And it was honestly, it was like I was a caveman discovering fire. I've never ever seen anything like it in my life. Hello, everyone. And welcome to Undoct. Hello, Royal. I don't know. Hey, doing. I'm well, thank you. And a very special hello to you, Peter. Hi. Right. You're, you're looking slightly bewildered. That we've invited you on this podcast. And that's probably because I invited you about 24 hours ago. So let me give some, give some context. Any anyone, anyone who's listening for a while knows that we acquired digital ship over the summer. And that meant kind of putting, putting heads down and doing a lot of work to get it up and running. And we've, we've got a really good team in place now. And so I ended up sort of popping my head back up and sort of seeing what's going on in the technology world about two weeks ago and realized that I had missed some fairly major updates when it came to AI. So I actually put a post out publicly on LinkedIn, asking for some help. And Peter, very graciously volunteered his time to tell me a bit about what's going on in agent to AI. And I kind of sat in the hour we spent together, slightly mind blown. And then I spent the rest of the day playing with these tools myself when I was so utterly blown away by it. That I decided that we had to get Peter on to the podcast to tell Raul. And then also to tell everyone in our audience who listens and is interested in the maritime technology world. Today's episode is brought to you by Fartick. In a digital bridge environment, poor visibility isn't an inconvenience. It's a safety risk. Fartick's online maritime monitors give marine professionals the clarity and confidence they need. They're specialists engineer every display with high brightness, anti glare glass and extremely wide dimming options for any lighting condition. Back by DnV certification and trusted globally, these monitors have been used in navigational system monitoring and CCTV systems. Integration is seamless with multiple video inputs, a cross fleet ownership, shipbuilding and marine systems integration. Fartick provides a partnership grounded in expertise, regulatory assurance and life cycle support. Discover the end line series at Fartick UK. Peter, do you want to just give us a bit of a background as to as to who you are and I guess how we know each other. Yeah, sure. So yeah, Peter Rossi technologist by trader, so 20, 60 years in the industry, designing building, scanning platforms, people. I started in Formula One with Macaron Formula One team and travel world then running track side IT. Then when I often did, did some consulting work or help redesign a number of sort of largely commerce platforms. Got into the BC world started that by doing sort of techy diligence or swalbc companies. Before techy diligence was sort of known as a thing that you did during M&A. It was quite funny at a time when that, you know, sort of speaking to them and then they were like like you literally don't know what we're investing in. Like several times it works, sometimes it doesn't, but some of the time we think it's an app that's already on the app store and it's just a picture of an app on a PowerPoint presentation. Then I've been involved in a cloud style. So that was we see backed and initially it wasn't it was it was interesting. The invested in my number of 1918 pop stars like half a spanner ballet and good. It was interesting interesting time. Built a SaaS company well sold that in 2021 into a private equity group and then when enjoying them we packaged and sold that business in 22 and then over the last four years I've been with them as the CTO doing. Basically a big buying build we acquired 22 companies internationally over the last four years and been been integrating those into the tech stack. So yeah, that's that's me I suppose. It's quite varied. It can I we sort of met when I'm now invested in in Beliga. And so maritime business and you were setting up the years at the time and we had some. How's an interesting conversations about what is like starting a business and and gang going? You went from a clone F1 to working with half a span of ballet as investors and that the natural lead on for that with the glitz and glamour is maritime. Right. So I did you end up investing into the maritime industry moving up the showbiz. I just saw like it was missing in my life. Like with all of these things right tech is a tech is a theme. It's an enabler for a lot of businesses right. I'm not big thing about tech to tech sake. I enjoy using it as a tool to to solve problems. So I met some interesting people at a time that were in the in the bunkering world and they saw an interesting opportunity to do something there. So I got involved with them and helping them to sort of build something and showed it up becoming Beliga way. Often talk about M&A and putting businesses together and certainly from a sort of data and integration point of view. But did I hear you right when you said that the current thing you work on is 20 22 companies. Yeah. 20. That's quite excessive. All in the same kind of space or is it what's the sort of level of integration is going on there with us. Yeah, it was it was certifications say I said certification might come across I said 91 or 27,000 one. So one of the more. Yeah, it's even more. Right. It's one of those ones where you're down the path and people get like what is it that you're doing? Oh, I'm going to lose. It's an interest. Yeah, again, I'm paying that for what it's an interesting commercial thing because it's business is like. No, nobody sort of as a five year old friend setting up a business and having an I say 27,000 once to get on the wall. Right. That wasn't that that's not the dream. I don't know if double business and decent test for all that stuff. But think about I say certification is that you. You enables you to sell more typically and the reason why companies get like 9,000, 1,000, 27,000 is say you built a SaaS company and it was going well and you were selling to customers and then all of a sudden. Do you go came along and we love your product. We want to roll it out across all of our user base and we'll pay you 5 million a year for in licenses. Can you can you just send over your I say 27,000 one certificate so that we know that you can you know you're securing it all properly and it's literally just the tick box exercise. And they and it's like, you know, and that's the barrier. So they and they then go out and work with certification business or consultancy or software company to help them achieve I say 27,000 one and that's why they're going to say the reason why private equity likes it is because it's like a new model right businesses. Get driven into buying it's not it's not a question. And they need it to keep selling so they can come back year on year for a certification and say your book builds year on year and you can just go from there. And it's still quite fragmented market so you know I joined up time when the business was just under 10 million revenue and about 3 million of EBITDA. And they acquired my business software company in that space and I just didn't some consulting sort of with the handover and I thought you know it's interesting to have to make the C suite and sort what they were doing that. So yeah, you know, so compared to that point is it was an interesting problem. Then it has CTA or a test actually at the time. So I wrote a strategy got to know them and then we sort of went on that journey. At the time of joining as I say it was about 10 million revenue and they've been doing it ever roll up in the UK bought their first business in Ireland and haven't redone the more they want to sort of take the model and try it in the US. Which again is quite good fun because normally like you know if you're doing VC back tech in the UK you want to get bought by a US company that's that's the dream right. It's rare that your UK company that's going out buying American company which was that was that's right. They must love that when they're on the run. Well just kind of just another point that just struck me there as well when you said about tech due diligence because obviously I've been through that quite a few times and the importance of that getting that expertise in to kind of actually check into. And we were talking about this through an a islands in an episode where we were talking about no code low code vibe coding and stuff like that because now of course you can give the small group of very effective application without actually be able to write a line of code so. The kind of that kind of DD works probably becoming even more important in the world of a. Yeah no it is it is I think it's still sort of changing because there's there's the fear at the moment that you buy the coded software is buggy and it's got less of security holes in it and actually there you've. People are finding themselves in a position where they're taking a product VC funding and the VCs are going like what tell me about how it was. I just told to an AI and it did it you know like there's actually no there's nobody engineering wise under the if it's that's doing that and to sort of get a counterbalance to that AI written software can be amazing if you if you've got somebody that knows what they're doing with it it will just rapidly accelerate them you can build something that's 10 times better than you would have built. I suppose I'm one of those people that I can end up I end up annoying other temperatures because I'm not pure as a tool I am you know there's always industry best practice and everyone gets very slow and bogged down in what the latest best practices today around.
of how you build things or technologies and use of packaging of C on CB pipelines or whatever it is. And really, I don't care. It's actually about what what was the business trying to do and what made sense to that business to work with. Yes, you could have a supersexy, best in class software stack that runs your business or client-pacing stuff. But ultimately, if your business is still growing, the product is still working, it's still underpinning all of your revenue, then actually, you know, we just need to look at the risks, you know, like where might you hit a scaling problem or where might it be a security thing that we need to tighten up? Or maybe it does need to go and it's rare, it's very rare. But a lot of my, a lot of my career being bought in bite boards and the funds to go and go and help companies where they've gone like, oh, they said they were going to re-build their core product and there's 12 months in and it's not done and they're moving forwards commercially in that time you're going on, it's like, oh, yeah, it's been obvious. You shouldn't have done it. You've got no experience of that, Royal, I'm sure. None. None at all. Don't know why he's talking about it. I feel like we could probably have a whole hour talking about private equity technology diligence, but I'm conscious that I brought you on here to talk about what is going on with AI and perhaps more specifically, Agente AI. Would you just give us your view on what's changed in the last six months that's taken us from talking to a chat interface to some of the stuff that we've talked about over the last few days? Yeah, yeah. I think the thing is, um, so, oh, actually, slightly going back into the tech due diligence bit for a second, you can carbon date a company by its tech staffer all the time. They start showing it to us, okay, it's a medium rail application right? This company has formed around 2014, okay, it's a classic SB, this is probably around 2004. It is generally most of the time and those companies get stuck on those tech stacks because that's what they know at the time and obviously the cost of moving something else is really high, so they tend to not do that. And what I'm finding with AI is that, you know, in the last three years, it's crazy what's happened. It's moved on at a rate of knots. And actually what you find is like a lot of people have their views on AI and lots of people using it in verticalments, but they're using it at sort of different capability levels that typically get locked in at a point in time, you know, they tried it or they tried it for a bit or they went into it and came out of it in the back, but you can tell when they started using it and then what their view is around how you continue to use it and what its capabilities are. So there's still a lot of people that view AI as a like a chatter to things and, you know, they got introduced to chat to you PT back in 22, 23, these that they're using a bit like Google to ask a question might got an answer out the other side of it could have been useful. They might have found that the answer was rubbish, hallucinated and that and that's it. The thing is sort of over time that's changed a lot. You know, it tends to break it down to sort of tier one, tier two, tier three AI. So tier one is it's like you basically chat to PT interfaces, Gemini or Claude. That's your, you know, you go in, you start a new chat, you have a conversation with that about something, and it's operating very much inside of that box. So it's, you know, it's very useful, very powerful, the models that moved along are rating knots, you know, and you know, sort of it's G-35 now, they're like, AI or 5.4 is over the years of the day and you've got OPPOS 4.6. Incredibly powerful models that, you know, you do a lot of ways. In fact, you know, if you look at what you get back to sub G-4 or before, you know, what that was mind-blowing about then, but actually it's, you know, to complete rubbish now and comparison to what they're doing. A lot of people are still locked there, you know, I talked to people, you know, Dean, you say, yeah, you know, you don't, I'm, you know, he's, you know, just talked to Claude and, you know, got lots of conversations in there and that's it. And that's their view of the world. It's never, never got more expensive than that. At the same time, all that stuff came out, open AI, launch APIs to be able to sort of programmatically interface with their models. I played around feeding, thought a few Chrome plugins and things and found that actually, you know, it's really useful for doing things like data, like the small models of brilliant, you know, imagine like you acquire a business and it's not like a mailing list. I mean, or they've got a customer database and you say, you know, we need to send out an email to all of these businesses and let them know that, you know, we, it's changed hands and the data that you get can be really messy, you know, you're, you know, it's richly, it's got a cell, then it's got three malagrester's, some of the number of apps in them, there's semi-colon, the way that they're placed, addresses, addresses are hilarious, database perspective, because everyone's on a site, you know, on how you should store it, right? Should be addressed one, two, three, towns, cities, it, what is it? And the lovely thing about, you know, a large language model is that, you know, it understands that really easily. It's a sort of thing you can eat to breakfast so you can just say, let his, his, the, his, the, his, the columns that I'm trying to get and then hit the data and give it back to me in the structure format, you know, you can sort of clean up data really easily and that sort of thing is really, really powerful and it's still so really difficult to programmatically solve that like with, you know, like, oh, if it looks like this, then do that with it and then, you know, it's, it's sort of thing you've probably spent just as much time to you are, you see, what if you're doing it manually, it's like the old adage of, you know, why, why spend 10 minutes doing something manually when you can spend two weeks failing to automate it. That's literally my life. So yeah, so that's your tier one, you know, the API stuff came out. Two tier is sort of where apps had started to really integrate it and make it sort of leverages of all the tools. So in the software development world, people that program, do programming inside of an IDE or an integrated development environment, that's got like the code of some of the always ones that you open up. Microsoft bought out co-pilot with Git and say that, you know, you got an AI live with you while you're doing software development and it's not just a chat window, it's got context about the thing that you're working on. So you can sort of highlight a function and go, you know, can you help me with this database pool, this bit, or something, and you know, very basic level that's sort of what it was doing. Yeah, that grew a lot and there was a number of different tools that came out after that, you know, cursor, which is really popular, and Windsurf and a few others. And it's explained that after that, you see an AI game built into absolutely everything. And then so after that, so tier three is really sort of when businesses started using those APIs themselves to start like, how do we bolt this into our existing sort of ERP and SaaS solutions and internal processes and actually leverage our internal data to do something interesting with it. I think that sort of parrots changed. And tooling wise over the time, you know, every time a new thing came out, you thought, you know, I'll personally deal to you again, like, wow, this is it, you know, this is, you know, like everything that came before is, you know, it's completely defunct. This is this is the way forward to nothing's going to beat it. And then literally through that space or something else comes around, it just eats the other thing for breakfast. So most recently, probably about 12 months ago, I would say that, you know, Claude code planning out was probably the thing that that made the biggest difference to the industry. That that was huge. And then most recently open-claw and being released. And I would argue it's the thing that parrots, Claude code would breakfast, although there's still still space for it. Claude code's funny one. Because I think it's the most badly branded product in history with in essence what it actually does and how you can leverage it and the power that it's got that you can use it for almost anything other than software development. If you go back to that point, I was saying about, you know, sort of your chat interface, it's the AI is in a box, right? It can really only talk to you a bit with chat. They've improved that, they've added functionality around, sort of projects, so you can add load files into it and it can reference them. And you can use sort of connectors to connect it to, you know, HubSpot or Git or your file system, so that it can go off and reference them. It works, but it's a little bit boxed in. Claude code was, what is, it's a tool that you can download and install on your laptop. If you're in the Mac ecosystem, then you're in luck because it's super easy. Property websites have looked at Claude code. The finally way to install it, you need a subscription, but it still works on the $20 a month's of fiction, so it's still a lot of the very cheap end of it if you just try. Basically, all you do is create a folder on your desktop. Go open your terminal, go into that folder, type Claude, bang, Claude's running in that folder, right? If you want to not have anything in it at that time, and you can just start talking to Claude about, right, there's chat interface. You don't have to worry about command lines or anything at this point. All you've had to do is type Claude in the end and you're in. You can then use it like you're using your chat interface with chat GPT or Claude or whatever it is that you normally use it. At this point, you've got the power of your laptop and your files and your file system available to drop into that folder, and you can start providing it with context and things to do. When I was talking to you, the day we were talking about social media management, it's always one of the ones that people come to today. I think product management is another really interesting one. I'll talk about that. Basically, I use Publa as a social media publishing tool, and there's loads
having these buffer or Sprout or PooT sweet or whatever. So I got a folder in there and I said, look, I'm using Pudler and I'd like to be able to schedule posts inside of it. I'd like to be able to pull analytics on the posts and going, I'd like to be able to monitor and change and schedule. It's going to go, okay, right. So it goes off and researches the API and goes, right, actually, yeah, if you could put the API in this environment there, I'll write some scripts that can do that. And then it stores that in its form test and goes, right, I took the cover now. What do I do? So then I've been given it, I've got another folder that's got my Tone of Boys and I've downloaded all of my articles over at NLINK 10 and published in other places, put that out into a text file. I'd like you to pass that and write a Tone of Boys style guide. Okay, it doesn't. It knows me. It can write like me at that point. And then at that point, I've got a very powerful tool. We can start building up some articles that we can post and publish. Where it then starts getting what power, we start layering in additional AI tooling. One of the ones I really like is there's an app called Grinola. Grinola AI is free. The main thing you can put it on your phone and you can install it on your laptop as well. It's an AI transcription. So if I'm in the car, I can just hit the chord, start tracking to like a rambled to it about all sorts of things until I'm thinking in my mind, oh yeah, that's been this idea. I legit about reason to talk to myself. And it's the thing, like your thoughts don't have to be coherent necessarily. Not only the way that you would have a conversation with a human, right? You can go all over the place with it a bit and it'll grab it all. It will then rework that into something coherent. I just said we'll all transcripts out there. A Grinola will turn it into nice AI nodes and then I go back to my social media folder and go, right? Here's a transcript from me talking to myself that what do you think of this? Oh yeah, this is an interesting article. We know based on articles that you've run before that these bits are interesting. So let's tweak this and then we'll schedule it or we can atomise it into a number of different posts. There's bits of that that you're describing there that I do already, but in a very convoluted sort of way and in a very kind of non-repetitive kind of way. I'm doing bits of that chain and honing bits down. But what you've described there is just like an incredible kind of workflow doing all of that for you as well. It's actually really nice. We're recording this on Thursday. I met Peter at 9 o'clock on Monday and I bottled it by the way. I went for Claude Co-Work instead of Claude Co-ed. But it's still very, very good. So I downloaded Claude Co-Work and I thought let's try this social media use case. And the two things I really love, I love coming up with ideas and I love fiddling with copy. Like I really enjoy the actual kind of working copy bit. But I hate all the admin that sort of sit around it. So I basically said exactly that to Claude Co-Work. And just to show you where I was at by Tuesday morning, I can now push a button on my laptop and talk into the laptop. It will then, and I'll say, take down a social idea and I'll talk through the idea. As Peter says, totally unstructured, totally sort of rambling. It then takes that, structures it a bit better and then saves it in a folder of my social ideas. And then I have another button, I can push a button and say process all my social ideas. It will take all of those ideas. It will just sort of work them up against my style guide. It exactly the same thing. I took like 200 social media posts, 15 articles I had written over the years. It processes them against my style guide. And I hate actually going into social media and typing it all out and cutting pasting. And so it actually, it now puts everything up into buffer as a draft for me. And then I can go into buffer and then I can fiddle with the copy and I can hit the schedule button. But the thing that was amazing about it was it sort of said to me, okay, so I'm setting all this up. I'm setting this workflow up. It said, could you get me the buffer API key? I was like, yeah, okay. So I went and got the buffer API key and I pasted it in. And then it sort of ran for a bit and it said, it's not working. Let me just see if I can do this manually myself. And then it just took over Chrome for me, went to buffer and then just clicking around the interface started scheduling posts for me. And it just completely blew my mind. And since then, I've built a whole lot of other stuff. I now have a, it now APIs into air table, into air table, which is our kind of core database where we store our interview transcripts. It can go all the way through all of our interview transcripts. We do about 300 a year. Pull out the key findings, do thematic analysis and then write research reports based on primary research. And it can do it in about five, ten minutes and took half an hour to set the workflow up. It's honestly, it's mind blowing. Well, and Peter, I think this is something that you said earlier, chimed with me, you said, I don't find technology that interesting. And I'm kind of a bit like that with technology generally, but with AI as well. AI as a subject doesn't interest me. Having an army of robots to do my bidding. That I can get behind. So that's a thing, Claude Cino is next. So Claude Coe and you came out recently. It was about a month ago. And this is anthropic scale. It's sort of trying to bring Claude Coe to the masses without people being scared about having a command line interface. The benefit with, yes, you get a lot of the power as you've seen. One of the nice things about Claude Coe there as well is that you can create, you can create skills and functions inside of it. So you could work with it to describe a process like, oh, I want you to go off to this website and research this and I want you to get a cross reference to it with this bit of information. Then I want you to get into search this over on LinkedIn. And I want you to write an article that does this and then process it. And rather than going in every time and doing it, you can say, right, now turn that into a skill. And then even more so, you don't need to run it yourself. You can then schedule that to computer and say, run Claude, run this skill please and do it. And there's no coding. You haven't had to sort of, I work through this sort of complicated process. I think the, well, I think I'd imagine the technology, say, will be screaming while that's security. So it's probably not that little bit. And since we've been talking about API 않ings, I think, you know, you've got to be conscious about what you're doing with it. Where that data is going, what you're doing. But they, you know, largely if you got a, you know, sort of a subscription paid subscription of the AI, the, you know, you can toggle in the settings that, well, you don't know, you can use that data for training purposes, so be aware of that. The, you know, sort of putting different passwords and API keys straight into, straight into the chat. It's not, you know, it's not West practice. It's not recommended. But you can, you know, you can pass it to it in, in environment files. And so, you know, particularly when you're working with Claude code and, you know, or over, it'll say, and can you give me a file and I'll go and put it in there for you and then you can use it. And there's still a lot of security farther around that. I'm only thinking people talking, because realistically, it still reads the file. It's a little, but there's lots of new, and that's the end of learning still at the moment about that in a new toolings coming out around credential proxying and all sorts of stuff. But the thing is, don't give it access to your credentials. Don't, don't let it make payments or give it access to your bank account or anything like that. That's the bit where you want to be a bit, you know, get more worried about and keep its hand box. It's worth saying there are stories of like entire databases being deleted, right? Just just permanently deleted by Claude going a little bit rogue. So there are very real risks here. Yeah, you know, I'm gonna stare at anyone with that because I think that's the thing I think, a lot of people do get very bogged down in it. And it's a tool at the end of the day, right? So, imagine an electric saw is a very useful tool for, you know, doing building work with, putting, you could also chop your fingers off. It's exactly the same thing. It doesn't, and it doesn't mean just because you can chop your fingers off, you shouldn't either. Just got to be aware of the fact that you're getting anything safety goggles and be careful about what you do. What I've found is that I've only really been comfortable using AI for, say, for example, analysing data. If I've got the power to correlate back and look at what I'm getting and question that against the data itself that I'm looking at, the more you use agents and the more you delegate that and the more you remove yourself from the process, is there a danger that that accuracy becomes an issue as well? And how do you kind of stop that from being an issue? Yeah, well, there's a couple of points there, really. One is around human judgment and human and the thing massively important still. Like, you know, even in the social media cycle that we've just been talking about there, Nick, Nick will very can press a button. It's only scheduling it. It's not actually sending it. And before it actually gets then, what are you doing there? Are you reading it? Well, I'm not just reading. I'm doing some pretty heavy editing because the quality of the actual post is variable. But you're right. There was an article yesterday that I saw around Amazon's junior and engineers are all being told that this is criminal into their contracts that you are
for what the AI writes and puts in. So you've got to, you can't just go, "Oh, we've got it wrong, it's fault, it's yours." Like, if you committed it, it's your fault. So, you know, people have to read it and make sure they understand it. So, human judgment is really important. And ultimately, I think the only sort of, at the moment, that's where careers are going, people are going to be moving into much more, like, less doing and more judgment. So, you know, like, what is it that, you know, what is it that's been produced? Is that correct? Is that the right thing for the business for me, for what is it that we're trying to achieve? And yes, that's good, you can go and do it. I was in process automation back in 2008, you know, and we were doing, you know, sort of, it was in a managed service provider. So, they had a very basic model of for every 35 servers that we bought onto Manage, we used to have to hire another person. And my target was to try and get that up to 200 servers with our head, with, you know, monitoring automation, and then using tooling off the back of actor, sort of correlate and then go and go and try and fix the problem. But, you know, realized very early on that actually automatically fixing the problem could be a bit of a nightmare, you know, like, get a low-dispatish you want to server. I don't want to automatically go and delete the database because yeah, so there's one problem, creates another, it's said what we did was, you know, sort of created tooling that went off and it scanned it, it then came back with suggestions to the and the actual humans in the team and it says, let, you know, this service got this space problem, it's all the files, this is what I think I should do, should I do? Yes. And that's much more scalable. That's an almost perfect analogy to ship management, actually, you know, most ship managers, you know, the kind of fleet manager job would look after anything from kind of 10 to 30 vessels. And certainly what we've started to see is, see in your leadership level, then the question is being asked, with the same team, can we get that up to 100 vessels, 150 vessels? And I think ship management is a full of edge cases, there's lots of safety, critical decision making. But ultimately, it is a very processed, driven world. And there's currently a lot of manual processes that live within that. And so that, you know, your stories sort of almost perfectly fits that in terms of taking the automation to a point where a human can judge it and also management by exception, I'd rather than having a human sort of pushing stuff back and forth. Yeah. So, you know, I think judgment is key, human and I think that's sort of where a lot of this is going. You can see there's a bit of an interesting problem, right, around the fact that if the client is to get the genie workforce and just have a sort of middle management, senior management doing judgment work, then at some point you run out of people that are capable of judging it because they've never done the work, they never know what's the right thing today. I think there's there's all sorts of ethical, existential issues with capitalism and stuff. Sustainability issue, yes, it's an interesting one. But I think Nick, you're onto something there in terms of where we should go with the discussion because I think there's also that thing that you mentioned there in ship management or the various layers through through technical management and crew management or all this where you are that oversight of lots of different, let's call them agents within the the workflow and being able to spot those kind of areas. That's one aspect of it where I think AI could be hugely important. I imagine that in most ship managers will come in and in the morning, they want to tell me what I need to know, tell me where I've got to kind of jump in what's my next best action. This is really areas where software can really thrive. But the second thing is you said it yourself Nick, the one of the things that's hampered technology adoption and people really digitization really living up to its promise in maritime is exactly the number of exceptions, the impossibility of putting any rules down. I remember throughout my career trying to map out like what's the rules here, what's the process and people will go and I think have I got it, is that it and then they'll go yes, but except except and that's when the whole thing falls down and you said yourself Peter earlier on that's where AI can be so effective because it can allow for those little mental leaps that says I'm making a judgment and that I mean what's your view from a maritime context you mentioned you had some experience with would you agree with it? Do you see it as a land of opportunity for AI? There's an interesting point there about that and a lot of the section handling stuff you go oh yes that is but right actually comes from like long long the tune of data right you know because three years ago you had this problem and that was the thing right AI won't forget right you can on a percussive basis you can have it logging absolutely every decision it's where I think it's going to be really interesting and medical particularly around that the you know you imagine a GP surgery somebody comes along with a okay they see one doctor nothing you know they go I was now they don't ever go away and then sort of three months later they come back and they've got some pain number else and they go oh it was this don't worry about it and a GP won't necessarily join all of those dots but an AI will just very quickly live at the notes and go no we need to send them for we need to go and send them for X-rays or blood tests on this group as I think actually due to X-Fine Z happening I can show you I can show you some pattern of the mission that means that this might be heading in this direction and I think that's where I think a lot of those exception become really interesting and just what becomes like data training for the model and I think that's you know you can end up with a lot more sort of domain specific models and capabilities that really help businesses the the the way the AI market is going at the right you need there is a much of a mode there for those companies you know it's all predicated around that point of view like somebody's going to win and then they're going to get a lot of SaaS subscription revenue from people using or leveraging their AI and you know there isn't much of a mode around the model itself and it is capability and it's thinking because it those have been open source to really quickly and you know there was a anybody saw the article the other day about and Thropic put out saying that the moonshot and deep sea could create it over 24,000 accounts and they were using cloths and trading and that's been and that's just an open source model now I mean it's up kidding kineatsy.5 and VLN5 are all you know they're they're ranking up there with opus right and so if you've got the hardware to run them it's for well you know you've got by the hardware and so really the mode isn't the model really all they've got the minute distribution and can keep asking you know people can consume from lump but again that's moving we and so on that you can say something. I was just going to say off the back of that what it's kind of two layers I'm thinking about this if you if we just take a you know a sort of mid-size shipping company that may have they've sort of got an evolving maturity when it comes to IT and there's a desire to start working with these tools. What would you advise the non-technical fleet managers do to start engaging with this stuff and then also what would you advise if you advise in the CIO what would you advise they do to engage with it but also manage all the non-technical people who want to be using this stuff yeah right rolling out across an organization there's a whole lot of the back and and there's a there's a couple of different things that I think over the next few years you're going to see a shift where you've got companies that exist as they update that are going to adopt AI or to try to adopt AI and what you're going to see there is that all structures not going to change what they're going to try and do is just drive more productivity with the existing structure that they've got. I tend to you know I tend to refer to to AI as a augmentation of people run replacement of them I'm sorry an interesting way of describing it seems like an exoskeleton that you can step into the sort of turbochargers your capability so you think you know same process is in place but more useful information and tooling available therefore means that the productivity can increase. I think where it's going to change over the next couple of years is you're going to get new businesses starting that structure themselves differently based around the way that AI is done so if you're assuming that AI is always going to be there in your business then how would you do it how would you design it differently for a product design stateholder management you are the development QA, customer service all there if you put AI first what do the humans actually need to do? Not what the humans actually need to do it's I mean you're going to need them what are they actually doing and where are you going to get the real value from that that's where it's going to change. So to go back to your point around you it's a real real world useful stuff. If you're a one man band out there and as to all this I definitely don't have a look at some of the tools and really lean into them and I don't mean just go and try it. If you're wondering whether I'm not a channel can't do something
I'll skip. You know, you'll start working with that and go to that social media thing. I'll get it to designs and social media, social media, post or maybe look at it and go, "Oh, that's not really how I would write it." And you start working on, "Okay, well, I'll take that and I'll rewrite it the way I would like it." Well, how about you go back to it with that one that you've rewritten and say, "This is how I would have written it. Can you remember that?" Yeah, I can do that. So that's the meaning of it. And that's learning a bit more about it. You started using it one way and then you've taught it to do something else. Saying again, the next step with that, you know, you then take that copy and paste it, put it into LinkedIn, done this now, schedule a bit on LinkedIn. So the way that you could automatically do that. Yeah. And then it'll work with you to help you do that. Very, very powerful tools. So I would highly recommend reading the mean to it. Right. So look at these tools, try them, push their capabilities. I, some of you hear about a lot, sort of people going, you know, it's dumbing down because they're not actually going out and researching, but it depends on the sort of person you are. Personally, I've learned a hell of a lot. It can explain things to me. I spend a lot of time getting it to explain stuff to me. And I would do you think about this and how have that come together and how could we do that? It's also massively aware of other products and tools and services out there that I was never aware of. Even from a tech standpoint, and I'm knitting this world, and you know, I know what companies are getting invested in. I know who's starting. I know what's exciting. I know what people are using. And it's still telling me about, oh yeah, you could do this way. I mean, all right. And it's it's really interesting. So lean into some of these tools and see what they can do for you. From an organizational perspective, you've got to look at it from a bit of a governance perspective. And I say, it's a bit of a boring conversation. I was going to say, we're doing the sexy fun actually using it. Do I start moving into further security? It's a governance problem because you give everyone access to a tool, and then you don't know how they're using it. You don't know what your data is going. You don't know what it's being used for, whether or not they're just they're acting as you're even in the loop, but they're saying yes. No one's actually done the training for them today. You need to read this. You're accountable for it. And then that accountability then comes down to employee handbook, process training, all that sort of stuff. Then I think the obvious ones to look at are co-pilot licenses from Microsoft 365. If you want to do actual AI integration from a development and data perspective, working with them as a year or because they've got the open AI and secure back end, so that you can use their API, you know that it's only doing it in the UK and the UK data center and it's not going anywhere. Same with AWS Bedrock. You can do the same with that, and you can run different models with them as well. So I think those are the safest ways of doing it. The benefit of using that like co-pilot is you can put all sorts of rules around, you know, who can use it. But all the nicer, you know, larger corporate tooling and security stuff there. But they also very quickly it hooks in really quickly to it can read all your e-mails. It can do, you know, it can access your file system, so it's got more access to your corporate data. The built-in tools, I don't know if anyone ever gets involved in supply chain questionnaires. It starts supplying stuff to a company and they send you a lovely debate in my life. Yeah, questionnaire every year. You answer these 400 questions, mate. About your business, if stated, the finances and then also information security, what you do with our data, all of that sort of stuff, then what we do, what we do is take all of the ones that we done over the last couple of years, along with our corporate processes and manuals and guidelines and trained a model on it so that we, and then the surface debt and the team's chatbot internally. So if anybody got a supply question that they could drop it in and it would, it would just read it and give them all the answers or they could chat to it and go, this customer's asking you this question, you know, so is the standard on this so that we would normally give to that? It's really powerful. And you can do that sort of thing really quickly with 365 and say, governance, see sort of a, it's not as sexy as I'd like it to be. I think everyone jumps when we think about the maritime industry, everyone jumps to really big difficult challenges like autonomous navigation and things like that. But actually, there's just so much administration that exists in the industry and so much of your life, if you're working in a shoreside role, is filling out compliance, questionnaires and all of this sort of stuff. And I think, yes, it's not sexy on the surface, but actually it's really what's going to move the needle in terms of productivity for an organization in a very short space of time, at a relatively low risk. And I think suddenly that that then becomes very, very interesting. I've spent a lot of my career day in effectively time and motion studies, when IT expected, and it's one of the things I enjoy a lot anyway, just looking at what companies are doing, where they're spending time, what's the race to people's time because ultimately, the human resource is human capital, and business, you know, being most expensive, also your most valuable, and you want to leverage them. You want them to be doing the thing that they good at, whether that's helping the customers, selling more things, buying the right stuff, you know, finding efficiencies in the accounting or whatever it is, that's what they good at. It's the human oversight bit, and everyone prides themselves on how good they are at doing that bit of work, and it prides themselves on how good they are at filling out form, yeah, it's just a waste of their time. So, using those tools to find that's really important. The thing is kind of, as well, identifying the bits where the human beings not really adding any value, as well, so just correlating data and, you know, joining up some of those processes, which are heavily, many like picking out something from an email, then adding it into a ledger over here, and all that kind of stuff, there's no value being added, and yet you're using their time. But I would say, I mean, I think one of the things here, and this is kind of the point I was making to the roller of technology, obviously the AI is being trained, the AI is acquiring more and more of those human skills and where humans add value, and I think one of the things to think about when you're rolling these technology is the fact that you're in many ways asking the workforce to make themselves redundant potentially in the way that they're training on these systems, you know, getting turkeys to vote for Christmas. And so, I think that I'm interested to know, like when you were talking to people about how they roll technology, about how do they take the workforce on that journey with them, so that people aren't feeling like this is something that, you know, they should be opposed to or should be trying to sort of put sand in the wheels of, do you know what I mean? Yeah, and it always get that. Tech change is far less about technology and far more about hearts and minds and process change and how you roll this stuff out across a business. If you go back to the point I made right at the start acquiring these companies, it's not about the tech, it's about making sure that people understand why we're changing things, how we're changing things, that they feel safe, that they're going to keep their jobs and where it's going, all that, you know, they drop my change, but to the better, right? And there's a reason for it, you know, the last thing, why is people sticking sand in the wheels, right? And really slowing it down. Yeah, you know, fundamentally work is going to change for a lot of people, but generally for the better, you know, if I go back to the point that I was doing process automation back in 2008, so, you know, what, it's nearly 20 years ago, I had the kind of same problem with the engineering teams then, they were sort of going, you know, you say it to take our job. No, it just, do you really enjoy spending time looking at servers trying to work out where the files are going to, you know, no, it's not, you don't, you enjoy doing, you know, more complicated debugging across networking and things like that, go and do that. Don't spend your time doing your basics that. And that's where it's going to go, you know, short term, until businesses really work out to restructure themselves, you know, this. I think there's the other thing to think about with that is, I personally would be worried being in an organization that isn't leaning into this stuff, because the organization that does is going to have such a big competitive advantage. The organization as a whole could end up, you know, being taken out of the market by a competitor that really leans in. So I think it's really important that at every level in an organization that we kind of lean into this otherwise.
other way, someone else is, someone else already is. - Right, and sorry just to say as well, to your point Nick about that's the worker as well, skills being left behind, right? You mentioned Nick with people who are tending some horrible old tech stack, if they then suddenly find themselves needing a job somewhere else and they've got their skills a completely out of date. And I mean, just to kind of, you started the program talking about the fact that you felt that you were falling behind. That will be a massive concern to loads of people, Nick. You and I are talking about AI virtually every week. If we don't fit, - I just took my eye off it for a couple of months. And that was enough to feel like, I mean, a different world that I don't understand. - Well, to that point, right? I've spent a lot of time talking about Claude Cade and Claude Cade work. - Yeah, other tools to exist. - Today, I arguably, I'd bought you up to mid 25. - I'm bought you up to today. - It's already me. - I'd say that. - Order it, stack with the book. - So you're just doing old stuff now, Nick. I got you excited about something that's out of date. The world fundamentally changed about four or five weeks ago when open claw got released. An open source tool for effectively agenteic workflows that sit over the top of the AI. The major thing that it boils down to is that something like Claude Cade or Chapchita Cade, anything like that's passive. It sits there and waits for you to tell it what to do or ask it to do something. You can schedule some work for it to do and it will go and do that. But it's only doing stuff on the teller to do it. What OpenCore really does is it brings in heartbeats and memories. Every 30 seconds, fires a heartbeat and goes, "Should I be doing something?" "Should I be doing something?" "Is there something that I'm working on?" "Something that I could look at." And you then through that framework, you can hook it into basically any data source you want. It's your email, it's your calendars, your social, your home automation server, your anything you want. Then it can then act on that and draw that stuff together. It can also do software development, server control, deployment. I saw a, if you have a look on OpenCore.ai, the website for the OpenSource project. And there's a showcase on there. And I'm a flitzer. I've been reading about it. I've seen stuff going on Twitter. And I had a look on the website. There was a showcase article about it. We're basically guys saying, "Yes, so I had a bit of an aha moment and the world is never going to be the same again after this." And he said, "I was working with it on a software app and all the tests were failing on it." So I got it to have a look at that, work out, log into the live server, have a look at what the problems were, understand where the bugs were, go and code a fix for it, push that into our source control, review it, I reviewed it, it then pushed it to project to live, fix the problem. I also noticed the UX bug in the tool. I discussed it with it, it looked at it, worked out a plan, presented the plan. I said, "Yes, it went and did it and deployed it." All done over voice while I was walking the dog. Yeah, I mean, time to look at this. And that's hours of work, for an engineer like a testing engineer to do. Still human in the loop. That is also something that I think is really interesting and that kind of plays into perhaps a discussion that we wanted to get into. You mentioned the UX. But what you're describing there, walking the dog, that's kind of like, space 2001 or whatever, style interface. You're just talking to this kind of sentient, or at least appearing to be sentient being, that's kind of doing all this stuff for you. And that, for me, when I think about the safest jobs that, if you've said two or three years ago, if you've said, "What are the safest jobs?" But I've said software engineers and UX developers, UX is going to be huge and it's a growing sort of industry. If you ask me now, I would put people off from both those things, because if I see the future as being an elimination of interfaces almost. Yeah, I was having this conversation with people about six months ago. If you look at it, you've got to fast forward a couple of years. Like, what, the internet, as we know it, what's the point and devices? Maybe our phones has been made them? What's the point? It's. Maybe our phones sort of exist so that you can use an app that shows you some structured data in a way that you then you look to understand it and consume it in a way that you want to consume it. It's just going to be data services with MCP, what a context project called running where, not a Bucco holiday. The AI will go off and talk to a couple of different providers. It'll come back with some, you know, his and course, so I can play with it. I've got an AI agent that runs daily looking at holidays for me, right? It's aware of the kids' school holidays. Right, it knows when they're on holiday, when we can go. It knows in-set days. It knows that I like to. And it's not just a basic, like, I want a seven days in tenorief or something on those lines. It knows that I like to play around with holidays to get good value where I'll play around with, rather than going for seven, we'll go for eight days or we'll get for six days or if we shift it to go on a Friday, rather than a Saturday you get a cheaper flight or something along those lines. It does all of that for me because I've told it what I want it to do. And then it just gives me, like, "Oh, yeah, here you go. This is a good option." Yeah. Wow. Mind blowing. But that leads us into a sort of a perhaps a broader discussion of what AI is really going to do to the software market. I mean, this is something you and I have talked about near on the shows, isn't it? Yeah, where does that leave SaaS? Where does it leave SaaS? Yes, there is a growing, this is the end of SaaS era. And then you get, you know, not many of the sales doors saying, "No, it's not." It will not. And you can sort of see both sides to this. Where is it going? Where is it going? If you go back to the point I made about what is SaaS, it's fundamentally it solves a domain-specific problem by putting some processes over the top of some data. Really, that's all it's doing, right? Does it matter whether it's sales, loss, up, supply? And, you know, the bigger the product, the more stuff it does for you. Sales force is a big tool, how's what's a big tool? What's actually going to happen? People at the moment are saying, "Well, I can just get it built on software." You know, I don't need to get paid for a SaaS license. I'll just get it to get and build some stuff, all these. Really, basic stuff, it can probably do it very quickly. If you've got a good engineering team to do that. Where is it going? You can probably do that even faster. Accelerate even further, what's the point? Do you even need an interface? Do you need a SaaS tool that helps your sales team use that every day? Is it even relevant? No. The data is relevant. And the AI model understanding your data in business is relevant. And then working with the AI for it to then surface what's important to you in your job and helping you do that is important. And how it's going to do that's probably not through a very structured SaaS app, because it's just not required. I think that's probably where it's going to go. Is it going to happen that quickly? No? One of my experiments this week was, I just said to it, I just needed to do this just to keep track of things I'm doing. I said this to Clau Co. I said, "Okay." And then it went off for a couple of minutes. And then he went, "Here you go." And it gave me a link to a web app. And it was like a really basic, just, you know, static HTML page web app, with my to-do list on it. And I said, "Oh, wow, okay." So then I opened it up in another tab and sort of typed something in and added it to the do-list. And I saw it wasn't syncing across browser tabs, because it was literally just a static HTML file. So it was not syncing across tabs. And it went, "Oh, hang on. Let me build a JSON file." And then we can store the JSON file and then it can call the JSON file anytime you load it wherever you are, and it works. And I'm still using it today. And that takes 30 seconds, maybe a minute tops. And I was expecting it just to just, you know, create a bullet point list. And it just went and built a web app, very basic web app. But I can kind of see that almost all of the value that's going to exist in the market will be in the kind of data layer. And actually having ownership of the data layer, how that data then gets presented in various interfaces is going to be completely customizable at the push of a button. - Okay, I think it's all going to change, right? I've said earlier in conversation about how, what is the actual defensible mode that these AI companies have got? And really they don't.
like two things that happen in the simultaneously, like the cost, the cost of the hardware is coming down and the open source capabilities and the processing requirements that you make of these AIs is coming down as you are, you know, so it's what does it actually mean for industry? It means that people are going to be running their own AIs soon for cheap. The tipping point previously, for just to give you an idea of how much people are leveraging AI. If you're a software business of a decent size, the tipping point is when your development team is spending more than $20,000 a month on API credits, that's the point that you would then go and invest in hardware and run that locally. And you'd be amazed in the amount of businesses that are spending that sort of cash a lot. And you would then go and spend 4 million quid buying some racks for your office. And that's a shift away from cloud, right? That's coming back to one. The tipping point this week was a few hardware provide AISIS have launched this AI super computer box. It's just under 3 grand and it will run effectively a frontier model 24/7 on your desk. And there's no token API subscription you've got to pay for that. It'll just run at one of these open source models. It's been trained on what can grow publicly. People will then get better at how you then train those data sets, how you work with them, because they'll have much more control on them, because they'll be used to running them locally. And then that's where things will really start to spiral. I think what you'll actually start to see is probably mass decentralization of AI, which then leads into power problems and has a good deal with that whole world. Well, that's potentially answers another point, because we've talked about this the other day that the compute power may end up being the problem again. So from a SaaS business point of view, you've got the SaaS businesses which are under threat from AI, but then they're also there. If they'd lean into AI and are really leveraging it at the moment, the computing costs are quite exorbitant. There was one example that we talked about on the show that you found, Nick, where the guy was basically the bigger his business got the more expensive it was. And we joked that computing is now becoming a restaurant model. In terms of the margins are going to go out the window. If I understand correctly in what you're saying about that decentralization and bringing that in house, you potentially solve that problem. Is that what you were saying? Yeah, basically. There's some other bits that have a stripe launched, and you've product. If you strike the credit card processing, SaaS tool that everyone plugs into, they launched AI token tracking tool for businesses. So if you built AI capability into your tool that people are using, you can make that consumable thing, but so that they then charge for. So rather than I've built AI in, then I either need to somehow bundle that into a monthly cost or work out, you can actually just say it's his the cost of token and we'll just build you extra per month. And that's built into stripes, stripes model now because it's definitely a problem out there. You say it like the bigger the business gets, the more expensive it costs to run the models. So you've got a hardware cost coming down and efficiency going up. And again, I guess that comes back to your point about people building their own custom models because it's become possible, but it allows you to only focus on the things that are really important to whatever output you're trying to get. Another one in the sense that most people were prepared or happy with the trade-off of saying, I'm going to put all my data into this application because I can then leverage all of the things that the application can do for me. The company is then sitting on all of that data which allows it to have some sort of real capital there. And almost like what you're seeing there is a potential breaking of the contract because if the company actually says, well, "Hey, I just want it to just focus on my information and be I want to have control of the servers and the architecture." And then the application itself becomes less relevant because I can either create my own mini apps which work on the data or I can get an agent that's working directly on the data. So for me, it is an inflection point, I think, when we will start to see some of these traditional business models break down. And that's exactly the point, right? I think I've just taken you on a bit of a journey there where you're now seeing it that it's going to be more about, what is the next generation of company and what does it look like? What is the org structure of that business? What's relevant to that business? How do you organize all of your humans around this new AI and data model to actually move forwards? That's where the big change is going to come. Yeah, I've seen in technology so many cloud mobile phones, when smartphones came out, there's such a big thing around 2008 and that wasn't the top-down change, it was a bot market change, right? It was in companies going, companies are going, we need BlackBerry when the iPhone came, I don't know, we can use BlackBerry enterprise server, we've got security, government was like, we've got falls encryption and stuff, and then MPs were going, yeah, but I'm using my phone, I'm using my iPhone because this is so much better. And then it changed from bottom up, not top down. I think it's going to be the same. You've got companies that exist at the moment, they're saying we're going to deploy this heavily governed model over the top of the way we currently work. Actually, there'll be a ground swell back up again with new businesses coming that are competitive or competitors that actually go, do you know what we can reorganize our entire business around how this works and we're going to change? Don't know how that's going to happen, yeah? Cloud, cloud was another one, and that was Turkey's waiting for Christmas, the amount of people that I went to sessions, they're like, well, business means that my job's going to go, so I'm getting that, but then to do something else. You talked a little bit there about like, what's going to be needed from business to evolve, but I mean, we don't want to go too far in the future because this is going a crazy lick. What do you see as the next big things that are going to be coming along and what do you expect to look like maybe in the next six months, 12 months? It's more a gentick AI, right? So, I think my life has not been saying in the last six weeks, since I've had that, it provides me with so much insight and access and capability to things that I can't even begin to explain how much it's having to my life and professional capability. The thing that you'll see about all this stuff is there's lots of people writing content out there, you'll see people going, oh, let me comment AI underneath and I'll give you my quarter of million dollar per month playbook for how to build marketing agency that they've made. There's a lot of rubbish like that out there, really, the people that are, people that are really benefiting from it, that really see it as a competitive advantage on talking about it. They're just getting on with it and going, I'm just going to say quiet about this because it's killer. So, if you look at it, the AI industry in general sees that, the guy that wrote OpenClaw, he recently, I don't actually saw, got accrued by OpenAI. So, it was an open source product, he was quite happily bundling along doing that, and released it and then they came and acquired him for a billion dollars. Let's say for 400 million in cash, he's 600 million in stock, it's not bad, not bad deal. And because it's open source, they weren't even buying OpenClaw, they were literally just hiring him, and he's gone to go and run their agents at the program. So, doesn't that mean he's the best guy for it? He's done a brilliant job when it's an amazing thing and they've also got an interesting thing, this product everywhere and people are using it. But it's open source, I've used it, I've got Claude Cade to rip apart the source code for it and take bits out of it and plug it into other things I've been doing because it's useful. The way it does, instant messaging, inspiration, and telegrams, really useful for something else I was doing, it's like, oh yeah, let's do that. What you're describing is, we started out about, I think it was three or four months ago now, I think we did an episode on the Super Worker, on how AI was going to charge the Super Worker. What you describe in the way that you've got that set up in the last six weeks sounds very, very close to what we were riffing about then. The thing that I think is kind of, my only comment would be, I kind of feel like it's almost more like the Super Worker is a freelance worker in a sense because it drives. Yeah, it's something that they haven't touched on. It's the, you know, the assumption is that if you give AI to a very capable person, then they're going to be 10 times more efficient. And it doesn't, it doesn't hold at all. I've tried to do this, you take a team of 10 developers and all very, very capable, give them AI tooling, 50% of them will get 10 times faster and 50% of them go, "Don't I get it?" It's not, not my bad, I don't like it. It's non-deterministic. I can't do it with it. It's just inference. It's, you know, it's gaslighting me. It comes down to a sense of personality. I think a way of describing it is, if you hand something,
somebody a hammer, you look at what does this do for me? You could say, what does this do for me this hammer? Or what could I do with this thing? And then you could start hitting different things with it and go, oh, this is interesting. Oh, I find it's quite useful for paying up shelves or hammering nails in. And I didn't realize that you could make a hammer screw in maybe. Oh, no, that's not great use for a very, very long. That sort of mindset works really well with AI because if somebody says, you use AI for this and then they get and try and do that and they don't get a great response. What's the point kind of at doing it again? Somebody says, he's a really interesting tool. Can't try and use it for absolutely every aspect of your life. See what it can do for you. Not everyone's going to do that. That sort of mindset that will win. And if I may just one follow up question, they're being educated for a world that's going to look very, very different right by the time they leave school. What they wouldn't say things that exist in the real world. Human to human, social skills, psychology, although to be fair, actually that's slightly getting them some dissimilar behavior, and plumbing, carpentry, building, all of that sort of stuff is product build. I think there is an interesting point about education in general. I don't know if you've seen some of the studies that they've done in the US with them. They're very basic level. They've got a class and gave them 50% of them the normal textbooks and said that this is the textbook they're going to work through and you're going to do your home work from this set week by week. And then the other 50% they got them to each of them to write their bio about themselves and say what is it they enjoy, what the FHTV, Pro-Bound Sports Shades, what interests them, what didn't interest them. And then they produced an AI-generated version that was very specific of that textbook for them. So the maths question might be about their favorite formula one team or might be about their favorite footballer or something. They did a lot better. It was easily. When I saw multi-modal AI in general the first time, I'd recommend you Google it and have let it. It was a dinner that I did a couple years ago. I think it's removed on that right and not. It was just a multi-modal means that it can do video and talk and audio, all of the text, all at the same time. Different modes of engagement. But it had a camera over the top of a desk and it was somebody that they put a piece of paper down and they started drawing something and it was like what's the other side. It's a duck and then they color it in blue and then I was like, "That's where ducks aren't normally blue." But then the thing that they did was they put a map out on the table and said, "What's the exit?" So it was the map of the world. So can you come up with a game, an educational game for this and said, "Yeah, okay, where did Kangaroos come from?" And they pointed to Australia and it's a bit tick. And it sort of played like that. My daughter has dyslexia. So yeah, but she's very capable, but words can be a bit of a barrier to her. But she can talk about stuff and learn and absorb through doing and watching and listening. And AI is an amazing tool for that. Absolutely amazing tool. It can teach her about anything, in a very personal way for her in a way that she understands and it can gamify it in real time. It can do all of that stuff. I think it's, if you look at the send crisis that we've got in UK, actually, they leverage AI in the personal basis. That always goes away. So can work out with doing it. People talk about it as AI is bad because you can cheat. Yes, of course, you could ask it to write your essay for you, but that comes down to the person that's using it. Do you actually want to learn or are you just trying to gain the system? And there will always be people that want to gain the system and they don't want to learn. But I've used it with my daughter for, who she got the link in the school show and she had a long script to learn, which you can imagine is quite complicated thing there. And I wrote a tool that she could then practice the parts through voice with it from her iPad. So she could, it played all the other parts and then when she spoke, it helped correct her. And you didn't get that bit quite right. And she learned it. Wow. That went fantastic. And I could say to you know, if she's doing a bit of maths homework, but just let me carry it. It's not like I can't do it. They've changed the way maths is done. And I can then say to it, yeah, I'm working with with my daughter, you know, and she's doing this bit of homework. Can you talk her through how to do it in this method, but then get with the answer? I know her more than happily sit there and wipe through it with her. So yeah. So stuff like that. Three years ago would have been two years of a team of five developers and a research team to enable, you know, to enable that kind of app to exist. And now you've just been able to do it in your spare time. Not even concentrating on it. One eye on the secret on the TV program, one eye on my phone. Peter, I've got two other questions for you. One, would you be willing to come back in like six to 12 months and just give us an update? And then two, you've been very, very generous with your time, both to me personally and also to to our audience here. Do you want to give a bit of a plug to either yourself or to beluga? Well, well, thanks very much. And so the maritime industry may be aware of beluga, you know, you launch that product originally back in 2020, sort of invested in it back then. The it's a bunkering platform. So helping it was designed originally to help people find better pricing for marine bunkering. I think overall one of the problems with it, there's other things that need to be solved. And also found that although it was a this lovely design web app and tool that chat and everything in it, this market typically operates on what's happened in the messaging and telegram and ice. So, what I've been doing with the team is the whole product has been completely re-engineered to be chat in AI first. So it lives entirely inside of WhatsApp and in telegram. And it's going to be lead gen for the for the supply side of the market. We're gearing out to relaunch that shortly. So quite quite excited about it. And it's interesting to see where it goes, what it does. It's getting back to that point around the fact that you can write bad software very easily with AI, but actually writing good software still takes time. People are launching that and that's a bit of that, do you? Awesome. And where's your people reach out to you if they found this interesting and want to pick your brain? Yeah, sure. And LinkedIn is easiest, best place for Peter Rossi and also it can email me,
[email protected]. Yeah, more than happy to chat to anyone about that stuff. I find it fascinating and I just love solving problems. So that's been. Well, you solved the problem for us today. So thank you very much for that. I really enjoyed it. Amazing. All right. Well, we'll hold you to that. Thank you so much. We really, really appreciate it. And yeah, you've opened our eyes and I hope you've opened the eyes of some other audience as well. Thank you very much. Thanks all. Bye. Bye. [Music]
PodcastTranscript AI