Latest / Under The Pallet Wrap / Episode 13 - Before You Invest In AI For Your Supply Chain - Listen to This
Transcript
- Dan Culverhouse: Hello all and welcome to this week's Under the Pallet Rock and it's a slightly momentous occasion for Keith and I as you may have spotted but you have a guest. Vince, introduce yourself. Vincent: Hello world, my name is Vincent. I'm here because I know this gentleman for a little while and I work in the similar industry. and I'm here to talk predominantly about the subject today which would be what I've spent the last few years involved in which is AI in and around supply chains. I myself am a retailer of many years since I was 15 and three quarters which got me into retail technology which got me into supply chain, supply chain technology and brought me here and I think I started working with AI powered systems or systems that claim to have AI around about 2019 or 2018. So it's been a little while. So even a little while before before ChatGT, ChatGBT was a twinkle in anyone's eye. So hopefully that Dan Culverhouse: we really knew what it was about. Vincent: will give me a little bit of perspective. Dan Culverhouse: Cool. And we had a bit of a Keith: It certainly does. Dan Culverhouse: conversation about it before, Keith, didn't we? Which was more about how AI impacted careers and supply chain. But it's something we have quite a lot of interest in. So we thought, hey, let's talk about it some more. And let's talk about, you know, how do we embrace AI? How do you make it actually help us? And so great to have you here, Vince. We're very, very pleased. And Vincent: Thank you. Dan Culverhouse: I think really, Most people have played around with AI, particularly since chat GPT became a thing. I think most people have just asked it, what should I wear? And can you write my emails for me? And Vincent: Hmm. Dan Culverhouse: things like this. there are all sorts of end of the world scenarios that people talk about because of AI. And yeah, we just thought it would be a nice idea to sort of. Maybe just sort of sit back a little bit and think about what AI does bring and what really is it about and how might it really help us. Vincent: Well, when we were talking before the show, I know the question was around automation. And so I'd be very good to talk about that because I think there's a lot of people are on a precipice of what do I do with it. I don't think it's worth going back over all of the usual conversation about AI. What is it? What it isn't? I think a lot of people are now starting to form opinions. A lot of them are maybe not correct though, on the various types of intelligence, what they do, what they can't do, versus what they're claimed to be able to do. And I think therein lies some of the problem. So when I talk to some of my customers, some of the things I hear, which are quite common as well. The exec are asking us what we can use AI for, which is never really a good starting point. A better question might be how much does it cost? Or maybe even some parts, can someone from the technology department pop upstairs and tell us what it is? That would probably help as well. So everything that... that I'll talk about, I'll try and keep within certain parameters of what we mean when we say AI. Is it narrow? Is it general? Is it what people sometimes call super intelligence now? What its methodologies are. So are you using it for statistical purposes? Are you using it for predictive purposes? Are you using it for generative purposes? And then. The one that everyone talks about now is, are you trying to give it its own fingers and thumbs and give it agency so that it can do things? And what people call, what people call, agentic AI. So I'll generally talk within those terms of reference and I'll try and relate anything that I say back to some element of that. So it has a little bit of an anchor point in what we mean rather than just saying. AI because it can get a little bit, you your eyes can roll into the back of your head and you can get a little bit when someone says AI. So we'll try, I'll try and be quite specific. Keith: Absolutely. So general business, the way they might approach AI, you know you've been involved for some time. What's your interpretation of each of these buckets Vincent: Hmm. Keith: as you've called them? Where do see the Vincent: Well... Keith: future? Vincent: Well, when I started, by the way, I should probably say I'm not a genius. I'm not someone who spends time trying to tinker with a neural network, fiddling with weighting parameters and doing all the data science stuff. But I have worked with some very, very clever people who've done a very good job of giving me some sort of degree of education and were very kind. to my intellectual curiosity, if you like, to help me along understanding some of this. But what I see is predominantly disruption. It's incredibly disruptive. I come in my latest iteration, I come from the SaaS software business. You might have heard of something called the SaaSpocalypse, right? So that's one of the big changes we had in technology going all the way back to, there's a computer, there's the internet, there's mobile technology, there's the cloud and subscription services, yeah? And now we're in this next revolution. And what I saw is an awful lot of disruption. Mainly to that latest business model of soft grows and service SAS which is people in buyers So I've I predominantly work with work with retailers So people in retail technologies and do I really want to sign a contract for the next five years for a few hundred thousand a month? For this application right now Because I don't know really Dan Culverhouse: Thank Vincent: whether I can do this in-house yet or not. I don't know what AI can do for me. know, my last sub-stack, one of my last sub-stack posts was lending a little bit on that, you know, what have the Romans ever done for us? What has AI ever done for us? And we don't really know yet. So. There's a lot of hesitancy and lot of disruption around the market because... Do I need to sign that contract? Do I need the contract I've got? Where am I with renewals? What should I do? Because the business case before was quite straightforward, right? So for SaaS systems, you took all of your expensive in-house IT that you paid an awful lot of money for, for server rooms somewhere on a property, to maybe run by a third party, you were paying them a lot of money to do God knows what maintenance and support and all the rest of it. Fine, do away with all that, move it to SaaS now. I know exactly what I'm paying per month, right? Well now, that could go the other way. Right? So now, now it could be well, instead of, instead of not paying any attention to that, it's just happening on a cloud somewhere, well maybe now I just need two or three developers and they can vibe cord me an app that now does a particular job I needed to do in the business that sits on top of my data platform, if I have a data platform. That's quite lightweight And I can maintain build it not license it deliver it and look after it myself Now that sounds great, but there's an awful lot of implicit assumptions in there Do you know what you're trying to achieve do you understand you know what what your business case is? Is your data good enough? Is your data in the right place? you have the right foundations in place? all that sort of stuff. it's easy to send. Keith: This is what I was going to mention because I've heard almost every business I speak to is they want to keep in touch with AI. And I've got a continuous improvement background. So I'm obviously all about root cause and improvements that actually add value. I do think that a lot of people are talking about AI as a game changer when they don't even know what their problem is. They don't even know what they would actually use it for, what value it would bring. They think it can save them time, but they're not looking at maybe the whole picture, identifying where it could help them. Vincent: Yeah, there's a couple of major issues that I think of and this may not be a completely exhaustive list, but I don't really think people understand how much it costs. They Keith: Hmm. Vincent: don't understand the mechanics of, especially if you're talking about large language models, so for being specific and about how tokenization of that machine conversation is going on, how that's being charged and what it means. to their budgets and there's some stories recently, I just forget the name of the company, I forget who was now and I shouldn't guess in case I get it wrong, but I think they burned through their entire budget in the first two months of the year or something like that because they didn't really understand the cost structure of using a large language model. So there's understanding, yeah, Keith: Yeah, all the parameters of it. Yeah. Vincent: so there's understanding the cost of it. And then the second thing, arguably most important is to your point, what do you want it to do? Do you want it to replace someone? Do you want it to solve a particular problem that your team haven't been able to solve for a long time? Or do you want it to make everything better? And how do you measure that? And that's really the third problem. Having worked around systems for a little while and looking at people going through digital transformation, one of the things companies are most reluctant to do is measure anything with any real teeth. Keith: For me, measure is one of the most important things coming from where I come from. When Vincent: What? Keith: I'm a believer that you can measure anything, you can measure absolutely anything. Vincent: You can, and we could talk the same language about define, measure, analyse, iterate, control, right? Keith: Yeah, well if you haven't done the definition, it's for me, it's hard to justify a solution unless it's absolutely obvious, if you cannot measure the benefits. Vincent: But how do you define a problem for a solution that you don't understand? So. Keith: Yeah, you can't even define it. I'm not sure whether... You know, I don't... Obviously some people have had success stories with AI. Vincent: hit. Keith: I mean, I'm keen to hear some of them actually because we don't tend to hear... We hear the bad stories like the business who got rid of their entire customer service team and then brought them back because they had so Vincent: Mm-hmm. Keith: many complaints. That strikes me as strange just because how do you know, you must know how much customer service costs, the impact, the benefits to just replace them. There must be a... Vincent: I think that's one of the easier ones. I think, who deployed some automation or AI technology into their customer service area and were able to redeploy people into what they, as a business, considered were other value add activities. So their benefit was they were taking people away from work that could be automated and giving them something you could argue less mundane. to spend their time doing. And again, you can get very quickly, very quickly you Dan Culverhouse: Thank Vincent: can get to automation before you've gone through any of the other stages of introducing this technology into your business. if I think about, if I reflect on the question that was proposed when we were talking originally about this conversation, it was, should I automate next year? My first question to a company that says should I automate next year is have you augmented already? Well, what does exactly, what does augmentation Dan Culverhouse: What do you mean by that? Keith: Yeah, I mean we had to... Vincent: mean? Well, are your teams getting better at what they're doing because you're able to deploy newer technology without cutting them out of the loop altogether? So my premise is that before you get to automation, step back and there's a step for your business, which I use the word augment. Dan Culverhouse: Yeah, that makes sense. Vincent: Yeah, so your people are using the technology to be better at what they already do, what you already measure you want from them. Now, I think there's a little truth to say that in most businesses now, your employees are using AI. Why wouldn't you if you want to make your emails better, if you don't want to so many hours at the desk with PowerPoint, which was obviously what I spent all the time of my life that I've missed with PowerPoint. Yeah. Keith: Yeah. Vincent: And now I don't have to touch PowerPoint at all. Now I think it's worth every bit of resource in the world already just for that. you see your people are already doing it. But I would argue they're probably not doing it consistently. They're probably not doing it measurably. They're probably not doing it with the same tools. And they're interpreting the task that they think adds value for AI themselves based on their focus of what they need to deliver into the team. Dan Culverhouse: Yeah. Yeah. And I think one of the things that always comes back to me whenever they're talking about this or if I'm ever trying to do something through AI and I'm just by no means any expert in it and I'm basically generally using it by trial and error to see what I can get it to do. I think one of the biggest challenges that you don't actually know what it is doing or you don't know how it's coming to the conclusions that it is. Well, that's my impression anyway. I don't, Vincent: Mmm. Mmm. Dan Culverhouse: you know, one of the things I think about almost like get into a point where you can trust what this is doing is I personally, I'd want to be able to understand how is it reaching these decisions, which I don't think is always that easy to get to. And I don't, is that just my naivety around using them or is that just the way that it is? Keith: I would argue, I would argue, Dan, that it's far from naive that you're questioning because you're using your expertise in using the tool. You're questioning it and you're trusting your own judgment against the efficiencies that you're getting. I'm not sure, I mean, I haven't witnessed people just churn out stuff that, I mean, obviously, I think you'd guess if someone was just chucking out reports that were... that didn't have any basis behind them you'd like to think Dan Culverhouse: be a naive guy. Keith: being experts in our fields. Yeah. But I'm not sure how many people aren't, you know, I think people see it as a game changer and a time saver. But whereas we are experts in our field and we can sort of see how it can help, how it can't, when you would use it, when you wouldn't. Do you think there's an element that would just use it, whatever the case? because it's quicker to use it for reports and therefore they use it. Vincent: But that's already happening. Dan Culverhouse: That's what I trying to do. Vincent: My argument is that's already happening. Dan Culverhouse: Yeah. Vincent: And Keith: Yeah. Vincent: if you're an executive in a business, my fairly solid bet is that you don't know to what extent. And you're certainly not measuring what it's doing to your business. You're not measuring what it's doing to your productivity. You could go from massive extremes. You could have someone you're a reasonable stack to. And they're doing most of their job from a telephone in the Maldives. because they've cracked it for their particular job, what they need to do. That's why they're never in the office in a meeting anymore, because they don't really Dan Culverhouse: they don't really exist. Vincent: exist or more. Now, I'm being extreme, but my point is, do you know what of your company's IP is being uploaded to a cloud service somewhere? Because someone in your team feels the pressure to shortcut the time it to process a contract. Dan Culverhouse: Yeah, and you know AI is using information that's out there right and it's like well if you don't know what information it's drawing on to reach its conclusions, how can you Vincent: Well, that's a different thing. if I reflect on, using AI to predict, so we're working supply chain, so let's say using AI to predict demand. Do you know how your brain works? Not really. Because Keith: night, the mystery of life. Vincent: when you hear a song and it makes you cry, you don't really know why, right? Or whatever the metaphor is, right? Or the analogy Dan Culverhouse: Yeah, that's fair enough. Vincent: is. So do you really need to know or do you just need to measure the executable outcome? So if you're if you're using a predictive system for demand and it's using 200 configurable parameters in a neural net that is being data science spend a lot of time to train and wait. each element of the weights in that neural net, you're not going to know how that works, you're not going to understand it, But what you Dan Culverhouse: Even if it's held me. Vincent: can understand is what you see in your demand signals after you've executed on the prediction, right? Dan Culverhouse: Yeah, I guess. Yeah, so I guess the answer to that is that, yeah, you're right. No, you can't really understand how it does it. But if you're measuring the outcome and I say it's predicting demand and since you've been using the AI tool to predict or demand your accuracy has improved, then I guess who cares how it does it right? It does it right. And it's doing it better than those 10 people I used to ask to do it. And so, you know, and I am a firm believer, I am sorry, Vincent: and I, and I... Sorry. Dan Culverhouse: I am a firm believer in actually not replacing people with this stuff, but enhancing people with this stuff, right? Vincent: Well, or augmenting then. Dan Culverhouse: or augmenting them as you say to use a better term. Vincent: So those 10 people, you're not, in all my experience, and I've been working retail now since 1985, right? So I'm gonna make a rough statement here that may be controversial. It says, don't care what your skew catalog looks like, you are not gonna get a system that's gonna perfectly predict all of your demand. You're going to have edge cases. You're going to have points of failure. You're going to have, you're going to have, I don't know, massive tariff changes that all of a sudden mean a significant supply route from somewhere in the world to another part of the world. You have to circumnavigate or whatever it is. But what can happen is rather than 10 people trying to do everything and not being able to cover everything because 80 % of their focus is busy work, all of a sudden you have 70 % of their capacity focused on value add and 30 % in checking the outcomes of the busy work that a machine is doing. Dan Culverhouse: Yeah, and that's valuable. That's valuable. Vincent: and that's real value. in the problem that I see, and again, being somewhat old enough to remember some of these cycles, the dot com years, someone in a C-seat, and I shouldn't always rag on the C-suite, but they're very busy people, they don't often have time to understand everything that's going on in the world. you know, the question comes down, do we have a website? Someone says, no. Well, I think we need one. Why? Well, I don't really know, but I'm pretty sure we need one. So go and make one. And I have a feeling that that's where we could be with AI and automation right now. Dan Culverhouse: I guess. Go on, on Keith. Keith: for sure. Well I think a lot of people just don't want to be left behind but the key word for me is value. What you're trying to do, what you're spending your time doing is what I love doing. It's going into businesses to actually tell them what they're doing, what they're spending Vincent: Mm. Keith: their time doing, the unmeasurables. But I think yeah it's a race isn't it? We look back at some of the early websites I mean absolutely... I mean can't even imagine what I thought when I thought there was a future in some of these things. Vincent: Well, the opposite pressure, I can imagine someone in a C-suite might say is, but Vincent, you have to understand that the margins I'm operating on, if my competitors suddenly see a step change in their margin because they've automated before me, I'm cooked. right and then investors investors Keith: Yeah, but automated what? Vincent: investors or VC will turn to me and say well where were you in this AI race well we were taking it we were taking it slowly and being very measured and trying to find use cases well that's all very well and good but now your competitor is that far ahead in market share to catch right so it like like everything Dan Culverhouse: Yeah. Vincent: else that happens in the c-suite I think we can get a little bit lost in our little world, in our little perspective of, well, you know, obviously you're an idiot if you try to do that, but that doesn't necessarily, it isn't necessarily going to help the business. You have to be able to have the balance, right? And Dan Culverhouse: Thank Vincent: I think part of that is in education, so that your... your advisors are advising you well on what your competitors are doing and you have a good breadth of understanding around the marketplace to stop you, what is it we used to call it, once upon a time, you're being on the bleeding edge, So you do something and then you cut yourself and it causes a lot of pain, or worse. But then being left behind is worse and ultimately it's a judgment call. So, you know, what I recommend, obviously, what I would recommend is that they all ask, chat, GPT, what should I do? Should I automate or not? No, I joke. But, know, we've got a balance. Dan Culverhouse: What are the questions that... Vincent: You've got, well, you have to be cognizant that there is that tension in the C-suite, right? And that you need to find a balance. Keith: Who do you think you should employ? We hear a lot about AI replacing people, but in order to make the right decisions, who do you think is a key employee of a business these days around AI? Someone who would be able to navigate their way around this stuff. What's your opinion on that? Who should businesses go to when it comes to making the right choices around AI? Should the tech team be looking into this? Vincent: Well, the thing is, you could argue both sides that one had sort of knew, yeah, who has that experience? mean, if you look on LinkedIn now, you'd be forgiven to think that just about Dan Culverhouse: Everyone. Vincent: three million people have implemented Keith: Yeah. Vincent: agentic AI frameworks in their business successfully. I'm here to tell you that they haven't. Yeah, so there's that element of it that it's so new. But then again, some of the business trade-offs remain the same regardless of the technology, right? So my personal opinion is the people who are best mandated to that are the people who you give the mandate to, who you give the objective to, who you allow to go and get the education, ask the question, but they're coming from a point of knowledge of your business. such as it always has. There's never been anyone so useful to me in the years that I've been implementing technology and that's been a long time now, other than the person you take out of the operation and you put into that project team because they know the business. Yeah. And they are the most, Dan Culverhouse: the doer. Keith: Yeah, absolutely brilliant. Vincent: they are the most, you can get all the consultants you like and we've all worked with lots of them, right? A lot of are very clever and a lot of them add a lot, terrific amount of value. But sometimes it's a flaw that they have, is that they understand that cleverness that they have and it sometimes stops them from listening to the people that are working in your business. Dan Culverhouse: Yeah, absolutely. Vincent: And in my experience, the most effective ones are the ones that are smart but are also smart enough to know, well, I don't know your business, so you can tell me about that, Keith: Yeah. They know what they don't know. Dan Culverhouse: Yeah, one of the things Vincent: Mmm. Dan Culverhouse: that I can see happening as, let's go back to the predicting demand thing. And you've got this tool that's given you, yeah, you know, this new demand curve. It's proving more accurate than the past. And then maybe you're in a SNOP discussion and somebody, I don't know, in the commercial team disagrees with it and challenges you on. Vincent: You Dan Culverhouse: right and says well you know I don't think that's right you know why have you come to that conclusion Vincent: Yeah. Dan Culverhouse: is it going to be an acceptable answer to say well we fed it into the AI tool and this is what it told us and predict in over the past six months it's been very good is that going be accepted because suddenly you've seemed to lost your ability Vincent: I don't think... Yeah. Well, I don't think... Well, I... Dan Culverhouse: to understand why you want to use this demand other than how you used to do it. Vincent: Well, the answer to that question previously was it was Mr. X with his spreadsheet that our supply chain depends on. He's been giving us that answer for the last however long and we've always Keith: Yeah. Vincent: trusted it, right? So Dan Culverhouse: Yeah. Vincent: all we're doing is Mr. X is getting his number from an augmented place. Dan Culverhouse: I could dial in Mr. X and ask him. Vincent: That, well, all I'm saying is that business race, he doesn't change. That tension, Dan Culverhouse: Meh. Vincent: that tension's always been there, right? What you're alluding to is, you know, is it a commercial that run the supply chain or is it supply chain that run the supply chain within parameters or? Well, you might have said it in the past. Dan Culverhouse: No, I don't know. I don't know if I'm saying that. I guess. But I think what I guess where I'm coming from is that, when it was somebody, well, actually, I may be I'm answering my own question here, I guess. You can you can challenge the AI, you can go back to the AI, so why have you put a 20 % uplift over there on that week? What's behind that? Because you can, you know, it will tell you. Keith: Give an explanation. Yeah Vincent: Well, I would again, I would say that Mr. X is being augmented because he now has a much better dashboard to measure those outputs because he's not spending his time with the pivot in the spreadsheet actually coming up with the number in the first place. But now he's. Dan Culverhouse: you spend this time interrogating what it is, what the number is. Vincent: Now he's running much better measures across a fully automated dashboard that allows him to dig and poke wherever he needs to. And through the use of interrogatable data with maybe an agent that's helping him, he can get his answers quicker and more effectively. Dan Culverhouse: Yeah, and more in depth maybe even. And I guess that is one of the things that even in the small amount of AI things that I've done is the great awesomeness about it that I love is the ability to just go, I don't really understand, explain that, know, tell me what is that number there? Why doesn't that, you it doesn't see, you know, you can interrogate it and it will quite happily explain it to you for hours on end, right, if you want it to. And I think that is a wonderful thing. Vincent: Again, there are things you have to be aware of, certain biases, machines telling you what it interprets you want to hear by its understanding of your question. Dan Culverhouse: Yeah. Vincent: So again, you have a certain level of skill maybe that's required because people will be screaming, yes, but they hallucinate all the time. And of course they can hallucinate, but we can all have a bad day. We can all come into work a little bit hungover. Maybe we're seeing double on the calculator. Dan Culverhouse: Big shit, Vincent: But it's again, it's the measures that you put in around how your people interact with those outcomes and how you keep saying the word ad nauseam, meritorious. Keith: And if you interrogate data, yeah, you interrogate data anyway as a living and you're skilled at that and you want to augment your skills, you're going to interrogate the answers it gives you. That's why I feel education as you say, Vince, is 100 % key. Do you think there's many good educators out there? Where do people go to get the most reliable education and AI? Vincent: YouTube. So I think, Keith: Yeah. Well, I've learnt most of, most things. Vincent: well I think my experience is too narrow to answer that question. However, what I would say is I think... how you construct it. I'm lucky enough to work for a company that puts some time into educating its people around AI and spinning up some courses and so on. And I that content exists and you can find it on LinkedIn. But the machines themselves are very good at helping with that. when I left my last gig five weeks ago, I decided that I wanted to develop software, which I'd never done before in my life. I didn't know how to do it. And the machine taught me. and you've probably heard the phrase vibe coding before and I didn't know how to use Keith: Yeah. Vincent: ages and now five weeks later I've got one software project in pilot I've got another one in design I've got two teams of agents and two project managers and a governance framework on top of that developing the software and I've been through the loop and pushed apps out onto the App Store and understood all of that and blah blah blah and And it's been through machine tutelage, because I'm not a software developer and I'm crap at maths. Dan Culverhouse: Let's just be clear on that, these people that, well, these employees or if you like partners that you've got doing that are AI agents, not people. Vincent: Yes, because I again, I was I was taught how to use them, what they were, what they could and couldn't do, how to use it, how it sits to a framework. And now I've gone so far as that I have on my local machines. software that I've written myself that allows me to spin up agents and either use outsource to a Chinese cloud frontier model or use a local model running on my local machine to do the work and it's now doing things on my computer for me without me doing it myself because it doesn't take you very long if you've got a certain level of interest to your point about being educated. It doesn't take very long to learn how to do it now. because the tools are out there. But, you know. Dan Culverhouse: And do you think there are roles that suit an agent kind of doing on your behalf and others that realistically are not quite so likely? I mean, think the answer is obviously yes. But I mean, you know, is it more the technical, analytical stuff that the more strategic leadership? Vincent: No, no, I don't think so. So when I say I have 16 agents, I have two software teams. And those software teams have personas within them. And each persona has a job description and an objective and its own little file that tells it what it is, what it's about, what, blah, And those roles... all work together to produce my software which I now get to, instead of worrying about lots of the minutiae of that, I get to decide what I want it to do and how I want it to look. And that I may want certain type of content, so one of them is a human to human piece of software and I have a behavioural scientist agent and their job is to research behavioural scientist stuff on the web, bring it back for content and say well we could look at it this way, that way, that And I do get a kick out of asking the team for a proposal on something that I've woken up and thought about and I see the little boxes come up as the project manager spins them up on the harness and they all pop up they have a little chat and I get a one-pager that's basically the output of the meeting with what we talked about who chipped in what they chipped in with and what what the recommendation is and I Keith: And do you ever, well I was going to say do you ever chip in yourself and say that's not what I asked for. Vincent: Sorry, go. Well, on the output, yeah, and I had to last week, because I nearly lost control of my bladder last week, because in the output, the project manager told me that, unfortunately, it was their job to represent that the solution architect had essentially got a little bit chippy about scope creep, because I'd woke up in the morning and had another idea. Dan Culverhouse: you Vincent: and the solution was, this is the second idea you've had this sprint, are you sure that you want to... How many times have I had that conversation for real? Dan Culverhouse: Probably less polite. Keith: Yeah. Vincent: At least it didn't have an emotional tantrum and walk out the building for a walk around the office, absolutely. Keith: Yeah, didn't get clouded. Vincent: But it absolutely was part of the conversation, yeah? Which kind of, well, it really Dan Culverhouse: Amazing. Vincent: made me laugh. But yes, I of course can override anything the way it's, so again, because I have a little bit of a background in this, I defined a methodology that I wanted to go from idea to functional spec to technical spec to testing spec to development sprint, blah, blah, whatever it was. But I basically got the process. project team to settle on a methodology and obviously I control that methodology and I kick off the sprints and all the rest of it. So I'm in Dan Culverhouse: Please. Vincent: full control but now I can develop a software product in the space of a week and I'm building too. Keith: Is there anyone in the team? Dan Culverhouse: Okay. Keith: Is there anyone in the team that you think, no, not much good, could get rid of him and put Vincent: Hahaha Keith: someone else in? Or is it an easy team to manage? Vincent: Ultimately, it would have been my fault because I didn't get the specification in the spec file right. Yeah. So I've been through a couple of iterations of developing the agent spec for a particular role. Interestingly, there has been one or two times when the machine has said, I'm spinning up two of the agents for this conversation. And I thought, well, why aren't you spinning up this agent or another part of the team? And I've actually had to say, no, I want contribution from the tester. I want to know how long this will take the test as well as is it a good idea to build. So interestingly enough, I do have to be somewhat mindful of who in the team is being included in the conversations. Again, very much like real life. Dan Culverhouse: So the leaders are safe then. The leaders are safe then. But they still need to be at the top. Vincent: Yeah, yes, and one thing about it is I can work with colleagues who have good ideas themselves. So one of the pieces of software I'm developing is not something I'm an expert in the domain. But a colleague has heard my story about what I've done and what I've already brought to a certain point. And he said, well, what about if we could do this? Because I've had this in my head. and now it's a project Dan Culverhouse: Ahem. Vincent: team building out his vision. Dan Culverhouse: Maybe we should talk after this, we've had some ideas too. Anyway, different, we digress. Vincent: Well, the problem is, and if people see this online, I don't want them to get the wrong idea because there's a million people writing apps in their offices or bedrooms. The Dan Culverhouse: Yeah, yeah. Vincent: problem is distribution. What's the problem you're solving? How do you get people to understand that you've solved that problem? And how do you get it to them? And those are the same business problems. Those are the same business problems Dan Culverhouse: Yeah, I had to them to hear about it. Yeah, amongst all the other stuff. Vincent: that ever they've been. But to your point. of your original premise. I can now focus on that, whereas before I couldn't even get anywhere near it, right, because I needed a software architect, a cloud architect and so on. Don't need those now. Dan Culverhouse: If it was an augmented team, those are partially agents and partially people. Do you think the agents can manage the people? Vincent: Not in the way that I have it set up. Dan Culverhouse: Yeah. Vincent: And I think the answer to that honestly is truly no. So we're a long way away from general intelligence. So to manage people you need general intelligence, you need experience, you need emotional intelligence, you need blah, blah, blah. To manage a task, no, you need narrow AI. But I have a project manager managing a task. But the main problem is he doesn't really have a concept of time because unless Dan Culverhouse: Yeah. Vincent: I schedule his automation, he's living in a cloud. And as far as he's concerned, he doesn't really know if it's Tuesday night or Thursday morning. So they're Dan Culverhouse: Yeah, very true. Keith: yeah, yeah. Vincent: not cognizant of time unless you schedule tasks and then Keith: Absolutely. Vincent: by all means he can spin up. check the schedule task, check an inbox to see that he hasn't had a reply to the email that he sent my colleague for some explainer on a particular use case, and then he can send an email from my email account to that person to say, I haven't seen that yet, so that the team aren't working on it yet. That can happen. Dan Culverhouse: Yeah, yeah. Vincent: But that's a narrow AI, that's a very narrow focus. One of my favorite things to do with super intelligence is I like philosophy. So I spend a little bit of time, I spin it up, and I say, I want Hegel to talk to Schopenhauer about the meaning of life. And I just want to sit back and watch. And then I watch a philosophical debate that I'm in Norway. capable and intelligent enough of cutting through their key principles and their views. But they I can and I can get a report and say, well, they disagreed on that. They agreed on this and blah, That to me is super, super intelligent. Right. And Dan Culverhouse: Yeah. Keith: Hmm, very... very different benefits as well. Vincent: lanes. And they're Dan Culverhouse: Yeah. Vincent: with very different benefits. And of course, we're nowhere near, I think we are superintelligent because I think things can be superintelligent because you can have a model that is big enough to understand most of human history. Dan Culverhouse: Ahem. Vincent: Going back to your question earlier, Keith, who's in a position to tell us? I don't know. I have to walk a balance in this conversation because at the minute it's very easy, like everything in our world, be polarised. You're either for AI, in which case let the robot masters come, I'll be perfectly happy, or you're absolutely against AI and you have a Luddite-esque conversation. I try to be somewhere in the middle. Has it improved my life recently? Yes, because I can bring products to market that I couldn't bring products to market before. Would I give it my credit card to do that? No. Dan Culverhouse: You Vincent: Because I honestly don't Keith: Yes. Vincent: know if it's going to suddenly wake up in the morning and I need more bandwidth for that question or whatever, and just pop off and buy some. Keith: How do you know? How do you govern it? How do you know until you live it and then each moment is a new moment? Vincent: It might have been Plato that said when society's music changes there should be laws to govern the nature of that change because music affects society. Or I might have just heard that on a record and thought it was cool but I'm pretty sure it's true. We go back Dan Culverhouse: And I think they're right. Sorry. Vincent: to similar principles. Dan Culverhouse: And to tie in, you know, I mean, I'm a little bit practical and pragmatic about all this stuff. And so if we've gone through this this period of, know, of really thinking about what it is that we want and where we think we need to go. And the decision is, yeah, I think we do need to deploy some of this technology. could really help. And for whatever reason, it doesn't really matter at this point. How does one how do you go about? finding because, you know, when you're talking about, you know, maybe, you know, a major sort of supply chain operation that you want to involve, you know, some, some, some AI tools to really support what you do. You don't generally go and ask chat GPT or Claude, right? I mean, there's how do you determine who should I work with and how do I? Yeah. What's what's what kind of AI tool do I need? Or do you just go on to ask Jack GPT, which is the best tool. Vincent: Well, there's questions of scale in there. So depending on the nature of the business you are, you probably have pre-existing vendors. I would be surprised if you're a supply chain business. I'm surprised if you don't have pre-existing vendors beating your door every day of the week with Dan Culverhouse: Sure, yeah. Vincent: why their new systems are better and blah, blah, blah, and more agentic or more AI this or more AI that. Dan Culverhouse: Hmm. Vincent: If you're dipping the toe in the market, then You could always go to RFP process but don't do that. But you you can find people, find some advisors that you can trust who seem to have an idea of what they're talking about and ask them to help you. But you've more than likely got people in your business who are probably coming to you and saying I think we could be doing this, I think we could be doing that. Listen to the people in your business who are actually already coming to your door or give them an opportunity to come to you. Like the suggestion box, Because AI is now the genies out of the bottle. Lots of people are having ideas and a degree of understanding and your next big business change idea might come from some... some lass that sat in behind a desk somewhere who just needs to be given the ability to say what's in her head, right? Or it might come from a trusted vendor who's actually made a reasonable step change in a piece of technology and you interrogate it properly. But I think however you decide on that, it's about almost more than a choice, it's how do you implement. And and Dan Culverhouse: Yeah. Vincent: we talk about different measures in you know speed to market or time time to time to value or whatever it is But you just have to be mindful that when you're deploying the technology that maybe isn't the most well understood because going back to what we talked about, know, neural network and outcomes and blah, Dan Culverhouse: Yeah. Vincent: you know, then make sure that it's sandboxed correctly in your business. Make sure you do all of the things that you would normally do with good governance in technology to deploy systems in the right way. Dan Culverhouse: Yeah. Vincent: And for heaven's sakes measure. My particular thing is before you do all of that, if you're in an existing business, start at the very beginning, go out and understand is your understanding of your business operating model up to date? Is it well understood by every layer of your operation? Is it as understood in the boardroom as it's understood on the warehouse floor, the shop floor, whatever? Do the measures that you Keith: Often not. Vincent: use to, often not, and are the measures that you use to measure the operating model well understood so that the measures that exist on the shop floor or the warehouse floor are reflected in the measures that are seen in a balanced, let's call it steering wheel, at the exec level. Dan Culverhouse: Thank Vincent: And if you're happy that that solid foundation is there, then why not look for opportunities to automate? Because all you're really going to do is re-interrogate that operating model and look for the opportunities. And if your people, your business, understand it that well, they're going to tell you what they are. Right? Because the problem exists today. Dan Culverhouse: It's all those good change protocols that you should do in any kind of development and project management is, you know, is, you know, don't try and change everything at once, know, test, pilot. Yeah, you you really need to. Vincent: Well, becomes philosophical. It becomes philosophical, right? So you can fail fast and work backwards, but it's always good to do that from a position of very strong foundation and strength, yeah? Or you can build forwards. And if I think about my days mired in ERP technology, because of its nature of being sequential, you start with a purchase order, you have to get that handling right, and then you do, do, do, do, do. Those projects, how long do those projects take? They were only supposed to, know, a scaler or what a hyperscaler, whatever they call them now, or a large vendor would tell you, it's only gonna take you 80 months. Yeah, six years later, right? Because you were in pursuit of perfection and it was very linear progression, right? So you can always fail fast and this technology allows you, if you've got good data. should say that it allows you almost to jump to the end say here's a bunch of data tell me what my business could have done better or tell me what the better outcome could be and then I'll work work that backwards into my operation is an approach you can take right it depends what you're trying to achieve Dan Culverhouse: Yeah, I think. Vincent: and with what and with what Dan Culverhouse: Yeah, I think what you said earlier about, you know, your term augmentation, you know, I think the worry is that people think, this sounds great. And, you know, I'm going to deploy, you know, AI to do all my demand planning. You put it in, starts to give you some good results. So you sack all of the, all the people that used to do it and then just rely on the machine and then something starts to drift and it starts to go on. The power of this is really using it with the experience of the people that you have to increase its power, its intensity, its ability. And maybe you'll be able to do a lot more than you used to do with the same 10 people, rather than say, Vincent: So. Dan Culverhouse: I've just got a machine now and I don't need to worry about Vincent: So those 10 people, the way I see that, if you think about retail business, those 10 people instead of spending 80 % of their time churning out the numbers and 20 % fixing what was wrong, they can be spending 5 % of their time. making sure the numbers are being churned out, another percentage of their time making sure that they're managing the exceptions, and a very large percentage of their time working with their commercial colleagues that they understand the numbers and working on the future plans, and further integrating commercial tensions and supply chain pragmatics to better outcomes in the future. Because all of a sudden you've given them that time to collaborate. Dan Culverhouse: Yeah, absolutely. And I think there is anything that people take away from the discussion we've had today. It's that that, know, really, if you really want to enhance or draw out the power that something like AI can give you, it can make your people add value to your business so much more because it takes away a lot of stuff that people do that is not so value adding and it's just grind. Vincent: Yeah. Dan Culverhouse: and you and I'm not saying you you want it to give you intelligence but I think you can do so much more with your people when you've got the power of AI supporting them rather than going in with this philosophy is how do I replace my people with a machine. Vincent: Yeah, that is a very flawed philosophy because you should be going into anything with how does my business arrive at better outcomes and how do my team do better. Dan Culverhouse: Yeah, it's a journey, It's a journey. Don't presume you know the answer or the final outcome. Just take it on a journey and see where it'll lead you. That might deliver you much more. Vincent: be aware of the nature of the AI that you're talking about because again if it's a large language model powered solution you probably don't understand how much it's going to cost you probably don't understand what your people are putting into it and you may not understand what governance is coming from the EU or the UK etc to be compliant to how you're using it Dan Culverhouse: Yeah. Vincent: which is something else companies need to be very, very mindful of. It's going to make GDPR Dan Culverhouse: Just. Vincent: look like a walk in the park, being AI Dan Culverhouse: Thank Vincent: compliant. Certainly if you're doing business in the EU as it stands. And there are businesses Dan Culverhouse: again. Vincent: around that can help you with that. And be aware of the same old things. You you put crap in, Keith: Yeah. Vincent: you're gonna get crap out. That ain't changed. Keith: Yeah, I think my takeaway and thank you, I've learned so much, but my takeaway is educate. Education awareness. Talk to the right people on the floor and in the industry the same as you would for any change management. I don't know why you would treat it any different. Walk before you can run. Understand what you're trying to achieve. But it's been a great chat. Thanks so much for being our first guest. It's been a riveting conversation and hopefully we can invite you back soon. Vincent: It's been an enormous pleasure. Dan Culverhouse: Yes, we Keith: Thank you. Dan Culverhouse: may ask you back. Keith: You Vincent: We'll try and make that blur filter work so that my home office disaster isn't so distracting. Keith: So like, share, subscribe and comment. We want to hear your comments and we want to hear your opinions of AI. What have you tried to apply in your own business and has it gone well? Has it gone badly? We'd love to know. Dan Culverhouse: Excellent. Yeah. Fascinating conversation. Vince, thank you. I'm sure once we stop recording, we're to carry on chatting because I think this could come on forever. But really, really a lot of fun, very interesting and insightful. I've learned loads today. Really great. Hope listeners have as well. Yeah, as Keith said, we'd love to hear what you've got to say about this topic. It feels like there was a lot in there that people might have views on. But let us know what you think. and of course we'll be back next Thursday. Thank you.