. So you can replace the part, you can identify which part of their brain is not functioning and they're not risk aware and then you can change that and make them keep their creativity while you add the risk awareness to their brain. Good example. Awesome. So I think one of my other predictions is AI literacy has to massively scale for, I think I told you one of the things slowing down when we talked about the multi-agent orchestration in businesses. So I think this year will be a huge year for investment in AI literacy in businesses. And there's a few reasons why I think that. I think human nature, even if you look back in history, has been to try to create, when there's something new, trying to start first by keeping very specific skills for that thing. So that's why you might, I don't hear it so much anymore, but at first, you hear things like, We need to hire prompt engineers. Prompt engineers is a role. And then you see people putting it on their LinkedIns, but that's a thing. But if you go back to when computers first came on the scene, That's almost the equivalent of, and I'm not sure who was around for the days, where you had computer operators in a business, you know, they would put computer people in a room, and if you wanted something done on a computer, you would give them a task, and that's how they thought of it. Now, everyone in a business is able to use a computer, and I think that that's one of the missing pieces that are underestimated to make the AI adoption piece a success, because there's an element of collaboration with AI augmentation that is underestimated and everyone needs to rethink how they work with AI to get to better decisions, to create more intelligent insight, et cetera, because that will be the bar. So businesses will be forced now. It should have already started, but businesses will be forced this year if they want to get any value add out of their AI to really invest in that. So that's one of my predictions for 2026. probably have a prediction of a barrier that we'll probably see in 2026. And I think one of the biggest barriers a lot of the time when we work with clients and especially AI models is context. and a lot of our processes and workflows are actually function based or department based. And for us to get the most out of AI, we actually need to look at redesigning those processes and giving AI context from different areas of the business. And historically, as businesses and humans, we've had very distinct, you know, this is finance, this is HR, and they're two very separate contexts that you operate within the same business. And I'm thinking that a lot of that is going to limit the ability of AI to be more context aware of the organization. And maybe one of the barriers that we're going to see in 2026 is when we recreate processes is the linking of information or the linking of models between departments, which will be interesting to see how people start navigate that because we've been hardwired that way from the start of work. What about you? I think in 2026 we have we're going to have a challenge and that is related to accountability for the output created by AI. The new trend that I'm seeing is that you get a lot of smart content from a lot of people but when you actually talk to them about the content they are not they don't understand the content so everybody it's easy to get AI to produce something for you but if you don't know what AI Has created, and if you are not familiar with the domain, you don't know if you've asked the right question or have given the right context. So, the example Yan mentioned, you can expose an AI agent to a lot of data sources, and if you just ask AI to do something for it. for you, because it has access to all of those resources by default, it's going to get all of that information. It may or may not be relevant. to what you're asking, but we need to make sure that we are using AI doesn't mean that we have to stop learning and we need to work on our communication because AI is essentially a mirror for us to see how... bad we are with communication because when we work with people, usually people blame other people that you haven't understood me. But AI is smart enough to understand you. If you're not communicating, it's not going to be able to deliver what you're asking for. Yeah. The old garbage in, garbage out phrase is still very relevant here. That's an interesting point because I actually think, don't quote me, but I think one of our 2024 predictions was around org charts implementing AI into their accountability structure. That was one that I would say actually we were way ahead of the curve on because we're still not seeing, like we've done it for a few years now and some of our clients we work with we've been encouraged to do, but we're seeing, sorry bear with the sirens, I think that's come through, I don't know if you can hear that, but I think we're seeing a lot of slow changes of thinking of how they safeguard that accountability. And I guess that's the piece I think we mentioned in reflecting on 2025 with the multi-agent situation being slower to adopt. I think that's a governance piece. So some people aren't starting or going fast enough because of that governance piece and they're aware that that could be an issue. Or some people go full on in without putting any guardrails around how they do that safely and responsibly and then they stall or get to a point where they need to pause. So I definitely think that that's something hopefully that, you're right, I think it still will be a challenge, but hopefully we see progress on people acknowledging that that's something they need to invest in. It kind of leads me nicely into one of my predictions as well, which is that I think this is gonna be a year where human capital in a business, especially people that are really engaged and collaborative with AI and AI output, are going to be one of the most valuable assets people have in a business. So I think people that are capable of engaging with AI, communicating properly, getting the right outputs, that is going to be a very, very valuable asset to businesses. But then the other side of it is, I think, I mean trust is always a differentiator in business, but because there is a lot of mistrust with AI outputs and but the expectation of what AI can do is still, there's still quite a, we're not quite there on the hype cycle, it's still quite high. I think there's gonna be a real tangible or intangible, rather, asset for businesses that can create a trusting experience with all AI outputs and that their clients see that value in their endpoint. So I think that will be an interesting one. Anything else that you guys would like to highlight? There's probably an exciting and nervous one for 2026 is humanoid robotics, AI, that combination. We've not really seen that available commercially, probably towards the latter part of 2026. I know there's a few companies with Tesla and BYD looking at bringing a few things out. So there might be a whole new, scary, exciting wave of of AI being implemented a lot more, a lot different ways to what we've seen in the past, which might be quite interesting to see. Yeah, there's already, especially in like manufacturing and logistics, you see a lot of robot cobot especially in the big, it's still in the big corporate level, isn't it, really? There's not many, there are some entrepreneurs bringing it into their business, but we aren't quite seeing it, I think that was one of our partial things too, wasn't it, that we might see it more mainstream, like coffee shops. We know, you know, your coffee shops are having barista supported, and there is technology advancing there, but how ubiquitous it'll be in our existence will be interesting to see how much that progresses. That is definitely an interesting one. And then I guess for me, I'll be interested, I have a hypothesis that speed will become, like the speed in which expectations are managed is going to, for all customers in all industries and sectors, whether you're B2B, whether you're B2C, the speed in which you get data and insight, the speed in which you receive your product or your service, the speed at which you receive communication just generally, and therefore the speed internally to decision-making in business is going to keep compounding this year. And as AI becomes more prevalent, I think the expectations are going to continue to rise. So I think this year will be a year that I would advise business leaders to really reconnect with what are my customers and what are my customer journeys and touchpoints, and my team, how are they delivering on those touchpoints? and what can we do to help them make better decisions faster, because I think this will be the year. where it'll be either the differentiator to keep businesses going, or it'll be where you start to fall behind and become a leg guard in your industry. So that's one of my 2026 predictions. Anything else? No? Well, I think we've done a very minimalist, Japandic maybe is the style. I don't know, gingerbread house. So I think we have our work cut out for us. We engaged too much in the conversation, not enough in our gingerbread house. Can we do a stress test? Yeah. See if it, earthquake. We gotta bring it over. Oh, there we go. So we've made it work, we just haven't made it look good. I just want to see Matt take it off the base. All right, guys. Well, thanks for joining us in the Innovation Room. Merry Christmas, happy holidays to everyone, and see you in 2026. Thank you.