**[00:00:00]**
Armando: Hello everyone, welcome to The Web Talk Show. My name is Armando Perez-Carreno and with us today is Ross Barnes. He's the founder of the Galahad Group. Welcome Ross, how are you?
Ross: I'm good Armando. Thank you so much for having me and it's an absolute delight to be here.
Armando: Oh, it's my pleasure. Especially before getting on air we were talking about a little bit of your whereabouts and how much I love that area. So I'm very excited that we're on the show today also long distance which always continues to amaze me and it's always good fun. Ross, can you tell us for all listeners a little bit about the Galahad Group and also how you got into this?
Ross: Sure. So the Galahad Group I founded 3 years ago and we offer AI consultancy, AI aided platform build and AI enablement for organizations to ensuring that organizations are making the most of what artificial intelligence can bring to them. How I got into this is before the Galahad Group I was a global CTO of a WPP advertising and media agency which under my tenure on the board grew from 15 people to 1500 and we were billing over 1 billion in media billing. So a big company but also realized that service businesses are incredibly threatened by AI. So the adoption of AI for them is really really important and wanted to set something up that helped organizations not feel scared of that threat but really embrace it and be first mover.
Armando: Wow, that is I mean if you look at it from an altruistic point of view, right? It's like you want to help business owners. I'm going got that of course it's a business, big business, but I like that concept because a lot of it is focused the energy is focused on enterprise and these huge corporations and then the rest of the backbone of the economy, which are like service businesses, etc.
**[00:02:12]**
It's yeah, people are confused, they're scared, they don't know what to do, how does it work, is it good, is it bad, is it evil? But at the end of the day, I mean, it's a tool, right? As you as you will know.
Ross: 100%. It's I mean, every technology that's, you know, driven a revolution of sorts has always been, you know, is the tool is the tool bad, you know, is the tool evil? And it's all about who it depends who's holding it. As with all of these things. Adoption and enthusiasm and curiosity are going to be the tools that that win the day in this era, I think.
Armando: Yeah, I agree. I think that a lot of bad press is put on it, but the same way that bad press hit the internet and the smartphone and the computers and the industrial revolution. Like everything when it's when it's as big as something like this and game-changing, people start attacking it from all angles. And at the end of the day, I mean, I was just looking at a comment yesterday in one of our shorts because one of the guests mentioned something about AI and immediately that comment was one of those like "Oh, you're talking about AI, I won't even listen to you anymore." or something like And it's because they see it from whatever they perceive from the media about it and have no experience with it and just think about it as this thing that will take jobs or is taking jobs. And I always like to see the other side and say, "Yes, well, it's like the internet. Yes, it did take jobs at the time, but can you imagine the number of jobs that it created after the fact? So, all that we use today couldn't be possible without it. So, it's If you look at it at a long enough time horizon, it's a good thing.
Ross: And look and the a reality that will, I believe, remain a reality is people buy people.
**[00:04:14]**
People want to talk to people. People want to have, you know, interactions. And I think to your point about perception is reality. There was is without a doubt a you know, a media groundswell that is, you know, is making AI seem a terrifying thing. And I think if you think about it long enough, you can certainly put yourself in a place where it's quite scary. But, we're not there yet. And doesn't that mean, you know, I think getting on board and finding out what you can do with it, rather than shying away and letting a competitor or a colleague steal, you know, steal yards from you.
Armando: Oh, for sure. That's an excellent point. Many businesses, especially business owners who still run their the day-to-day, and they if they don't embrace it and just stay out of it, out of ignorance, out of fear, out of whatever, they don't they don't they just don't want to try it, their competitor, like you just said, they're using it right now. And it's just even if you don't use it for because this is one of the things that people complain about, creating AI slop and content, like take all that aside. Just using it for helping you do a repetitive task that your staff hates doing anyway, and takes a long time, and it's prone to errors. And have it just do it automatically quickly and give your staff something else to do that's more engaging and actually uses our human senses.
Ross: Well, that's it, right? I mean, that is it. Remove frustration, remove friction because AI just it doesn't care what its task is. It will just perform its task. And therefore, you can give your employees, your colleagues, your clients just a much more enjoyable experience by just genuinely being more human rather than you know, on a on a conveyor belt.
**[00:06:24]**
Armando: Yes. I when I when I think about support and a lot of these companies don't have support. Like just companies in general. You don't like you see a contact page, a support page. You shy away from even sending something because before AI I yeah, like nobody's going to answer you. Now with AI, a lot of them can if they want to, they can have an agent that can help you with things that are common that might solve the issue quicker. And some might do it well, and some might not do it well, and that's okay. Like there's people who are not going to do a good job at it, but they're trying at least. And if you like to think about it from from also the business's point of view, not not only the consumer, you have to think about like if you want to support hundreds or thousands of customers, you're going to need a lot of staff. And it's not that you want don't want to hire staff, it's just that it's not feasible for a certain size of business to have a hundred reps on the phones all day answering questions that have already been answered because they're generic questions or they're account questions or they're things like that. And again, you can have a voice agent, a regular agent, whatever, to help just take some of that off and still have your team there. I, for one, think it's a good deal.
Ross: I completely agree. And let me give you a flip side view on that. Is that what AI is the enemy of bureaucracy. And this is something that's happening in the UK at the moment, and there's quite a lot of research. I think I posted about it on LinkedIn this morning. Is that complaints to government bodies, since 2022, the advent of ChatGPT, have increased exponentially. Because people are now able to go through the bureaucratic steps that an organization may have put in place to make things difficult easily, because the AI will just do it.
**[00:08:26]**
And if you're not if you're not responding to that as an organization, then pretty soon a snowball is coming towards you where, you know, and literally, I imagine with the UK government, the paper is stacking up on your desk as more and more consumers, because AI has got such a low barrier to entry, more and more consumers are saying, "Do you know what? I'll just I'll just get ChatGPT to do that for me." Thanks.
Armando: When you see people doing this in the UK, specifically, are they using these tools to allow them to go through, let's say, forms that would have otherwise taken them a long time to fill out versus taking a long time to fill them out?
Ross: Except, I was speaking to someone, recently, about they had a HR complaints about a colleague. And they brought the complaints to the HR department. And the HR department reverted with, "Okay, well, you're going to have to find, significant evidence for this. And here's the form you have to fill out. Here is the evidence you need to find. Here is the amount of evidence it needs to be. And this person had Copilot on their email. And just gave Copilot that form, the body of evidence, and there are the emails. And HR, who normally Well, I'm not saying this is definitely true, but would have hoped for that to be a deterrent for a really large process, have suddenly immediately got the evidence, and they've got to do something with it. And, you know, that is indicative of any customer service business that is trying to put hurdles in the way of customers contacting them. It's not a thing anymore.
Armando: That's incredible. When when I hear that, I immediately think of like obviously there's people saying, "Oh, you can't give it access to your email because then it can blah blah blah blah blah, right?" But at the same time, like depending on the tool that you're using, you still have to be careful.
**[00:10:39]**
But some of these tools will not necessarily have to read your email every day automatically. It's just that they have access to, and you can tell it to do something. And in that moment, it will go and do the thing like you would have done. It's just faster.
Ross: Exactly. Exactly. It's not as if it's actively scanning for something. It You say, "We've got this particular task. Here's the things I need you to look for. Can you read and look at the relevant emails as well?" It's not going to read everything. It's very It's very And we must be careful about not making them people, but it's very clever. Yeah, it doesn't want to consume all of the information. It only wants to consume what's relevant.
Armando: Let's dive a little bit deeper into that because you're an expert on this topic. And people when they hear email, for some reason we imagine, "Oh, I gave it access. Now it read all my email." Come on. It's not capable of reading all the email. First of all, because it can't fit in its context window, but also it would need a huge database. You think about your email inbox size, gigabytes and gigabytes, and it's just not not practicable for it to actually go and read everything, but instead it will do searches and do things similar to what you would have would have done. Is that correct?
Ross: It's absolutely correct. It's similar to, you know, doing a Google search, but it's more intelligent. I mean, we've got to think of these AIs or agents as they are they are programs that are designed to complete the tasks they are set. They are waiting to be told what to do. And sure, you could say scan my email every second. Sure. But you know, what's what's that going to do for you? But you can give it tasks or triggers. If an email from this person comes in and if an email from Armando comes in, I need I need to see that immediately. Give me that. So, it's not reading, and it also wants to complete the task as efficiently as possi- as possible. And that is not reading every single email.
**[00:12:41]**
Just like us. You know, as you know, as humans, we see an email coming and go, "That's junk. I'm not going to read that." It's similar.
Armando: And even as software developers, I created a tool for myself because the like tools for managing people and projects and things like that. There's There's tons of CRMs out there and things like that, right? So, I have a very specific process that is different from a corporation or from many types of businesses. And so, I created my own little tool called Janus that I just use internally. And it has a part of it that reads email. But even as I built that, I told it just like the only things I need you to process are the inbox, because I clear my inbox. My inbox is always empty. And I read every single email. You're a better man than I am, Armando. I know very few people who do that. But I like to keep it that way. And this tool allows me to do that. And when I clear something off, it'll it'll put it in the right buckets for like tasks, waiting on someone or myself or things like that. And that way I don't have to use email like as a task management tool. And but still, I have it only read that inbox. Not everything because everything else is out of context. I just need you to process this, do what it needs to do, and that's it. If I ever need more information, it can go and fetch things. But in reality, you set it to something that's going to be practical because things cost at the end of the day. Tokens cost, memory costs, bandwidth costs, right? So
Ross: I mean I mean if you apply that to how you might work with a colleague, that's exactly how you would work with Yes. A person. You would say, "This is what you've got access to. This is what I need you to look at. This is your defined task. If you're doing that, then you are performing a good job. Rather than oh my god, can you just know everything about everything? Because that isn't that isn't helpful for either you as a user or an AI or a colleague as a as a task.
**[00:14:48]**
Armando: My friend, Mike Carlo, he runs Power BI Tips. It's a lot of
Ross: Oh, nice.
Armando: Power BI related fabric, that sort of things. And we chat a lot. And one of the things that that he talks about and we talk about together is when you're trying to implement AI in a company, you have to think about just like you just said, as a new recruit. Somebody comes in, you're not going to give them the keys to the kingdom. You're going to give them a an email address, and you're going to give them a specific role. And that role can only access these parts of the system and you have control of that and you can see the log of what it can do and can't do. And if you do it that way, then it's very secure because even if it wanted to do something, it's limited to what its role and capabilities allow it to.
Ross: I mean there's so much in that to unpack because you're completely right. So when I do one of the most frustrating things I hear when I talk to companies about AI enablement is it's like we've given them all Claude and surprise. So why isn't it why isn't it working? And it's like well that's not what it is. You can't just give everyone a hammer and expect them just to be finding nails everywhere. So when I train people, we spend you know, the first day of a three-day course, we won't even be touching ChatGPT or Claude or Copilot. We will be talking about understanding workflows, understanding process because you can't truly implement AI without really you know, getting very close to organizational design because I think one of the big mistakes businesses make is they try and overlay AI onto a human problem whereas actually an AI might want to do it a very very different way and not follow the same process that you know, a naturally inefficient person might might follow.
Armando: What's very interesting is that recently that changed quite a bit.
**[00:16:56]**
Before you would need to be very specific on how you wanted to proceed and do tasks and now it matches just what you said where you'll give it an outcome and if you define the outcome and the input and output properly, it will find very interesting ways to solve the problem that if you just told it step-by-step, it might have been less efficient, had more errors, etc.
Ross: But I And I think I think you're absolutely right. I think there's something super crucial in there that people are I feel missing is that it will not necessarily explain that process to you.
Armando: Oh, yeah.
Ross: Therefore, your process the work of that workflow in your organization becomes opaque. So, therefore, understanding what the workflow is and then understanding what the AI does with that is super important. Otherwise, you know, I've been in so many training where someone like I'll say like, "Work through your workflow like this and figure out what goes in where and what flows into what." And then someone will come back to the, you know, the show and tell and say, "Oh, I'll just ask Claude to do it." And I said, "Okay. Okay, brilliant. You just asked Claude to do it. Okay. What happens if this goes wrong?" And they have no idea. You know, they aren't able to track back any errors. They aren't able to you know, look at guardrails or risks because yeah, Claude will do it for you, 100%. And yeah, and it will it will usually get to the right result, but usually isn't good enough.
Armando: I think that the overlying issue with just telling it to do things and stepping back is that if whoever's doing it has never done any sort of engineering or design thinking they can accept it as is.
**[00:19:01]**
But there will be so much context lost in edge cases like you're saying where you don't really know what it's doing. And if you ask it to explain it will explain it in a very strange and convoluted way where you're like, "What are you saying?" Even as an expert in software.
Ross: Yeah, yeah. Imagine having an employee that did that. You gave them a job and they went, "Don't worry, I've got it." And just was turning up with the right answers all the time but then one one day you got something catastrophically wrong and you said, "Why did you do that?" And you know, they answered in a way that just you couldn't decipher. It's the same thing.
Armando: I like particularly telling it in the rules and we might want to go into how all this works, but sometimes in the in the system settings or in a file that explains how how things should run, etc. I like to give it some guidelines into how it should respond to human requests such that when it's explaining something to me or to a client, which might be different and I'll and I'll tell it to and it will explain it to me in a way that I understand. And it will explain it to a client in a way that they will understand, but it will no longer explain it in the way that it typically explains because it's super unclear if you just take the out of the box way.
Ross: Completely. And well, my AI that helps me run my business, it has like a it understands all of my colleagues, all of my clients, and it has an adaptive style that is built in. So, it analyzes the way that they communicate with me, the way that I communicate with them, and then builds whenever we're sending something. It knows that client X really likes to the point detail.
**[00:21:04]**
You really Whereas, you know, client client Y might like to be asked how their weekend was. And that kind of thing. You start to build these communication profiles. And then you start to think are all of these systems that we're going to be using now as consumers, as professionals, going to be adaptive in that way? Are we start Are we going to be able to start you know, rather than going into a settings and picking things, are we going to be able to start systems adapting just based on how we communicate with them? And who we're communicating with. And I think that's quite quite a profound thought.
Armando: It is. When I don't know why this happened, but when they started adding this these extra rules to Claude, for example, for watermarking, for for EU regulations. I started finding that though it was quite easy to spot when Claude was talking before, once you read enough, we're very good at pattern recognition. So we sort of figure out as humans. You catch on pretty quick. But when they started making those watermark changes I don't know about you, but I now it's just like oh, please no. Like because it's being put into a box of the potential that it can do to write, so that it can agree with those rules that have been put in place. What are your thoughts on that thing?
Ross: And couple of things on that, I think. I think it is your responsibility if you're using an AI to produce content to give it as richer context on how you as an individual communicate. And you and you can do that. And you know, your emails, your posts can appear as you, as close to you.
**[00:23:05]**
But then, and I spent a lot of time doing that, and then, you know, people would ask me, are your emails AI? And it was like, well, of course they are. Because that's just an easier way for me to do that and get information to you more quickly. The actual question
Armando: Yeah, it's like, are your emails spell checked? Well, of course they are.
Ross: Right. Yeah. The question you want to be asking, was the information contained in that email factually accurate?
Armando: Yeah.
Ross: If it wasn't, then may have a problem whether I've written it, whether an AI has written it, whether my dog has somehow got hold of my computer. Yeah. If it is accurate, though, it does not matter who has written that email, yet security aside, it all that content, it comes from the person and the desire to deliver that information. It is purely a medium.
Armando: I agree. And also, you very likely read it and prove it, or have done it enough times that you now trust the system to be accurate enough to send it. So, it's not like you're just blindly Some people do, and I that I'm not, of course, I don't agree to, because you get these emails, you're like, you didn't read any of this. Like, why did you just send me all the whatever the thing spit out? But, in the context that you're talking about, yeah, I think it's perfect, because it'll give people the right answer, exactly what they need, to the point, in the way they want to hear it. And if it does that, I agree with you. It just like it checks all the boxes, and it clears up both people's time to continue doing what they need to do.
Ross: And you know, and I don't mind now if I receive what is obviously an AI written email in response, because then I go, okay, cool, you're probably not reading what I'm saying. But therefore, now I no longer need to read what I'm You know, we can just we can just let them talk to each other, because that's where you are. And then, you know, we'll figure it out. I'm My systems will judge an email's content, you know, based on risk, based on importance, sensitivity, and then will surface the ones I do need to check.
**[00:25:07]**
But if something is, you know, I'm on the 7:06 train and I'll be home at quarter to 8:00 to my wife, doesn't need to be checked because, you know, that is, you know, factual factually ac-
Armando: When when these tools keep getting better, it's I'm still blown away. Like we used to create parsers and things for to be able to read digital information and or physical information. And it was a mess because you had to have all these different scenarios of how things are coming through and can we parse them out there errors and can they inject things, etc. And but these these newer models and agents and harnesses around them are so good at determining that sort of thing. Like even the language, like how is the person speaking? Is it obvious that it was just given by an AI? Did they just use it for spell checking? Did they write it themselves? Or likely, did they anything. Like and the context like you're saying, like did is this important? Is this clearly a sales pitch? Is this something that I irrelevant to me because they misspelled my name? Or did they use the wrong person's name because they're just using a scraped list? Like they figure it out right away. Like it just read your email and just like, "Oh, Okay, so you can get rid of all the noise if you're getting 150 emails a day and you just get the ones that are important, but they're really good. And I think you can tell me from your expertise, but sometimes people are afraid of, "Oh, I don't want it triaging my email because it'll miss important things." They are very good if you set them up properly at like literally figuring these out better than yourself sometimes. And when it has doubt, it can tell you, "Hey, I'm not sure about this. What do you think?"
Ross: And you and what's amazing about the doubt part is the doubt part becomes self-learning. As you know, as long as you keep telling it, you know, "No, you could have deleted this one." Or "Actually, this one should have come come to me." It'll go, "Right, cool.
**[00:27:10]**
I've got that." So, therefore, the next time this happens, I will perform the action that I should have performed. And you know, that just makes you know, that just makes a system that is self-learning. I think you know, interesting thing about parsing and also your interesting about your own bespoke tool is something to be really aware of is that you know, now well, not now, but more I think it's more well-known that these tools are now getting very good at talking to machines. And what is interesting about that is that they are able to break down what were previously locked down systems. Let's not say anything nefarious. Let's say on consumer hardware that you know, the guy that you know, had reversed engineered his Amazon Fire TV to allow you know, the firmware to be you know, what he wanted it to be. You know, or being able to produce PowerPoints by a direct access to PowerPoint. And then you start to think, "If these machines are talking to machines all the time, what is the role of software? Is the role of software something for you and I to use? Or is the role of software to talk to the AI that I use on a day-to-day basis?" And that just changes the whole equation of to your point, I want to build a tool that does the things I need it to do, rather than buying a tool and kind of trying to make it work around around hopefully.
Armando: You'd love Mike. We were talking like just about that a few days ago. We he was bringing up the point of like should we then as engineers start building CLI tools more than more than UI's for for many things. And I was like, "Yeah, I found myself a couple of weeks ago creating a I needed to send some some credentials to someone.
**[00:29:17]**
And typically you either use a pastebin or something like that or you have a password manager that has that software. And so I created one specifically so that my agents could use. Then it's all CLI based. There was no UI, no nothing. It was just a way for for my customers to send me their credentials securely and I can only unlock them on my machine with a specific key. And it's all terminal based, but the agents can use it very nicely. They don't see the passwords or anything. They just just move them around. And so then I thought, "Oh, now that I have that and I didn't lose time on UI, I'll make one that can send them things." And that's when okay, I can do like a little UI so that they can receive it via email and click and get it whatever you want. And it's all client based. But that thought that you just said about should we be making software for machines to use is it's amazing because it can have so many features and there's no time lost in oh, is the layout good? Does this look good? Are black or white? Is the contrast okay? Does it fit in a small screen? Like just forget about it. It's just going to read the code anyway. So
Ross: You are right. I mean this is just absolutely my biggest passion point at the moment because I think what is hard for I think the I won't say the general populace, but a lot of people to understand is that it's this process that the machine does not care. Right? So, your point of that is the that's a By the way, I'm going to build that because that's a lovely a lovely And also, that's the thing, right? You can just have a chat and go, "Yeah, I like that. I'm just going to build one of those." Which which which is crazy. But, you know, normally someone would want to go, "Okay, that's a really cool idea. Can you show me how it works?" And it's like, "Well, there there isn't there isn't anything to show you. It's just happening. And that's quite that is quite hard when you're trying to demo, you know, workflow solutions because there is no UI.
**[00:31:22]**
There is no lovely little app. And I think it's getting into that system-based thinking that I think it's holding a lot of businesses back because they're thinking, "Okay, so we need to produce something. We need to make something." And well, you just need to solve something. And you know, and that's it.
Armando: And there is a technical part to this for for those listening who care that if you have your AI browse a website or use an application, it has to take screenshots. And those parsing those screenshots take a lot of tokens. So, every time it does something, it has to go and some equivalent of this and read the screen, understand it, use tokens for that, do something, read the screen again, etc. But, when it's using a command line tool, and sorry I didn't explain it, CLI is a command line tool, then it can just do the actions, get the output, do the actions, get the output. It could do it in milliseconds, first of all, and it doesn't waste tokens on trying to understand what it's seeing because it doesn't need to at all.
Ross: So I mean, I've this has happened to me today. I have had a I've got a mesh mesh network here that is where and we're quite out in the country and it's quite unreliable. And you know, you just have a thought and go, I wonder if I could give the credentials to Claude, it could go in and just build an MCP. You and by and by an MCP for your listeners, a model context protocol is the way that a AI will talk to a system in a very similar way that Armando explained. And like you said, it's just gone on, done exactly what it needs to do, found the problem, read the logs, and everyone's got lovely Wi-Fi now. And it's like it's like it's almost a world where if something is annoying, just point Claude at it or point ChatGPT at it and you know, see what can be done.
**[00:33:30]**
Armando: To your point earlier where where you said, oh, I like that idea. And the bringing that to your earlier point of do we really need software, right? Because now it could it's really about the ideas. And I find it brilliant that you could say, hey, that's great and I could say, oh, one second and I could just tell my agent, give give me a quick prompt that I can give Ross so he can build it. And it'll give me a little MD file, which is just text to send you and I don't have to send you source code, I don't have to send you anything. Just with that, then your agent is able to build it the way you like building things and it'll finish in like 10 minutes and you'll have the full thing. And like, how it's very meta.
Ross: I was in Belgium last week with the client and one of their head of AI's we ended up watching some football in a bar and I think it's really interesting and it's like you and I, right? When you get people who are very passionate about this kind of thing, we share our ideas freely. And not only freely, but for free. You know, the open source market is, you know, it's exploded more than it you know, it ever will. But what you think about is what does that mean for commercial software development? If you and I just going, "Hey, that's Do you know what? That's cool. I could add that to this thing I've built. You could add that to this thing you've built." And we've solved a problem that maybe a business might have nee- needed to be spending thousands of, you know, dollars on a license.
Armando: Exactly. And you find very interesting use cases after the fact. So, like, with Mike, we do this a lot. We're sharing We're like, "Oh, did you see this tool that I made?" "Oh, that's great. Can I use it?" And then now now we have an a GitHub org that we share.
**[00:35:32]**
And we'll we'll put projects up there, but then I'll add stuff to it or he'll add stuff to it. And so we collaborate with software that we're both using separately with different configs and things, but that we Sometimes I'll get something that I'm like, "Ooh, that's nice." And now I have this new feature because he thought of it for his channel. And then I have something that he's like, "Oh, that's nice." Bec- And now it's just there. But it gets to a point where suddenly you have this great tool that maybe some people want to use. And you can sell the tool because you've already solved a bunch of things by using it yourself.
Ross: Yeah, but then you've got to, you know, you've got to commercialize that tool. Whereas you could just say, "Hey guys, yeah, look at this great thing." And that's And that's what people are doing. I think a key part of what I do, you know, the Galahad method of training is I when we don't when we're not touching the tools in the first couple of days is I try and get people to think that kind of everything has a shape. Everything you're trying to do is a shape or series of shapes. And I've got six shapes that I tell them about what you prioritize something. Find something. Write something. And then when you start to think about the things that you know, you've built and you might say, "Well, I've got this thing that you know, looks at my emails, it prioritizes them, it then drafts responses, and then it sends emails." I go, "Okay, so let's prioritize, draft, action. I need something that looks a bit like that. I can take your tool but adapt it for my specific use case because those shape There's things we're trying to do. You know, it's just input and output. The process, they all look the same.
Armando: I like that. If you look at things like shapes, then it's better for people who are not used to finding patterns to now be able to say, "Oh, there's only six. There's only so much I can put things in." And so it's very easy to just like, "Oh, this is my process.
**[00:37:36]**
This is what my staff does. This one fits here." And it's almost like a building block thing. I really like that concept.
Ross: Uh-huh. 100% and what we'll do is we'll we'll we'll work through a what appears to be a very complex workflow with you know, a very senior executive and we'll go, "Oh, actually, that's a 2 3 4 1 5 2 1." There you go. And then they're able to then you know, go into whatever tools they've got access to and start to build that in a in a very simple way. You're not you know, you're not talking about connectors and MCPs and plugins and APIs. You're talking about things happening in a process.
Armando: It sounds like music when you're when you're playing, especially if you do improvisation in jazz or other types of music where you have these chords and you have progressions and you have scales. And it's all, just like you said, like you can sound really complex, but then if you look behind the scenes, you see the musicians, they're just doing like 1 6 4 5 1 8. Like they're they're just going through those patterns and it works that way. And I think as humans, it sort of resonates.
Ross: That's quite nice someone that might see if I can make people's use cases into little tiny pieces of music for them. That would be quite cool.
Armando: Yeah, yeah. And it's the other day I was watching, I enjoy all sorts of music, but particularly, I think gospel music, like church music. It they the one that has really powerful like bass and chords and drums and they're very talented musicians. And if you look at them, because suddenly I'll watch a video and then YouTube suggests a bunch of them. And so I'll get these people that I've never heard of, but they're behind the scenes, they like do these little videos or shorts. And you see them doing that live, like when they're playing, the it's they don't know what the pastor's going to do sometimes. And so they're just following and then there's a leader and he's just like shouting things to the to the rest of the guys, like two, five, three two one.
**[00:39:41]**
And then they like they and then and then they know exactly what to do because they've got those structures that now they can perform.
Ross: And what and what's amazing about, I think, music when you view it like that and if you view it in this way is there is only a finite number of places that that can go without being unpleasant. You know, unpleasant to listen to and also unpleasant to play or and I think if you think about that rhythmically with AI, it's a similar thing. You know, you the reason you're doing some your process A is probably to feed process B. So, how many process B's have you got in your business that could benefit from implementing process A?
Armando: What do you find as a difference between a senior executive at a larger corporation, medium large corporation, versus a founder of a business that still might might be very productive and very well, but maybe they have 10 to 20 employees. When you're when you're coaching them, when you're when you're training them?
Ross: Really interesting. I think there is a huge attachment to legacy. There is a huge attachment to this is the way that we've always done things. And there is a real need, I think, for senior executives to tick the AI box. And often that looks like signing a contract with Anthropic or with Microsoft. There isn't really When I see a founder, no, a founder will want to achieve We talked about outcomes earlier. To achieve the outcome in the most efficient and effective way possible. And that may be AI, that may be automation, that may be what you That may be human.
**[00:41:42]**
But, they're willing to do all They're willing to explore all three of those things or a combination of those three. And the other thing is that I find super interesting, and which is why I've really really enjoyed this chat, is you never know when you're going into a training session or a meeting, what someone says, what someone actually means when they say, "I know a lot about AI. You're never sure what that means because I was in a meeting this week and the guy was like, "Yeah, yeah, yeah. I know all about AI. Don't worry. Don't worry. We do you know, you know what? I set up something that every morning my AI tells me what meetings I've got that day." I was I was like, "I don't you know, in 2026 I don't you know, yes, that's good. You're adopting something, but you know, there is there is so much more that is that is possible than like the things we've talked about today. And I think once as an organization or a business owner you start to not think of isolated use cases like that and you start to think of all these things compounding and helping you run your business end-to-end, that's when you're really leveraging you know, what you can do. Does that make sense?
Armando: Yes, and that's a great point. I've had times when I've been called to do some sort of workshop and you hear "Oh, by the way, this group just took another training, so take that into account." Like they already did the And you go in and you yourself you're always like, "Oh my god, so another all going to be experts. Let me do this expert level." But so I stop myself and I'm like, "Okay, let's let's make dynamic, right?" And so you get there and you ask specific questions to see where people are and you realize they're like still less than 101 because maybe the other training wasn't as effective or it didn't land or whatever or it just covered completely different topics than you expected.
**[00:43:53]**
So this applies to AI, to anything that that you're training. And sometimes we sort of shoot ourselves in the foot because we trust the what we told what they told us before actually going in and prodding. And when you prod, like you said, now it's like, oh, let's let's begin here.
Ross: You're absolutely right. And you I look you again, you're better than me because I've gone into meetings where some These guys are These are And this is no shade on these people, by the way. I mean, the reason that they think they're great is because of, you know, the circles they move in, but things are just moving so quickly. But I have gone into a meeting where someone said, like, these are These are an incredibly advanced team. Can you show them the most advanced things you do with with your AI stack? And I did. And people said, well, I don't understand what's just happened. What have you just done? And I was like, And then you have to go, okay, I see. I see. All right. And, you know, it's it is no It is no shade at all and it because, you know, I spend a lot of time in this. This is my full-time job. And it's just It is so hard because everybody can use AI for free. It is so hard to judge people's level of expertise. For you as a trainer or an enabler or a builder, but also for you as an individual. So, I always think I am not going to be the best person in the room. Because we've discussed things. I've learned stuff from you. Hopefully, you've learned stuff from me. Because that's how you get better. Is by not always thinking you're the uber expert.
Armando: I love it because I sometimes get and very often nowadays very pleasantly surprised with non-technical people using AI and how they use it. So, we'll be chatting about something and I might know all the like really advanced things and then like you said, like about certain things.
**[00:45:58]**
But and you talk to them and they're like, oh, yeah, yeah, I built this application. This is this is this is this is this is this." And you're like, "How did you think of that?" because they're they're so out of the system thinking that there's no barriers to stop them from, " maybe this won't work." And they try some stuff that you're like, "Wow." And you learn from someone who's not technical because their use of AI is so drastically different from that of a software engineer that it's mind-blowing. I mean, I enjoy it.
Ross: I completely agree. And then And I think that is the mistake that a lot of organizations are making. They are handing AI over to the technical people.
Armando: Yes.
Ross: Where you want it to be with the people who want to solve a problem. Who wants to do something. You know, I remember being in a meeting last year where one person who was not technical had to deliver reports to a client and they needed to get the data from an Excel report and put it onto a web app. And there was a We had a member of the technology team, a developer, who was responsible for what connectors were allowed. And the conversation was "I need to get data from this Excel report and send to a client. So, I need I need an Excel connector." And then the tech the tech then I'm paraphrasing. And then the tech guy was, "No, we are not allowed to install an Excel connector onto your AI system. That is not a That's so we are not able to deliver that process." And I was like, "Well, well, hold on. Hold on. You know, what you've actually said is you need to get data from something to put into something else. Is there a Is there a way to get that data into something that isn't Excel?
**[00:48:01]**
For example, CSV. Oh, we could do that. Okay. Is that a problem? No, that isn't a problem. We just can't have an Excel connector. Though there would you do do do the request was meeting on a technical versus non-technical and you know, there needs to be something in between that was like, "We just need to solve the problem, not the tech." Does that make sense?
Armando: That's an excellent way of putting it. I love that example because it is many times the technical team will say, "Well, we will be quick to say, 'No, that can't That can be done. That can work. No, that won't work.' That because of your prejudice from before, past experiences, etc. Or rules and guidelines, etc. But what you just said puts it so clearly that they need something Many people think Excel. They're not talking The client, the customer, or the user is not really thinking Excel. They're thinking a table of data. Yes. They don't care Excel is the only way they know how to view it, but they don't need Excel. They just need the tabular data.
Ross: That That's it. Like, they don't care. They And that's you know, that's almost how to think of the AI. It doesn't care where the data comes from. It doesn't need an Excel connector. It just needs the data.
Armando: That's amazing. What in your experience from from what I gather, you do a lot of training also with with different institutions, right? So, what do you find that there's a benefit or not a benefit of having multiple people in the room from different departments when you're doing these trainings?
Ross: Incredibly I'm incredibly relevant question, a great one. And I think also it talks back to the shapes part. You know, is So, there's two parts to it. It's when you have people from vastly different departments who realizes that the problems that they are trying to solve are 80% alike.
**[00:50:06]**
And, you know, they're just trying to get something from somewhere, do something to it, and get it to somewhere else. And if you get them thinking like that, that they can, you know, stop thinking about the Excel and start thinking about the little thing they want to accomplish. But, the other thing that's really important is allows departments, colleagues, organizations to think of the knock-on effects. So, I had a client who was a gaming company, heavily regulated industry. Their engineering team were really good. I mean, really good. And, you know, they were using you know, an enterprise solution, and they were shipping features a hundred times faster. You know, after a month, get into all of their into all of their platforms. Heavily regulated, though, but the every feature before shipping needed someone from the legal department to look at it.
Armando: Wow.
Ross: Legal department didn't have any AI at all. So, all that was happening is they're amazing, the technology team, the engineering team, they are amazing, but they're just piling to this poor legal person who's going, "Where are all these emails coming from? I just don't have time to do that." So, what we did when we engineering their training is what I would use to do is I want you to educate them so that they can make your work actually see the light of day. And I think that's why having multiple departments working together, you could just be pushing a button like elsewhere and feeling very pleased with yourself.
Armando: Yeah. That's a great way to put it. It's a not only do you learn from what other people are doing and notice patterns, but also you might see that what's actually stopping a big part of your performance is that someone in the loop either doesn't understand something or is doing something in a slow way or they don't have the access to the tools or what you're giving them is a completely different type of input that they need and for you just changing your output to match their input could get drastically affect the chain.
**[00:52:24]**
Ross: Well, we in fact that's we had an exercise where yeah, we had four or five complex joined workflows and workflows four at the end they were like, well, we're just going to have to we can't really define what it is until we find out what we're getting from workflow three and I was like, well, you're the last mile in the chain. You should be telling workflow three what you want, what you need and then they should be telling two who should be telling one. You shouldn't be being you know, beholden. You need to deliver. You're the delivery point. Therefore yeah.
Armando: Nice.
Ross: Like like you said. You they can they can adjust their output if that's what if that's what you need.
Armando: Yeah. This is so much fun. I enjoy talking about all this a lot because it's something so new and then the frontier and I am very interested in seeing all sorts of businesses, but specifically smaller businesses and medium business being able to take advantage of this to close the gap between them and the enterprise and grow themselves, grow the economy, grow the city, grow the etc. And so I find these tools being super useful and getting it not only getting it into their hands, but training them like you're doing at the AI group for them to be enabled and upskilled to do all this is an incredible feat and super helpful for for the economy at large.
Ross: I hope so. I hope so. I think hardly any of this conversation has really been about AI. It's been about how you think about it. Not about unnecessarily how you use it. You know, it is not should I be what which is better chat GPT or Claude. That isn't even a question. If you're asking that, then you know, then have you go and play with them. You'll find out. You'll find out which one suits you best.
Armando: If you were to leave our audience with a remark or suggestion based on your experience and what you see everyday in the trenches, what would that be?
**[00:54:39]**
Ross: I would go to work tomorrow and I would spend some time in my working day to find out apart from my boss, what is really annoying. And find that process that you are doing everyday and it is drudgery and then pointing an AI platform at it and saying, "How might you do this differently? Here are our constraints. Here's what we need to get out." Claude, have a look at this. And I'd be doing that all the time. And once you get into that sort of mindset, you start noticing things that you have allowed to be unsatisfactory and you can fix them. And you know, there's a lot less friction in your life then, which I think is a it's a good place to be.
Armando: That's excellent advice. I've found that people, once they do that first one Yeah. It's I don't know. I don't know what's triggered in our bodies, but it's like, "Oh, it's sort of release."
Ross: It's like, "Oh, if it can do that, can it do that?" And then that's it. That's it. Yeah.
Armando: Amazing. Ross, thank you so much for coming on the show today. It has been a brilliant conversation. I had tons of fun. We should do it again and go deeper into some of these topics. Thank you so much. What is the best place for people to find you, find more about the Galahad Group?
Ross: Galahadgroup.co.uk is our website. It's probably been co.uk. We are global. We work work with clients across the globe. I'm at ross@galahadgroup.co.uk and you can find me on LinkedIn with a photo where I look a lot younger than I do today.
Armando: Amazing. Well, for everyone who's listening, thank you for listening to the show today. And please, if you haven't, please like and subscribe because that way you will be the first to know when an excellent episode like this one with Ross is released.
**[00:56:46]**
Thanks everyone, and we'll see you in the next one. Thanks, Ross. Thanks, Armando.