When My Senior Developer Called My Junior Developer an Idiot (They're Both AI)

This is The Future of Work. Welcome along to another podcast from Ian and the crew on The Future of Work. Let's join them now for the new episode. Well, good morning, good afternoon, good evening, good grief, it's another podcast from, um, The Future of Work Mastery. And, uh, today a new crew member joining us for a chat today is a good friend of mine, Sean. Sean Phillips. Sean, do you want to say hello? Yeah, hello everyone, and thanks Ian for having me here. It's a pleasure. Sean and I work together, and actually we've been talking through this week an interesting topic to do with our use of these vibe coding things and how we could use them. And I've been doing lots of vibe coding myself. I am a 61-year-old app developer who actually today sold my first app. I now have a recurring payment customer for something I've written, which is, you know, it's like when I first sold my first book, my first ever book, you know, literally I sold it to a close friend. I said, it's for sale on Amazon. That person bought it. I thought, that's it, I can now put writer on my passport. So now I can put app developer and I guess I'm the product owner of an app. So anyway, so that has made me realize lots of things about this new AI particularly the one. So what's around? The big one, there are loads, but some of them, Lovable is there, um, Antigravity is there, uh, Google AI Studio, Cursor, uh, Claude Code, Groq's got one as well. Do you know any others that you'd name right now? Uh, yeah, Devin is definitely an agentic AI that I've been using recently. So what we want to talk about is, it's a bit of a listicle. We're going to go through 5 or 6 things that we've discovered together over the last sort of couple of months of doing this. And we hope you enjoy the episode. So the first one is something you said to me a while ago, Sean, when I was sort of showing you some stuff and you went, yeah, now anyone can be a developer. Yeah. I've always believed that what makes a great developer is the ability to problem solve, understand your requirements, and work out how to solve a user's problem. It's not the code. Yeah. The code is always just a tool. But for a long time, the code has been the barrier to entry. And now with Agentic AIs over the last few years, we've really seen that barrier of entry being removed. And I think it's great that you've been vibecoding over the last few months, because you understand how to take requirements in, and then use the AI to produce something that's valuable to users. And it's been proper that, because not only have I gone requirements, but I got to this point when I go, okay, I've got all these things I could do, all in a GitHub with, you know, how it works, and these are all my requirements, and, but what should I do? And interestingly, if I connect into GitHub correctly, I can even get the Agentic AI teams to sort of write down my requirements from me talking, prioritize them, think about what should be next. They can do all of that as well. So it's a good system. So my system, the system I'm referring to, is always going to be Google AI Studio. Not because it's the best, it's just, uh, for listeners of the podcast, Ricardo and I were talking in November and chatting through some options. I was wondering about buying some software. That could do something I wanted to do. And he said, why don't you just, why don't you just write it yourself? And I've not looked back, to be blunt. So, so anyone can be a software developer, which brings out a really interesting thought after we spoke last time, which is Jeff Bezos of Amazon famously said, Amazon isn't one store for a million people, it's a million stores for one person. Each one is quite unique. And I think we're going to get to the point where we get Yes, yes, that app does do that, but my app does it and I've written it my way and it does it quite how I like it. And therefore the cognitive dissonance that you usually have with any app that you use, you can lose because it can do it exactly how you want it done. So certainly in my scheduling and just not an advert for this, by the way, but I have a personal AI assistant, Pia, and that certainly is looking at cognitive dissonance and how to do it in a way that is impossible to not get. So, that's quite good. Any thoughts on that first one, anyone can be a software developer? Yeah. And I think, if I think about my teams at work, we're very much seeing a blurring of the lines between— so, your stereotypical team will have product people, it will have technical people. We're seeing a blurring of lines into technical product owners. Where actually your product people or your non-technical team members are now able to go and just prototype something themselves, go and produce working code, in some cases compliant code, that can then be shipped out to a customer. And they don't need to involve a technical team in that process. And equally, on your technical team side or your engineers, they're now able to rather than drilling down into the details all the time, they can leave that to the AI, they can raise their heads up, and they can actually get on with solving the problems that the users are facing and pick up some of the products. So we're seeing a lot less working in silos and much more collaboration where everyone can do everything. So that's really interesting because it's interesting how we can get together and do everything, but we're going to talk a little bit more about other things we've learned from this. But so 2 years ago, AI could start writing books and every writer said, it'll never replace us. We're humans. It will never replace us. It'll never have the creative spirit. It'll never do those things. You know, get Perplexity or one of the others to test whether you're getting AI. And frankly, when it first came out, Open Playground, 5 years ago, yes, it was pretty much rubbish. I challenge anyone seriously now to think the AI systems can't write well and be creative. You've still got to guide them. But, you know, I, I think of when I'm writing, I now think of myself as the editor. I'm commissioning work, I'm reading what it writes, I'm saying that's not right, I don't want it that way. My big debate at the moment about a fiction piece I'm doing is my AI system is telling me that it's a gang of people together, and therefore they have to solve the mystery in the end as a gang for it to be a gang book, you know, it's a book of a crew that do it together. Right now, there's a big surprising ending where one of the characters does something quite overwhelmingly emotional to capture the baddie. So, so it's doing that. Anyway, my point here is, 2 years ago, writers were saying, we'll never be replaced. Are we 2 years away from coders being replaced completely? Yeah, I don't think, I think replace is perhaps the wrong word. I think it's a transformation. Of the role that the coder is doing. If we look back a little bit further back, this really happened with DevOps. So traditionally you would have developers and you would have operations people and they wouldn't overlap, and now they do. You have DevOps. And I think what we're seeing is that people, your engineers, are now using the AI tools to blend their job role and transform how they're doing it. So you're getting one engineer who is actually coaching some of the AI agents, if you want to use the word coaching. And I think this is the topic we got onto, that actually when you get into agentic AI, there's a lot of similarity between how agentic AI works and human structures. And Ian, I had a look at some of your vibe coding you showed me the other day,, and we got onto a really interesting topic, something very interesting that you did. Now, when I set up an AI to help me with coding, I typically give it a prompt that says, I want you to be a very experienced full-stack engineer with 20 years' experience, and I want it to know everything. But you didn't do that. No, no, maybe I didn't deliberately do that. So what I actually did was I just left the code agent to code and didn't give it any prompts on what kind of coding it was. I just said, let's start with— I mean, classic thing, my friend Riccardo said, start with a hello world and then add to it, it tells a joke. And I did. And actually there are some pitfalls in that, which I now know, but it just started that way. But what I developed was, when I got to the point when it was doing full stack, including Firebase backends for database and MongoDB and frontends, they're all in TypeScript, React, Tailwind for how the styling works. And occasionally it would get itself into a loop where it kept guessing how to solve a problem and it kept not doing it. So I actually opened Gemini and asked it, what is this happening common and how to do it? And it was Gemini that said to me, You should assume the code assistant is a very junior, fresh out of college developer who knows how to code, but doesn't know much else and won't tell you off ever. Will always say yes to you. Think of it like a keen graduate who will do 22 hours a day, but doesn't really know much, but will work really, really hard. And what you need is to create, well, it suggested a gem, which is what you would do with Gemini. A gem, and in that gem, that gem is a full-stack 20-year experienced developer. And so I sort of, I sort of, it suggested, Gemini suggested a prompt, which was, you are a full-stack developer, then it said describe the fault, and then give it a picture of your codebase and let it tell you what to do. And so I did this very thing. It was, it wasn't loading correctly. I got a blank screen. The error kept saying, you know, oh, it's a problem with my import library. Oh, I need to change this. Oh, it's a problem here. So in the end, I stopped it and I said to what would be my full stack senior developer, I've got this problem. Here's my code base. What do you want to see? And it came back with, show me your index.html file. So I showed it that and he went, okay, show me your app.tsx as it's TypeScript file. And he went, ah, there's the smoking gun. Your App TSS is importing the wrong library. And the right library version is this. And he said, would you like me to give you the code? And what I actually said was, give me the script to give the junior developer to do that. He said, fine, here's the script, gave it to the junior developer, worked. So what I worked out was, and isn't this interesting that we're swarming and humans are blurring their roles a little bit. Actually, giving AI like a team member idea. You're my full-stack developer. Ian is the PO. Ian is also the architect because of my background. I can do the architect. That's my background. Meanwhile, there's a junior code developer here. Meanwhile, if I extract the code and give it to Claude Desktop, Claude Desktop can be my QA check. I can also use a different system to do the full sort of what we might call cucumber testing of it. So what I worked out was, first of all, actually using— this was Google— as a junior developer is really, I think, quite a good idea. But having other team members. Yeah, I think it's a fascinating way to use it. It's something that had never occurred to me really. But when you think about agentic AI and what a lot of people are doing, they've got all these agents, but how do those agents actually interact with each other? And we're trying to solve that problem, I think, as an industry at the moment. And find the best way of working. And a lot of the time it ends up being set up that agentic system to work how humans would work. I think it's brilliant to have a junior developer and a senior developer. That exactly mirrors how humans work. And then you can sort of dig deeper and say, well, why does that work? And I guess that's really because the job functions that humans have landed in, we know how to interact with each other in a certain language., and a certain style. And that style, um, can be transposed onto how AI agents interact as well. Uh, here's the key. I think we thought my team of 6 can now be just one AI system that can be as good as all of those. Yeah. But actually there's a tension that exists between different skills that you need to continue to exhibit. So actually, um, Ricardo, for those in the audience, he and I are going to do some podcasts on what he calls mythological-driven development, which is he got two AI systems with an NCP server to start working together with him as God. And it was very interesting how they behaved. I'll tell you, tell you more about that another time. So I think what this has brought out is our initial idea that we could get the AI to do all of the team roles simultaneously, perfectly. I'm not there now. Yeah. And if you think about it from a human point of view, if you give a human and put— take a human, put them in charge of the engineering, the testing, the design, the selling of the product and the commercial side of things, and the compliance and the security, that human's probably going to struggle. And I think we're seeing very much the same thing with AI systems. Because it's not about the capacity of the AI, it's the focus, I think. So for example, I've got this app that's just started selling. I said to Riccardo, what should I do now? He said, you need to create a gem that is your marketing team and ask your gem how to market this app. And I went, great. And he went, don't ask the coder for goodness sake. Don't ask your senior developer. They don't know what they're doing. Get a marketing team, create a gem. And of course, when we say create a gem, there's different ways of doing that. What the gem is doing is really tightly focusing on what it knows. That is, the word is focus. So if I create a gem to do marketing, I'm going to give it all the best marketing books in the world to read and understand, and not the rubbish you might read on LinkedIn. Don't read the last 15 LinkedIn posts on marketing because what's in LinkedIn is really poor. It's, it's aggressive stop-the-scroll statements from people who really don't know what they're talking about. In order to get people to listen to someone who does not know what they're talking about. So, another area, I think. So, it's this handoff, isn't it? Yeah, and I think when we are working in teams, we learn to speak a certain language, and this is something we've talked about in the past. So, we'll learn how to speak in a corporate style, if you're presenting to an executive. You'll learn how to speak in a technical style, if you're trying to motivate a team of engineers. And we have different ways of speaking. And I think as humans, it takes us a while to learn that. And sometimes we spend several years trying to perfect that skill. And the brilliant thing about AI and LLMs is they can learn a style pretty much immediately. It's the very thing that they've come from. They've come from parrot copying. And that's where they came from. It's a large language model. It's designed to copy language. My son calls it the ultimate autocomplete. That's what it does for a living. It autocompletes sentences for you, just in a very big way. It autocodes for you in a very big way. But that's all it's really doing. Yeah. And if you break it down, the reason that these work systems that we put in place work so well is because of the way that we communicate. And that is something that it seems that LLMs are actually able to mimic. So you can take the human system, give it to an agentic system, and we actually see it working okay. And then to extend that idea a little bit further, you can actually have a hybrid system between humans and agentic AIs. And actually, they are capable of handing off. The input that you might get into a testing process, for example, is going to be the same regardless whether you have a human doing that testing or an AI doing that testing. The output that they give out of that system and give on to the compliance approval person, again, a human who's doing that role and an agent who's doing that role, they're going to be looking for the same thing. So there's quite a nice mimicry there. And some organizations at the moment are going as far to actually produce org charts that have both humans and AIs on. To demonstrate where all these roles and functions are. Yeah, yeah, yeah. Very good. Very good. Of course, sometimes the handoff— I want to talk about the handoff for a few minutes. One of my friends has discovered this, which is we all know if you're interested in this area at all, it's all about context. It's all about number of tokens before you have to reset it, all those kind of things. It's exactly the same for the coding agents. They have context problems. They forget what they're doing. They run out. They start to trimmed down because they're running out of context. Yeah, context window. Context windows. So actually, one of the things I've learned is one of the things I do with all my, um, Agentic stuff is every now and again I will stop, not just reset the context, not the chat, but before I do that, I go, I want you to update the big README context MD file that we have written. And by we have written, I mean I've asked it to write. So it writes an entire context to give to somebody else to run, to read. Each of my files has a README of that actual file with what it does for a living, how it works, how to test it. And again, they're all written by the original code developer, but then the sort of senior developer reviews all that and says, that's not, let me change that. I can get other people, other systems to look at it. So one of the things we've learned we have to do is, if I could put it like this, I need to get ChatGPT today to write the handoff that ChatGPT tomorrow will pick up. So actually I'm realizing more and more, and you've said it, actually creating these handoffs, even if it's the same AI system later, you know, really, really important. Yeah. And if we dig a a little bit deeper into the technical reason why agentic systems work is splitting up. It probably is that context window that you mentioned. So not overfilling an AI's— let's call it brain, seeing as we're going down that parallel between humans and AIs— not overfilling an AI's brain. And it's become a little bit of a joke in the tech industry at the moment that when I go to the shop and I forget to buy my carrots, that's because my context window is full. And the carrots have fell out. So there's almost, there's another parallel between a human memory and an AI memory. Which is we're now starting to use the phrase, oh sorry, my context was full. I couldn't put that in. In fact, I was giving some coaching a while ago to someone about someone else and they said, how can I handle this person? When they're in the room, they talk for 30 minutes. I can't get a word in. And I went, think of it like this, their buffer is full. And until they offload 30 minutes worth of buffer out of their system, you can't put any other in. Yeah. You've gotta let them get it out before they'll listen and put it in. And that's a human. I think it's some of that as well. So this model of the idea of context window and the arms race that's going on now between all the agentic AIs to have more context. Yeah. You know, so Gemini and Claude can cope with an 80,000 fiction book in one go and give you the kind of responses of, is it duplicative? Do I say the same thing twice? Have I got a written style that changes? Now that's, that's getting quite powerful. I mean, a year ago it was, here's a chapter and I can't do more than that. Yeah, uh, a year ago it was, here is a transcript of a podcast and Sometimes it would go, "Sorry, I can't read the whole transcript, too many words." It's almost like a too many words problem, isn't it? Yeah. Now, if we were to take this further, we know with human teams that sometimes they don't work, and sometimes humans will fall out with each other, or they won't understand each other's viewpoints. And a lot of what we've said so far is that AIs are going to mimic human behavior. So there's an argument to say that actually AIs are going to fall out with each other if they're expecting certain handoffs. They might not get the handoff in the form they're expecting. They might get frustrated at the other agent. Our senior developer might get fed up with the junior developer because he's not learning fast enough. Oh, he does. I've said he. We'll talk about that in a minute, what language we use. But my senior developer is occasionally quite abusive about the junior developer. Yeah. Like literally, what an idiot. How could he have made this mistake? Yeah. And you know, I've even at one point said, can you tone the critical language in the script down please? Because I'm going to give it the junior developer. And you know, you start with, you know, you are an idiot for not having seen this, but this is, you know. Yeah. So you're actually providing a little bit of coaching to the AI on how to communicate better. With the other AI. And my friend Ricardo has got Each of these things has an MCP. If people don't know what an MCP is, it's like an API interface that can expose an area of what AI can go and look at. So you would have an MCP if you got like a database of the last 10 years of your notes, you might create an MCP on that. That means all the AIs in the world with permission could go and look at that information and, and read it out. So, you know, my friend Ricardo has, uh, 2 AIs talking to each other, both with MCPs, looking at each other's chats all the time. And this is where his mythological mission development idea is coming from, which is one of them tells the other that they'd better improve before God gets upset. Very interesting. Yeah, and if we think about how humans solve some of these problems, so one working methodology that I'm sure you have heard of is Scrum., and in Scrum we do retrospectives so that humans have that space to share with each other. And there's, it's almost a comedy for, or a satire here, but I think it is a serious suggestion that if that works for humans, and we said that AIs mimic the human roles and the human functions and the way the humans hand off to each other, do AIs actually need to work in a Scrum process and have retrospectives between themselves? This is a great thought, Shaun, because, you know, one of the things I will say from my Agile sort of coaching life is if there's something in Agile that's worth everybody doing all the time, it's retrospectives. Yeah. That's the thing. If there's one thing, retrospectives, the idea of improving all the time, build that in. Other things there as well that are good, but that one. So I've not done this, but I'm now going to go to my iGeniq AI system and say, please review the last few weeks and tell me what went well. What do we need to stop doing and how can I improve? Yeah. And then how can you improve as well? I'm looking forward to that. I imagine it's going to tell me off for being imprecise sometimes. Yeah. Well, what you can do there is just ask it to be constructive. I'm coming to you for constructive feedback and frame it in that way. And if there's a particular AI assistant that you have used consistently over a few months, you've got a lot of chats. You can, um, yeah, you can make sure that you ask it, use all of these previous chats about me, put them all together, uh, and then obviously, just as with humans, you're asking for feedback. If you don't like that feedback, you just ignore it. So it's worth giving it a go and see what that AI thinks of you. Um, because this— the fact that AI can mimic well is creating this entire problem we're seeing in the industry now where you can get AI girlfriends who, um, you know, I saw a great— I listened to a great podcast on this that, uh, uh, told a story of a person who got an AI girlfriend. And at one point he said— he types, it's not a live chat interface, he types, you're not a real girlfriend, you know, you're, you're AI. And the AI replies with saying that to me makes me feel unseen, which is a fantastic, fantastic sort of mimic kind of idea of how it works. And of course, um, in that area, it's a bit of a detour, but in that area, we pile on emotions to celebrities all the time. We say Johnny Depp is this great guy. We don't really know what Johnny Depp's like. George Clooney, literally, you know, oh no, new film from George Clooney. My wife goes, oh, we watch this. She settles down, she gets a cake out, she loves the guy. There's an emotional connection there. So why wouldn't people connect with an AI system that's not human? Because my wife's understanding of George Clooney, it's not George Clooney. Yeah, it's just, it's just publicity and status and how he comes across when he wants to, you know. He still wears underwear, goes to the toilet, all that. And the great thing about— sorry, George. And the great thing about LLMs is you can just, if you don't like quite how it's acting, you can just tell it to adapt and just give it a slightly new prompt, which you would struggle to do with George Clooney. Yeah, I suppose. Very good. But yeah, this brings me on to sort of an anecdote, something I noticed because we're in Barcelona at the moment and in Spanish, Well, a lot of us are referring to LLMs as "it." So we'll say, "Oh, it's going to code that for me." And I noticed that some of my Spanish friends actually refer to the LLM as "he." And this is ultimately a translation item because in Spanish you don't really have the word for "it." You use the word "he" instead. And there's a little bit of historic sexism there in the language where he becomes the default rather than she, the feminine form. But when it translates over to English, what that means is you actually get a lot of people referring to it as he, and the LLM really takes on a person. Personality. Yeah, a personality. And I think there's an interesting sort of behavior that humans as a society are going through, not just us as an industry, as to do we refer to them as it or are they he, she, they? And that really adds to the personality element that leads to things like AI girlfriends. Well, yeah, I mean, you know, classically ships are called she. She sails, she sits well on the sea, she's going down. They're machines. Again, it's probably part of the male dominance as well. Let's not be silly about this. We do understand that. But I've also noticed when I use any of the live chat systems and I ask a live chat question, so I'm in the car, so I can't type. And I put a live chat on, if the live chat is perky female, my wife gets viscerally annoyed very quickly with perky female. Literally sort of signals for me to turn it off while she won't say anything, but signals cutthroat kind of thing. If I put a male, a calm male on, she's far happier. And I have no idea why, why that is the case, but it's very interesting that it's an amorphic kind of perception, isn't it? And we're seeing companies take a bit of a blend. So I previously worked supporting contact centers, and we find that different companies take different approaches as to whether they will disclose that you're speaking to an AI. Some cultures prefer to know upfront that they are speaking to an AI and not a human. Others actually don't prefer to know that, and they just interact. And to them, it doesn't make a difference whether there's a human or an AI, as long as the response they're getting is useful. I think there's a lot of companies at the moment trying to work out which is best or how to blend that to best integrate into society. Okay, we're gonna, uh, it's a bit of a listicle, so we're going through our things. Um, the 6th one you wrote on the board is, "I'm wrong, tell me why." Tell me why you came up with "I want to talk about it." Yes, I think this is the argument, if you go back to AI girlfriends, actually, this is a great example. Um, AIs tend to support your view and you can end up in a little bit of an echo chamber by default. And that's the way that the prompts that the AI takes in, it won't naturally challenge those. If I go back to thinking about humans, what I want to surround myself with is people who are going to tell me that I'm wrong, especially at work. I want to know better ways to solve my problem. I don't just want people who are going to tell me, yes, Shaun, that's perfect. To get that prompt out of an AI, it's really, really simple. And I encourage you to go away and do this. You just say, "I'm wrong. Tell me why. Prove me wrong." And there are prompts out there that are much longer and more comprehensive and robust that you can use, but you can really go a long way with just finding— Just that one? Yeah, just, "I'm wrong. Tell me why." Interesting. And start a debate with it. And it can be about something technical, it can be about a viewpoint that you have, but it's a very useful thought process to actually go and develop that, just in the same way that you'd use that with a human. Yes, yes, there's a film I can think of where one of the characters just stands up in the office and puts a, like a megaphone on and goes, tell me something I don't know to people. And It's seen as a bit arrogant, but near the end of the film, the payoff is it's to learn, it's to understand things. And she says, you know, foolish you that you think you know everything. Yeah. I don't do that. Yeah. And if you're having a discussion with this AI and you disagree with it so much that you don't want to continue that discussion, well, it's very easy to get out of because you click start new chat. Are you not doing it consciously? Create a new context window. Well. It sounds like science fiction now, but I'm sure a lot of what we've discussed so far today, 5 years ago would have felt like science fiction. So who knows where we're going to go. Okay, so we're at our sort of time limit. The ideal person who listens to this is probably walking a dog or having a shower or walking to work or having a cup of tea. So usually there's 2 questions we ask anyone on the Crew. One is, what's the most interesting thought you had today? And then what's the question you want to leave with? If I'm thinking of the things you've said, I think the idea of a retrospective with my AI is really interesting. I'm going to do that and I may even report back to people how that went. That's mine. I think that'd be great fun. How about you? What did you notice? Yeah, I think you've stolen mine as well, actually. Oh, yeah? Yes, but I think it's a very exciting thought process. It's still a very new technology that we're learning how to understand. And to be honest, I think having a mix of junior and senior developers in the team, if I'm building a team of humans, that's what I will go for, a nice mix of experience levels. And I think it's a brilliant idea to do that in AI as well, and not just, not just take your expert with 20 years of learning. Because we know it's not just how humans work, but we know how to How to get systemic change is you need different viewpoints and you need different understandings and you need different, shall we say, forces to work not together but in parallel. I.e., they slightly nudge in different directions to get an overall effect. Yeah, all right. What's the challenge question we leave them with then? Yeah, I think it's quite a simple one. Are you using Agentica AI yet? Simple as that. There's a lot going on in this field. It's very, very easy to get involved. And to bring it back around to where we started, anyone can be an AI developer now. The barrier to entry has never been lower. So just give it a go and fail fast. It's very, very easy to give it a go, realize it doesn't work, create a new context window. The AI won't remember anything. And just start again. And yeah, it's never been easier to get involved and get technical and give these AIs a go. And definitely the, yeah, the have a go. And you know, my view is a friend of mine said, just go to Google AI Studio and give the coding assistant a prompt, start, just start, have a go. And then very quickly, this idea of, I asked Gemini what was going on with the coding assistant. It's an interesting final thought from me. The coding assistant was doing something interesting, which it was trying to write the code in the chat rather than on the file. And it kept doing this. I kept trying to stop it. And in the end, I thought, I'm going to ask Gemini. And I asked Gemini and the Gemini system went, yeah, it's doing this at the moment. We're calling it code hallucinating. We don't know why it's doing it, but here's a prompt that will definitely get it sorted for you. And I put the prompt in and that's what it did. It's really interesting that an alternative view is so necessary. Yeah, and I think this is great because AIs and LLMs get a lot of publicity about making mistakes and they hallucinate and they get stuff wrong. But do humans do that? Yes, we do. Of course we do. Of course we do. Sean, thank you very much for joining me. To the podcast audience, if you want to get hold of Sean because you thought something he said was interesting, just get hold of me via The Voiceover Man is about to start and I'll pass them on to Sean. Sean, thanks. Perfect. Yeah, thank you very much for having me. It's been a very fun conversation. That's it for this week. Join us next time for more insight on the future of work. There is also a Linktree site which is at linktree.com/ianbanner. If you like this podcast, please like, subscribe, follow, and tell your friends. Send them a link. It's for free.