#596 AI Agents Need Managers. Not Prompts | Ross Barnes

[00:00:00] 

Mehmet: Hello, and welcome back to a new episode of The CTO Show With Mehmet. Today, I'm very pleased, joining me, Ross Barnes. He's the founder of the Galahad Group. Um, today we're gonna talk about AI a lot, and we're gonna talk about, like, different use case of AI, mainly in, in the ad tech and the agentic AI and maybe the mix of both.

Like as each episode, what I love to do is I keep it to my guest to introduce themselves, tell us a little more about themselves. So Ross, tell me a little bit about you, your background, your journey, and what brought you up to having, you know, Galahad Group, and then we can start the discussion from there.

Ross: Yeah. Yeah. Um, and thank you, uh, Mehmet. It's, uh, an absolute honor to be here, so thank you for having me. Um, my journey is largely, um, a corporate one. I started in digital media and marketing in 2001, uh, publishing my first [00:01:00] book on SEO that year. I then moved agency side, so m- media agency side, um, where I was global CTO of a, um, global, um, agency, um, within WPP.

I left that agency two and a half years ago to start my own business around AI, and now I run the Galahad Group, and the Galahad Group has two arms really. One is about AI enablement and adoption for, for, um, for clients, and the other is we build and sell our own, our own AI platforms, um, which is, um, a lot of fun.

And as you say, everybody's talking about AI, right? 

Mehmet: Yes. Everyone is talking about AI indeed. So Ross, you've seen an, an evolution from all the way from DMPs, data management, like to, to agentic AI systems. What fundamentally changed in, in how value is [00:02:00] created for brands, and what have stayed the same in your opinion?

Ross: Yeah. That's, I mean, that's a great question. I've seen so many acronyms, um, that's the thing. Um, what I think has changed in the last 24 months that is fundamentally different from any of the other digital transformation moments we've witnessed in, in media and marketing, you mentioned DMPs, the launch of Google Search, all those ki- types of things, is that AI truly changes every domain.

Um, not just in marketing, but in business, commercially, and also humanity, um, and changes how we interact with technology because it is everywhere. It is omnipresent. And I think one other thing that's really important about, um, AI is that the barrier to entry is pretty much zero, meaning that- The [00:03:00] spectrum of understanding, usage, and expertise is incredibly wide and is getting wider because anybody can do it and anybody can become an expert.

Mehmet: Right. Now, talking about this, I've seen that you say you are building the cognitive infrastructure, right? So m- when we talk usually today about AI, Ross, so the first thing that comes to mind, of course other than the chatbot tools, is like the other, uh, tools which are like kind of specialized tool or vertical tools.

What's the difference in practice when you talk about the cognitive infrastructure versus like me going and searching for a certain tool? And where do you see people getting this wrong the most? 

Ross: It's a, it's a great question, and I guess it really cuts to the heart of myself as a technologist. I don't think about technology first [00:04:00] because I don't think that's the way that you generate adoption.

I think about people first. And what I think AI has the ability to do is provide support for us to be more human and ensure that we are doing the things that as humans we should be doing, and that is what the cognitive scaffolding framework is, and that comes from a very distinct and personal place. In 2023, I found out I had ADHD and autism, and that was quite a journey finding that out at 40 something.

Um, but what that led me to do is to really examine how my brain worked, and I realized that I'd built coping structures and coping mechanisms that could be replicated through artificial intelligence. And I started to build myself, and I've got the Ross operating system that is a number of 30, 40 [00:05:00] agents that support me in my day-to-day life and make my life better and allow me to avoid the things I don't like but still accomplish them, but focus on the things that really energize me.

And I'm bringing that to corporate businesses so that their staff are happier, more effective, and more efficient, and the company will perform better. Does that make sense? 

Mehmet: So if- I want to say it in, in different terms. So the, the key here is to understand ourselves first, right? So, um, so, so because just throwing...

A-and I think it's on not only AI, Ross, like this, we s- we saw it again and again repeating, um, when new technologies comes in, uh, if, if we don't understand our real problems and w- you know, uh, what, what are like our challenges, i- if you go through any technology, this will make it like more complex and cumbersome, I would say.

So what you said [00:06:00] 100% makes sense. I, I think you need to add something. 

Ross: You're absolutely right. Technology without a purpose for a human is, it's pure theater. And I think, you know, that is, is what AI has, has been. That is the people on stages are... It's innovation theater. Really, this has to start human first.

Um, and at the Galahad Group, we've developed, um, a diagnostic framework, framework, which we call IKIGAI or IKA-GAI AI. And IKIGAI is a, if you don't know, it's a, a Japanese process, a, um, a Japanese methodology of how to live the perfect and happiness, happiest life. Good. You're nodding. Brilliant. That means you're, you're striving towards it, as we all should be.

But what this does, this diagnostic framework, is it looks at the things that you do in a day-to-day basis and distills the four H's of humanity, that we call them, which are hear, hunch, hunger, and heart. The things that AI is never [00:07:00] going to be able to do. So hear is presence, you know, touch or, you know, or, or being on stage.

Heart, which is about showing empathy and sympathy and emotional intelligence. Um, hunger, which is, you know, the drive, the ambition, the ability to overcome failure. And the last one is hunch, instinct. AI cannot be instinctual. And what we do is we, we put tasks and jobs and job descriptions into this framework, and we decide whether AI should even be involved at all or what AI can do to support you to achieve those tasks.

And I spend a lot of time with CEOs, and I get them to give me their job description, and we put it in the framework, and we show what their job would look like if they were augmented by AI, and they realize how much time they're wasting and how much they're not doing the things that really, really can impact the business.

Mehmet: Right. The reason why [00:08:00] I smiled and I nodded my head because, you know, this is one of the books that, you know-- I, I rarely read a book twice, which I should do more. But this is one of the books that I read a couple of times, "Ikigai." It's a, it's a book, like, available. Um, a-and, and you know, it, it... I used to hear about it, but when I read it and I read the stories there, uh, and as you said, like, they, they go to, to these Japanese villages and, you know, they, they, they sit with the, with the local people over there and just to understand, like, how they live long and happy.

And, um, I, I advise everyone to go and read the book. Yeah. I don't want to spoil it, but it, it, it's a good place to start. Uh, I call it, like, to, to, to, to know your purpose, right? Like, wh-why, why you do what you do. And I think if you know what your s- your purpose, it becomes easier to understand in business perspective and technology perspective, like, why I need to get something done or, like, why I should adopt this technology or that technology.

[00:09:00] So 100% on this. Now, we're living again in AI, and AI is changing. As you said, it's an innovation theater. I agree with you on this, Ross. And, you know, we are seeing businesses trying to... B-because I think this time what everyone is trying to do, they said, "Okay, we've seen or we've learned from history, like the ones who missed some technologies, they stop to exist," right?

So this time people are keen actually to adopt this, and now we're talking about AI agents, agentic AI, whatever you want to call it. So everyone talks about it, and maybe, correct me if I'm wrong, maybe few are deploying them the right way, right? Um, from your perspective, like, if you can give us maybe two or three examples of, of agentic workflows that actually work today, um, in [00:10:00] marketing or anything else.

Ross: Yeah, that's a great question. And I, I almost think it's not about, for me, it's not about applying it in the right way. It's about applying it. To your point that companies that don't exapt-- adapt to new technologies, you know, you think of your Kodaks and the like, you know, can get left behind. But as we go back to that barrier to entry point, anybody can be doing this.

Any business can be doing this, and also any individual in your business can be doing this. So there will be, in your business, shadow AI going on. There will be agents running on systems that you don't know about. There will be data leaking out of your organization and training other people's models. Um, but good use cases are ones that make your life easier as an individual, that automate That have good guardrails, that understand the [00:11:00] risks and understand the, um, uh, understand and don't overreach.

So things I use my operating system for are relationship marketing, and not necessarily even customer relationship marketing, but ensuring that I know my relationships with people, the status of those, the, um, the next best, best actions. And I think CRM is a real place f- in marketing where, um, a-agentic workflows are, are, are really powerful.

I also think data analysis, um, performing complex functions and analyzing the output. I've seen and worked with a lot of corporate clients, and also keeping people honest. I think that's really important because what AI can do is it can remove, um... Have you heard of HIPPOs, um, in [00:12:00] meetings? The highest paid person's opinion.

Because it's the truth. It has no ego. It has no hunger- Mm ... hunger, and therefore it can be a great leveler in, in making the, the truth come out.

Mehmet: I'm gonna go to the ego part in a moment. Um- Yeah. But one thing, you know, and it's, it's a top of mind for everyone, uh, whether they are running a small business, whether they are like, um, startup founders, um, or maybe even large scale, uh, corporate leaders. Um, so w- we know that, you know, we can scale the output without increasing headcount.

It's, it's a pro- it's proven, you know? Uh- Yep. 

Ross: Yep ... 

Mehmet: and this, the, the reason I say it's proven because, you know, and, and I shared like many examples on my LinkedIn about things that when I think about how I was doing it and how long it used to take with [00:13:00] me, you know, like years ago, and how much I felt like as if I have with me seven, eight team members although like I'm my- by myself.

So imagine this is... And, and this is just, you know, for like, uh, I would call it, uh, like consult- in the consultancy or advisory- Yep ... space. But, but think about someone who has to, to go to work every day and put all the effort. What are you seeing this? So st- sometime but still I'm seeing, Ross, people who are denying this.

They're saying like it's a hype. That's too much. You can't still replace headcounts with, with the AI. So how we can differentiate this, um, a- and make it clear to people that this is not theory, it's actually a reality? 

Ross: There's some really interesting things to unpack there, Mehmet. Really interesting. And I think one of them being, um, [00:14:00] the point of you said when you now, when you're working with your agents, you think back to what you used to do and how hard that used to be.

And I think that is a very human trait, is to think of the work you did, let's say yesterday, which was twice as hard, versus today. You think-- you see that as a sunk cost, and I think that's why some people kind of don't want to get on the AI boat because they feel that they have been working so hard and running so fast that that is what, you know, that is what fuels them.

And I think that is a, that is a genuine fear for the adoption of, of AI in business, is kind of exposure that you'd been doing it wrong. But you hadn't, you just hadn't had access to the technology to, to do it better. Um, and I think the-- and I think fear is, is a great barrier. [00:15:00] People, people just na-naturally abhor change, don't they?

Um, and this is an institutional change that, uh, you know, it's a societal change, and that's pretty scary. 

Mehmet: I-I-It is, yeah. So it-- I, I don't deny that even sometimes myself when, you know, I reflect on the things that we can achieve now with this technology, uh, it, it makes me think... 'Cause I'm not scared, but I would say, you know...

And back to the ikigai thing, by the way. 

Ross: Yeah. 

Mehmet: Um, like hold on one second. Like, for example, I used to spend hours when I was in my 20s on things that now an AI can, you know, do it in seconds, right? 

Ross: Yeah, yeah, yeah. 

Mehmet: So, uh, like was it really the thing that I wanted to do at that time? Probably not, because now I-- you start to divert your thinking from, you know, doing these [00:16:00] tasks towards something more meaningful and how I can come up with something, you know, that can really help people, help businesses in, in, in thriving and moving forward.

Saying that, we know that as any company, you know, whether it's a startup or, you know, if, if maybe it's a small, um, small project that you're working on, um, maybe you are just an entrepreneur, you're starting now your, your journey. Um, we learned how to build what we call the go-to market, right? So- 

Ross: Yeah, yeah.

Mehmet: Yeah ... a-a-a-and build up on the story. So if I want to reimagine, to rebuild this using AI today, what's my stack would look like and how- You know, I can... H- how I, I changed, you know, what I have learned for years and years, right? Maybe from books, from blogs, articles. [00:17:00] So what are the things that I need to unlearn first in order for me to build the right stack now?

Ross: Yes, great question. Great question. Um, for me, it start with your personal stack first. Build, um, and outsource to AI as much of your cognitive friction, or you mentioned, um, tasks that you did that you didn't want to be doing. Find out what they are, and then build agents that can help you do that. Now, a lot of people ask me, "What is the best model?

What is the best ChatGPT, Claude for me to be doing?" And, and my, my stock answer to that is, if you're asking me that, then it probably doesn't matter for you, because you probably aren't using models for particular use cases. Just find one, adopt one, and start to explore, because [00:18:00] effort in this area compounds.

The more context you give your agents, the more context they will demand, and that is how you start to build multi-agent workflows that can make you more efficient and more effective. But the biggest, the biggest barrier is paralysis. You have just got to do it. I mean, we are at a point now where, and I'm seeing this when I'm talking to some startups, where the...

Because the, the distance between idea and execution is becoming closer and closer to zero, the moment you're not building your idea is the moment someone could be building it instead of you. 

Mehmet: Right. 

Ross: And that... Just get to MVP as quickly as you can, because [00:19:00] then investors, customers can see there is something, and therefore won't just build it themselves.

And I think that is a clear and present danger. 

Mehmet: Right. Do, do you think, like, we're, we're shifting from, um, you know, a campaign driven to system driven, um, you know, um, I would say, uh, from, from, from marketing perspective again. Uh, so because if, if we know that the campaigns, to your point, like we, that we used to do before, it might not work now, so now we have to make the systems right.

So h- now here is the question: How I can structure my team based on, on this new, um, way of doing things, um, and how I enable them? 

Ross: I, I would, I would say we should always have been, for marketing and advertising campaigns, systems driven. Um, always start with that infrastructure [00:20:00] because building as you're flying just isn't, you know...

While it, while it's possible, you need to have as much, um, down, uh, uh, much in place as possible. And now because of hyper-personalization, because of the different discovery surfaces, we have so many different ways to talk to so many different people, meaning that campaigns, while still useful, we are so informed by data that we are audience and systems driven.

Um, and at my old agency, that's exactly how we approach things, and I think that thinking makes the, the modern, let's say, media planner much more an architect, um, in terms of building a way that, a way that a campaign can flow through a life cycle, and to an individual can appear like a campaign, but really it's part of a [00:21:00] system that is being served or s- shown to different people in different ways.

Mehmet: Right. So, so, so you talked about visibility and discovery, right? So I- I've discussed this, like, recently in couple of episodes. Last year I had a great episode also as well, uh, with one of my, Vikas, uh, you know, who, who was trying to build something to solve this issue. So we know now as consumers, uh, a- and I'm think- I'm talking here, you know, how consumers are now discovering brands versus how they used to.

Mm-hmm. So in, you know, and starting again from myself- ... the be- behavior changed from searching on Google to going and asking one of their LLM, whether it's ChatGPT or Claude, right? So- 

Ross: Yeah ... 

Mehmet: um, so now the search recommendation and decision [00:22:00] making, all these are changing. So how companies should be rethinking visibility, uh, in, in this new AI mediated world?

Ross: Oh, Mehmet, it's something I'm so passionate about, this. Um, we've got, we've got a platform, um, called GRAIL, which allows you to build authority in LLMs and allows you to, um, surface your content in those surfaces. But I won't talk about that because I... That's interesting, but what's really interesting is, so I, as I said, I wrote my first book on SEO 2001.

Mm-hmm. Um, and that was around the times of doorway pages, white text, meta, meta tags, and- You know, if you think back to that time, your point about pe- your content was disintermediated by an AltaVista or a HotBot if you were a brand, right? People went there, and then the Google algorithm came in, and then Google Ads came in, [00:23:00] and then app stores came in, um, personal algorithms came in, and there was a disintermediation.

But you could always kind of pay to be there, and you could al- there was always a known way of doing these things, of, of, of optimizing. Um, LLMs have changed that completely. Um, it's such a different surface of discovery in that, you know, you're not just going for what's the best car anymore, because the answer to that has so much context behind it of everything that user has ever searched for before.

So what, what you as a brand need to do is find a way to both talk to the human and also talk to the machine. And I... My hypothesis is that we're going to get to a world where effectively we have a content iceberg. As more and more AI [00:24:00] slop is produced, LLMs are going to naturally have to, to avoid model collapse, have to actively search for human content, actively search for premium human content.

Now, that will probably be video or, you know, it, it will probably be harder content for an LLM to consume, s- so they'll also need context. So you'll be building content systems where your content will be premium, human, beautiful, but you will also be providing a layer of context to an LLM, so you get the authority from the humanity, and you get the, um, intent and the query analysis from that machine structured content.

And hopefully that will mean the death of AI slop. 

Mehmet: Hopefully. Now, it, it, it's good that, you know, 'cause I... You mentioned this because I want to [00:25:00] come back to something, two things that you mentioned previously. Mm-hmm. And I told you I'm gonna go back to the ego thing of AI. Yeah. Um, because I want to talk about safety.

Um, and the reason I, and the reason I stopped when you said the ego, and this is something, maybe it's only me because I interacted with AI a lot, uh, I mean, especially the chatbots like ChatGPT- ... Claude, recently more Claude. And one of the things, of course, you know, their, their whole, their, their, their whole, I would say, um, they don't men- they, they didn't mention that at the beginning That way straightforward.

But now we know it's about the personalization how, and how it can get you the thing that, you know, are important for you, the user. 

Ross: Mm-hmm. Yeah. 

Mehmet: Because, because of course, you know, if you remember ChatGPT added memories and [00:26:00] then of course the rest th- they followed, and then later on they started to have this ability to kind of yourself train the AI on the tone you want.

Like do you want it to be friendly? Do you want it to be this, to be that? Now on the ego part, and this is why I'm asking you, and this is related to safety by the way. 

Ross: Mm-hmm. 

Mehmet: Um, so if I'm, if-- how far can we trust, you know, an action which is actually driven by an LLM telling the agent to do what it has to do?

Because remember, agentic AI is just the layer that al- uh, i- it's just, you know, the, the gears that allow LLM to looks like it's active, right? Because, you know, usually LLMs, people think they are only spitting out words or images or videos, but actually they become the engine or the gears, whatever you want to call them, to have these actions and you have like multiple agents working together.

Now, from... You [00:27:00] s- you talked about the safeguards, and the reason again I'm gonna go back to the ego thing. Um, sometime I feel the, the a- the LLM has an ego actually, right? Um, there's this funny guy I'm not sure if you have seen it or not, um, who challenged Sam Altman, and, uh, I think he, he's-- I'm not sure if his name or his nickname is Husk.

So- Okay ... he, he, he recorded himself, uh, uh, telling, uh, the voice model of ChatGPT, like, "Hey, uh, can you, can you count for me? Uh, I'm gonna go do a run and come back." And then he just like do it for two seconds and then ask the model like how, how long I did, and the model said like, "You did seven minutes." He said, "No," like, "I, I didn't even move.

It was like seconds." And the model was kind of arrogant rather than, you know... Uh, uh, it was, "No, no, I'm sure I [00:28:00] can do this." Even they showed it to, to Sam Altman, and Sam Altman said, "Yes, we are aware about this issue, and we're gonna solve, you know, the awareness of the LLM of time in a future, uh, release." Now, of course, this is like kind of a fun game or fun- Yeah

thing that the guy is doing. But let's think, let's, let's take it out for a moment and think about an action or an output that was given into a, um, you know, corporate or, or company or business perspective Uh, you said no ego, but if the AI thinks or tell us, like, it's doing the right thing, which actually not, so I think it ha- it...

we can say it has an ego here. Now, I'm asking all this about how we safeguard this. H- how, how we make sure that actually we're, we're using the AI in, in the safest way so we don't have... Of course, we... [00:29:00] hallucinations are decreasing with time. We know this, the slope you just mentioned. Yeah, we want to get rid of the AI slope.

Ross: Yeah. 

Mehmet: But how... till we reach this ultimate situation, what we should be doing? 

Ross: It's a really, really interesting question, and lots of interesting words that I think I'd... Trust is a really interesting word that you used. How much do you trust a coworker? 

Mehmet: Yeah. 

Ross: And I think that is the kind of attitude to have.

Would you tell a coworker or a junior in your team or anyone in your team how to do a job without telling them how not to do the job? Without telling them the things they weren't allowed to do, would you assume? Hopefully not. That is w- the same with an AI agent. You mentioned about, [00:30:00] um, ta- tasks. How would you give tasks to a coworker?

You wouldn't give them 100, you'd give them two, because giving them two allows you to control those tasks, be very clear with what they're supposed to do, and be very clear, um, with, with what they are not allowed to do. We've been spoiled by ChatGPT and Clau- and Claude, because we expect agents to know everything.

Because they want- Right ... to please us, they try and prove they can do everything. They can't do everything. So make your agents very discreet, give them very clear boundaries and guardrails, and give them the opportunity to fail and let you know they failed or let you know they're about to fail, just as you would do with a coworker, because being fallible does not mean not being useful.

So in your thing about ChatG- uh, ChatGPT not knowing how to, [00:31:00] how to count time, fine. I won't, I won't use it to do that. You know? If I've got a coworker who's not very good at m- uh, maths, but they're very good at something else, I'll get them to do the something else. We'll find another way to solve the math problem.

And so for my... when I go running, I'll use Strava. I won't use ChatGPT because it can't do that. But we have a tendency to look for fallibility as an excuse to rule out a technology for every other use case, whereas you should just rule out that use case. Fine. Find another way of doing 

Mehmet: it. Don't take shortcuts.

Don't take shortcuts 

Ross: Exactly. Don't take shortcuts. If, if you want it to tell the time, get it to write a Python function that will tell the time. It's, you know, it, we can't expect everyone to be able to do everything the same way you would with a coworker. 

Mehmet: And, you know, back to the point we discussed about, like, the [00:32:00] headcounts and all this.

So now I believe maybe there are one, maybe there are two, some people they can combine them. But in my opinion, like, two skills, traits, whatever you want to call them, for us humans that we should, you know, be mastering in this age of agentic AI. First is what you just mentioned, how I assign the right task to the right skill- Yeah

whether or not it's a human. Ac- actually it, it, it should have been the case for long time, but with AI it's amplified. So now we understand the importance of, you know, choosing the right person or right agent for the right task. The second one is the attention to the details. Because if you assume, you know...

And this is why, for example, what I start to do, and because you asked about do you trust a coworker, and this is what I, I start to do. Like, if I give a task to Claude or ChatGPT, the first time or two times I would go and check the work, [00:33:00] at least if there is some calculations or numbers. And I know for a matter of fact sometime they can do mistakes.

But once, you know, I do it two, three times and see always the output is good, so I know like, okay, they trained it the ri- the right way, no hallucination, it's doing the right thing. So I think the ones who are able, with the amount of this amplified work, to be able to spot these flaws- 

Ross: Yep ... 

Mehmet: and, and just to fix them, by the way, not to cr- to critica- to criticize, sorry, the, the, the, the AI as, as a coworker.

I'm, I'm with that. But, you know, so you become, you become, like, a, a super important in, in, in your, in your business because first you know how to assign the right skill to the right task. Second, you have great attention to detail, which again, will make you, like, super powerful in your, in, in my opinion.

Like, this is my, my, my- Agreed ... my, my, uh, two cents on, on [00:34:00] this topic. 

Ross: And do, do you know, and what you're describing there, Mehmet, is a good manag- manager Someone that knows their team, whether they're agents or whether they're coworkers, can assign the task to the right person, agent, skill set, and then is also responsible for that person in their team...

Person or agent, sorry, in their team's output, and helps them improve that output- Absolutely ... rather than just firing them or shutting the agent down. So those skills exist in businesses. It's just a different surface, and really it is about... And we, we, we do, um, anthropomorphize, anthropomorphize, that's the word I think.

Uh, we make the agents seem human. But there are some things that are very useful to think about how to approach them. But it's good management is crucial. [00:35:00] 

Mehmet: It is crucial, 100%. Now, if we want to compare enterprise versus startups, Ross, where do you think startups have, with AI of course, an unfair advantage, um- Done

in the adoption? And where, in your opinion, still enterprise would win? 

Ross: Yeah. Now, okay, this, this goes back, again, it goes back to our ego point, I think, is that we as humans are incredibly, incredibly passionate about legacy. Um, the reason, one of the reasons that, um, Claude Code is so fascinating for software developers is how software used to be built, or still is built in some organizations, is it's always something on top of something else.

So you are building features on existing, an existing code base most of the time, and that's because it's quite [00:36:00] hard to rip down a, an entire code base. You don't understand the old code base because you're a new developer, or you're attached to it 'cause you built it yourself. Whereas Claude Code doesn't care about that.

It will refactor the entire code base to build a new feature if it needs to, and it will rip down the legacy, and that analogy is analogous to startups versus enterprise. There is no legacy, and in that gap that we've talked about from idea to execution, legacy gets in the way. Startups have speed and the ability to scale so much quicker than they did.

Now, where enterprise still wins is in trust and brand and safety. They're having, you know, the ability to, uh, you- Li- assume liability for [00:37:00] anything for other companies. You know, there are so, there are so many stories of datas leaking out of startups about AI, um, chatbots sending, uh, uh, for example, uh, I read, read a story the other day of someone who was in an, um, who had an AI, um, um, transcription agent that, um, was actually would listen to her on her device.

And she was having a personal call, but in her diary, the, um, agent, the, her listening agent thought she was in a corporate meeting. 

Mehmet: Mm. 

Ross: What, what it, what that agent then did is it took the transcript of her personal call and forwarded that to all the attendees of her corporate meeting that she wasn't in.

Mehmet: Ooh. 

Ross: She'd given it permission to do that, but she didn't know she'd given it permission to do that. So there is [00:38:00] still things that are, you know, the edge cases that startups need to just be really aware of, that enterprise has either probably learnt the hard lessons already or has done the hard yards to make sure those things don't exist.

Mehmet: True. I agree with you on this. And, uh, it, it's, it's, it's pretty much, you know, the... I don't like to call it trial and, trial and error but, you know, it, it's a learning curve I would call it, so, um, but to your point, you know, the, the, the legacy is, uh, is, is always in the way, right? So- 

Ross: Yes ... 

Mehmet: uh, a- and we can understand this.

Like, we can see it also. And w- I think this is why, for me, like, my opinion, like, I, I would say both, they have advantages, right? So I would not say, like, only startup, they will have an unfair advantage and the existing ones. It depends on how you act, on how you adapt, how you, [00:39:00] back to, to your introduction, right?

Like, how, how we, we understand ourselves, right? 

Ross: Yeah. 

Mehmet: Uh, and, um, so both can thrive, and even maybe do collaboration as well. Like, that's what I wish to happen, like, you know, when, where startups can go to big enterprise with legacy and try to help them and, and vice versa also as well. So this w- will be the ultimate world I know, but, uh, why not?

And AI is, is allowing us to do this now. Um- Really ... you know, as we almost close to the end, just, you know, I know, like, you scaled the agency, you know, from £80 million to over a billion pounds. 

Ross: Mm. 

Mehmet: Um, so what lesson from that journey still apply when building AI-native companies? 

Ross: Yeah, great question. Great question.

Um- I think the lessons from growing quickly, um, is an, a true [00:40:00] understanding of your culture and what you want to achieve, and ensuring that everybody understands that culture. Um, our culture was ask, um, forgiveness, not permission. We needed to move quickly, so we moved quickly. And to do that, you have to have faith in people, and you have to have trust in people, back to our, to our agent point of view.

And I think that is, in an AI-driven business, that is as important. But as a leader or a manager of agents or AI, you are responsible for the output of those systems. You are responsible for the behavior of your, um, staff, the performance of your staff, and I think that's really important. Moving forward but maintaining accountability throughout the business and throughout the systems that you deploy is really important.

Mehmet: Great insights, I would say, Ross. Um, [00:41:00] finally, you know, this is a common question I ask to all my guests at the end. Where people can get in touch and learn more? 

Ross: Yeah. Um, our website's at galahadgroup.co.uk. If you go to galahadgroup.co.uk/ikigai, you can do the Ikigai diagnostic, find out, um, which parts of your job could be, um, could be enhanced by agents.

Um, I'm also on LinkedIn, um, as Ross Barnes, and on, um, o-on, on X as Ross Galahad, um, where I post about some of my, some of my thoughts on what's going on in the world of AI. 

Mehmet: Great. Uh, again, Ross, thank you very much for joining me today. It was a very insightful... I enjoyed the discussion myself. Uh, a lot of- Same here

balls, you know, like it, it came on top of my head. Um, a-and, uh, you know, thank you for sharing your time. Uh, and do-- you know, the links you mentioned, they will be available in the show notes, so if, if people are listening on their favorite podcasting app, they can find the links in the [00:42:00] show notes. If you're watching this on YouTube, you'll find them in description.

And, um, this is how I end my episodes. Um, this is for the audience. If you just discovered this podcast by luck, thank you for passing by. I hope you enjoyed. If you did so, give me a small favor. We're trying to do kind of an impact here and sharing knowledge, so subscribe and share it with as many people as you can, friends, colleagues, family members, and so on.

And if, if you are one of the people who keep following us, keep sending me their messages, keep taking us to new, you know, horizons and limits by moving the podcast into the Apple Top 200 podcast chart across multiple countries. Thank you very much, because this cannot happen by itself. As I mention always Since 2025, last year, I'm seeing the podcast never ever it went off it-- one of the country's top 200 podcast charts, and this cannot happen without people recommending and sharing.

So thank you very much. And as I say always, stay tuned for a new episode very soon. Thank [00:43:00] you. Bye-bye.