ChatGPT Agent and the Future of Digital Marketing

Welcome to the digital marketing podcast brought to you by targetinternet.com. My name is Karen Rogers and I'm Daniel Rolls and today Daniel we are discussing AI agents. Okay so let's just define what an AI agent is. Talk about the evolution and then talk about why we're here. This very specific period. What are we here? Why so profound? Yeah. Well it's going to get a bit profound in this one actually. So, definition of a agent based AI is it uses sophisticated reasoning and iterative planning to autonomously solve complex multi-step problems. Okay so this started when Chattatubiti and Google Gemini released deep research modes. I mean it started wait before that but that's what's going to brought it to everyone. So deep research mode if not familiar you go in, you've in Chattatubiti for example you've got your little plus button as you have in Gemini and you could set deep research mode you'd give it a task maybe to do some research and it would go off and ask you a series of questions. I what target market are we talking about was the time period you're interested in etc etc and then it would go off and complete a task for you autonomously. Now the thing is that's not quick. It might take 10, 15, 20, 30 minutes to go off and do that thing but that was their first kind of move into agent based AI. Then what you also had was Chattatubiti were building this thing called operator and operator was the ability of Chattatubiti to use a browser not to search in a browser but to you know log into a website to go through to fill forms in all those kind of things. What's happened and we've just done a new zepisode on this is the Chattatubiti released their agent. So Chattatubiti agent and Chattatubiti agent suddenly takes deep research the ability to use a browser to manipulate documents and gives you this new thing that can go off and do stuff for you. It is a game changer now. I should say I got myself in trouble with this so I got hold of this a bit earlier. But by the time this comes out it would have been out for about 10 days Chattatubiti agent and I got hold of it really early on and I started playing with it and I started creating lots of test cases and I tested the heck out of it and I then went look what about I want you to research content, create content, log in to my profile on LinkedIn and post it. How I bought you to go into all my LinkedIn posts and I want you to analyze all the time we work content I should do in the future. How about I use the Canva connector in Chattatubiti. If you're not familiar there are lots of connectors in Chattatubiti now that can you connect to external tools and I was able to connect it to my Canva account and I could do this to the agent either. I could get it to log in to my Canva account, take the content he had created and create a load of carousel posts for me and then take those carousel posts and put them into LinkedIn. What could possibly go wrong? So it didn't. The thing was nothing went wrong. It did a really good job. But there was a risk right now what I should say with all of this AI agent stuff there are human intervention points. So for example when it goes to my Canva account it doesn't know my password so I have to take over. So I click the button I take over the browser for a moment. I put my password and my username in and it logs in and it carries on my behalf. So let's just take a step back for that. How this LinkedIn post got me into trouble was that I said this is game changing technology. I reckon I can replace a marketing exec with seven decent prompts. What I went on to say was that means you have huge cost savings. These agents don't need holidays. They don't want bonus pay. Their dogs don't get sick and all this kind of stuff. And then I did add which is the people that a bit of no one seemed to read was that you'll also have no talent pipeline because you won't have any junior people anymore. You're going to have no diversity of thinking your organization. You have any young people potentially. You won't have a working culture and that's going to have quite a big impact on businesses. What do you think businesses will do about this? Classic social media. Nobody read the bottom. I was being a little bit facetious. You have them all. You have them all. We don't need. We don't need these people. And my favourite one was that come I said you clearly didn't really test this. And it's like my professional integrity was cooled out to quite a large extent. But anyway, what I normally do when I get angry about social media post is I refuse to respond to it until the next day and I've normally calmed down by the next day. So I went away from the edge. It's very politely pointed out to have on the actually. I wasn't saying it was 100%. But so there's a couple of things to pick out there. First of all, I can now get this to do things I could never do before. Like log in to my LinkedIn and analyse all of my posts. I don't have to export it first. There's phenomenally useful things to do. I can get it to log in and post social media stuff for me. Is that a good idea? Probably not. It's kind of risky. But I can get it to log in and buy something for me. So I don't think we're very far away from the first legal case of saying, "Chachy PT spent all my money and I told it not to." So which is, yeah, there are lots of safety features in place. But my point is this is game changing because this is the first iteration. And we know how quick this stuff is moving. In six months, this will be radically better than it is now because it's learning. And therefore, you will be able to very simply say, "Okay, what do I do every day? What is my marketing executive every day?" Let's turn that into a really good step-by-step prompt. Go to this website, do this thing, download this, go here, do that. And then just take my role and break it down into that series of tasks. And I can automate those. Now, we've had robotic process automation for a while, but it falls over because of the fact that actually, when the website changes, it can no longer handle it. Well, actually now the AI can overcome that. The first time you use, by the way, Chachy PT agent and you sit there watching it browsing your website and telling you what it's doing, or that didn't work, I'm going to try this. Oh no, that didn't work. I'm going to do this. It's quite a mind-blowing experience, I thought. I felt it was like I really, you feel like you experienced something very new. It feels very science fiction the first time you kind of do it as well. Well, we were chatting on text, really, last week. Have you tried it? I'm like, no. Anywhere like, what have you got access to? I don't know. I went in and looked. I've become a bit obsessed with this, but it's the little icon that they shoved in there. Kind of looks like a little robot paw print. So I'm seeing it. Maybe a wonky mushroom. I'm sure that wasn't there intent from a design perspective. No, I know. It's, well, I think while they're trying to, like, illustrate the point that it can open a web browser and click and explore. That's what it looked like to me. But it does look like a bit like a square foot like paw print to me. Like if K9 had paws to me, it's just to screen with that point of clicking on it. That's a paw. It's a paw, you can see it. It's a paw, right? If you agree with me right in, it's a pointer and some little dashes that are showing movement, you think. But anyway, it's kind of exciting. What it's clever about is when they put the word agent on it, it sounds really cool. But is this really that different from Gemini's deep research? It is wildly different because deep research can go off and do things. It can look at things and read things. But this can actually fill forms in navigate through a website. It's not just reading the text of the page. It can interact with the page. So if we go beyond just using a web browser, I mean, that's the main functionality. It can use a web browser, like a human can. But also, if you give it, it used to be, you'd give it a spreadsheet, it would analyze it and then it would give you some sort of output. Well, now you can give it a spreadsheet and it can put the answers in the spreadsheet. So it's got the ability to utilize things that we'll put it in the show notes. So I get into it. I've got a full slash podcast. The functionality it has at the moment. But the point being is that before I could look stuff up, I couldn't actually interactively do things. But now I can go to a logged in website. I can say to it, go and find me a recipe for this many people that have this many dietary requirements and then log in to my Tesco account and buy that stuff for me. And it's able to go through and do that. And if it doesn't got a particular product, it makes more alternative options for me. And there'll be human points. I have to help it log in. I have to say yes to purchase and things like that. But what's really interesting is all the marketing videos they created. They gave the agent a task and then they shut their laptop all the way. And there was a very big push in the market. And then what happens with actually big tears you get a notification on your mobile device when that task is finished. So it's like it's doing stuff for you. It is your agent. It's going off and completing tasks for you. So and you're absolutely right. It is deep research mode combined with the ability to use a browser and some other stuff as well. So it's got more tools. It's disposable. And therefore it's disabled to choose which tools it's using. It's quite fun to watch it while it works. And actually I could see it was at one I don't think I shared this in a previous episode but I bought a laptop using the agent feature with intract GPT. And at she at one point I could see it was writing a whole bunch of code to compare some of the data that it scraped and sourced and do something clever with it. But you have to watch it. It happens quite quickly. Let me give you another example. This is amazing. It basically I asked it. I gave it an image. And I've done this three months ago and it was an image of a with 99 dog faces on one picture. And it said how many dogs in the picture. And three months ago I'm sorry I can't answer. I can't work it out. It's too confusing. This time it went right I'll go through and I will use an algorithm for facial recognition. And then it doesn't work because the facial recognition is set up for human faces. But what I'm going to do is quickly write some code that uses it just to count the number of eyes because I know the dogs are up to two hours. and it counts a number of eyes and it got the answer right. So essentially what we've got is something now that has a number of tools, it can do web scraping, looking stuff off the web, it can go to its large language model, it can write code, it can use a web browser, and it's using those tools to make a decision what's the best way of solving this problem. Now it's not game changer for summer fates when you need to know how many jelly beans are in the kilnager. So, you know, the ability to do things quickly, effectively, in a more creative way is certainly there. What's interesting is that this replaces tasks that we do every day now. It's not that I'm interacting with it backwards and forwards, it's like, go and do this and it'll go off and do it and I'll get back. The outcome is what I need it to be, it's not giving me something I can then do something with, it's actually doing the thing for me in many cases. So my experience of this is yes and it gives some very impressive results, but you sure as hell have to double check and sense check what it's done or you're going to get egg on your face pretty quick. And I only know that because I've produced truly wonderful things using this that I'm very proud of. But I know I'm very fortunate, I work with somebody who really their head is totally in detail and you make any claim and they'll double check it all because that's how they're brain wired and it's brilliant because actually it means you're forced to sense check stuff. I've trodden on that mind too many times and I'm not having the curing. Did you actually check this? Remember when we were first playing around with chat GBT and it was given, we were asking it for latest news in SEO and it's given some sorts of absolute rubbish. This is a few years back, right? But you know, I still think that that element is there, especially when you're playing around with different models you've not used before and you have to sort of sense check it and work out where are the boundaries, what's a good at and what isn't it. So that was going to be my question to you, like if people are wanting to test this out, what sort of things would you recommend people because no, if you haven't used it before you're definitely going to want to play with this, what would you recommend Daniel people do? Well, that's a really good question because it leads us into the prompting piece which is that I think a lot of people use this first of all when I was alright but it's not great because like they did like a one line prompt, go to my LinkedIn, tell me what my best posts are. Okay, so it will go, but are you best posts are the most engagement, are the ones that drove your desired outcomes, are the ones that got the most views, you know, you haven't really defined what best means. But it will be able to log into your LinkedIn and then it will go to each of your posts and it will try and get the stats and then download them for you and it will, so actually that's the kind of thing you can get it to do, that's I would try it out with those kind of things, get it to do something, you have to log in to do something. So to a social media profile, to a shopping website, to, you know, something like go to Amazon to buy you something, whatever it might be. So I haven't gone that far with it, but how do you, how do you do that? Should you give it the log in details within the prompt? No, so you just give it, you say go to this website and do this thing and then when it gets there, it will, it will open up its browser and it will get to the login page and go, okay, well, I think I need you to log in and it will give you a button to take over. Right. And then you take over its browser, you put your login details and then you hand the browser back and at that point, it then does it, but it will remember that's where you go. So now it knows my login filling team and if I ask it to go LinkedIn again, it's able to repeat that step. So depending on which tool you're using, that's kind of terrifying because as far as I'm aware, we don't have two factor authentication on our large language learning model logins. No, so what will happen? That's why you want two factor authentication, switch on everything. Yeah. Because it can switch it on which FTP is stuff like that now as well. Yeah. So you can have two factor because what you want it on FTP and you want it on all of your tools because then if someone, if you accidentally are doing something with your agent or if, you know, somebody's accessing your agent, you want the security of knowing on your mobile device, oh, that's not me. I don't want them to log in. So yeah, security becomes even more important before there are some mess ups with this. But what I'd say is that I would look at what you do every day in your day to day role. And I would think about which of those tasks are repeatable and which of those tasks are purely admin. All right. So I know when we create this podcast, there's a number of steps I go through. I take the podcast. I put it into Descript. I edit it in Descript. That is not something I'm going to get agent to do because there's loads of nuance and subtlety and what's that? Then I'm going to go through and I'm going to create the show notes that go onto the website. Well, actually, I can get AI to do that because I've trained AI to go through, take the transcription, put up the key points, structure it in a certain way, find any links, put those up. So actually, that's a step I go through. Then from that, I've got to log into my website. I've got to put that on tour web page, create the page, fill in all the standard bits. I've got to then go through into Lipson and Publish it and Lipson. There's a lot of stuff in there that I can get the agent to do for me automatically. But it's so business critical. There's some bits I'm not going to do. I'm not going to let it publish it to Lipson without double checking it and all those kind of things as well. It's not 100% at the moment. It will do a task, get a little bit confused sometimes and then just kind of stop. Because it's not, it's got a bit concerned about what's going on or what have you as well. But this will iterate and this will change. The important thing from this is that one, this is going to help you be more effective and efficient. That's great. Hopefully it gives you more time to focus on the creative, the strategic, the critical lateral thinking, whatever it might be. But what I know it's going to mean is that as this gets more advanced, what's going to happen to junior roles? I was being a little bit facetious, right? It will replace a marketing set. If it can replace 80% of what I'm marketing set, that is, do you think people are going to still going to keep recruiting? And everyone's getting, yeah, because you need a talent pipeline. You do. But you will know what companies are like big companies about efficiency. Already, there was a report in the Times when I had this show note, so you can see I'm not making this up. Junior roles across the UK, and this is, there's a global pitch that goes with it as well, 60% job in recruitment for a couple of reasons. One, because there's stuff that's going in the economy. And also, because the fact that, well, actually, I might be able to get an agent, I might be able to get AI to do this. Do I need that role at the moment? Where do I need to focus my efforts? So people are just taking a bit of a step back. It's not necessary replacing them, but it is making people think about recruitment. Take this six minutes down the road, this will radically change. And actually, a lot of the stuff that we've got junior roles doing will be able to do. And be able to do it with these agentec tools really easily. So it's a game changer for the workplace. And it's going to change all of our roles. It is going to replace. I've been preaching for ages. It's not going to replace roles in, you know, it's any certain roles at all places. Mostly it's going to change roles. I think this is going to replace lots of things that people are doing. I just, you've just got me thinking about the security of this though. Right. Like here, we've got a back door, which could any, anybody in your team could be hooking up to all your business critical systems, giving it logins. And actually, the, that's outside of your organization. You don't even know they're using this tech. I bet there's a lot of you out there that are in that instance and you won't even know it like that. And that's a, that's a, because best dream right there. That's actually quite hard for IT teams to even manage. Well, let's talk about something with that then because I had a really interesting conversation with a very senior person in a large retail organization. And it's a great conversation because they said, look, what they're concerned about is at the moment, I want to go and buy something and I'm going to go to the retail website and I'm going to, because I trust them, I like them, product quality is great, all that kind of stuff. I'm going to go there and I'm going to buy that stuff. Instead now, I go to my agent and say, go and buy me this thing, go and find me the best one of this based on user ratings and then buy me if it's less than £10. Right. So, I'm not going to buy you a new item, but I'm going to buy you a new item. I'm going to buy you a new item. So actually now Amazon gives me good personalization to it. be a good reason to go there. Well, actually, if my agent learns loads and loads about me over time, it can personalize stuff about me a whole course of whole range of websites, and suddenly that data that Amazon have got isn't valuable to them anymore. Because yes, they can personalize, but I can personalize via my agent as well. So this whole thing of light is do these people have lots of data about us have really good value with this they do with the large language model like it, yes, they would, but sooner or later, they're not going to need it. Because what's going to happen is that they'll learn all that stuff themselves and it will become a relevance. So suddenly, the AI agents and AI platforms have a big security portals. So there's a huge level of risk that goes without. They're actually learning more about us in one place without really needing to use cookies and all that kind of stuff that was limiting people like Google previously. And actually, they're taking kind of commercial organizations out of the conversation unless they choose to put them in the conversation. So if, for example, I am sort of a rogue AI company and everyone's using my platform and I aside, I don't like that brand. I'm never going to recommend anyone goes to their website and therefore I could hit their commercial revenues. And that's been the same with Google for a long time, remember? In that Google, you can decide what showed up in the search rankings and didn't. And I've had lots of legal cases about it. I've had to prove there is unbiased as possible. It's the algorithm and so on as well. So these are really big conversations that I've come back really quickly. And then on top of that is the impact on jobs. The problem is these conversations are happening post-event not before. You know what I was saying in that other episode about wisdom, thinking about the implications of what you do before you do it. I'm mapping that out a bit. I just see this all the time. It's just a race to beat the competition. And what I think is really interesting is that even the organizations that report on this are struggling to keep up. So I've been looking at some of the reports and data on this and people like McKinsey and all the major US banks are reporting on this. But the pace of change is so fast. A lot of the information that's going to the decision makers is way out of date. We're waiting on the next iteration of chat GPT and there'll be no news or evidence of what it's going to do or how it works. And then it happens and then it changes everything and it's massively ahead of any legislative curve. Well can I give you the only solution to this then? You've got a couple of things and this one is regulation. The problem is the regulators can't move as fast as the technologies that's not going to happen. You'd have to have universal global agreement that's not going to happen. So the only opportunity you really have here is one that we've spoken about in digital transformation over a long time, which is the organization that is the most agile wins. I'd come back to the thing I've been saying for a while, which is that you don't need to know everything. You need to know a bit more than your competitors. And that's why podcasts, live learning sessions, a learning culture. And I would say I'm on a training company. But the point being is unless you embed a culture of learning into organization, you're going to have a problem. So iterative, always learning, always learning, always trying, testing. But creating organization that is agile enough to change quickly. And the bigger and more complex organization is the harder it is to do that. But that's the same of all your competitors if they're all large and complex. Where it becomes tricky is if you've got competitors that are smaller and agile more nimble than you, they will take advantage of this. So I think that we're going to see the destruction of large organizations quicker than we will, smaller ones, which has been the case for a long time anyway. But those large organizations also have the cloud to go off and try and invest in this and so on as well. The thing I'll also bring you to is as we trust everything less and less because everything is deep fake and everything is, you know, is it an agent? Is it real? Whatever. That leaning into our humanity thing that I've come to a few times before, which is like, what makes our brand human? That's the stuff that's really important still. Let's really kind of focus on that. I would think about that. So I wouldn't panic about it, but I would be highly concerned. And I would just say, what is it that I can adopt from this that's going to be useful? What don't I want to adopt? But just looking at the fact that it will really change the market. And actually, I think we're in a for a period of disruption in the fact that we will come to adjust to these things and kind of learn what we want and what we don't want. But in the interim stuff is going to happen that isn't been planned. We are going to stop recruiting junior roles potentially, not we, but you know, a number of organizations will, some people will go through ruthless efficiency and they'll cut loads of roles. And then we'll go, well, what happens when you haven't got a talent pipeline? What happens when you haven't got a diversity of thinking in your organization? What impact does that have in the long term of the organization? So I think it's making sure our short term thinking and our long term thinking are actually married up a little bit and try to do some strategic planning. There's a danger in all of this. I had a conversation with my bank and I was frustrated. A PayPal payment hadn't been made and PayPal was telling me it had been made and the banks telling me it's not right. How do I sort this out? So I was having a chat conversation with my bank and I said, like, this is the situation. Can you do anything to help me with it? Like I just casually mentioned in the chat. Yeah, I'm a bit stressed about this because it's sorted. And suddenly the person that I thought I was speaking to clearly is an AI agent because it starts going off on one about, you know, do you need to talk about the stress? Like in a way, it was just unnatural. It's just unnatural. And you just think, gosh, how many of how much of that's going on? No one's looking at it. How much damage is that being done? Can I'm like, I'm not actually even speaking to it. So they've driven the efficiency there for sure. And they would give them damage their brand, but they've damaged their brand. And I think that's a good argument for actual people. I've had it for a long time with, you know, customer service. It says nothing like being given great human customer service. And I think if you slash and burn too quickly there, you will reap the negative awards for sure. You know, there's no reason why a machine should be able to outperform an actual human and a task like that. But I know everybody, every company seems to be vying that as a real ripe piece of use for this technology. Some of the voice say, I now is incredibly good. Wow. You just look at 11 labs at the moment. They've got the new voices where you can go to say it's so castically whisper it and things like that. It's epically good. You can hear the breath, right? Yeah. Which is like so, so, so much better. But is it as good? Actually, where does this position you? Is there a, an advantage within the marketplace of actually having real people that genuinely care? There is, there is. And let's come to that a little bit. I'm getting lots of people saying to me in a lot of my students, a lot of people that listen to the podcast saying, I'm really worried about this. I'm, you know, I'm a junior marketer. How am I going to get a foothold in my career? It's the first thing anybody does when you show them a genty K.I. Yeah. So they start testing it to see, could this replace me? The fear is there and I was, and when they realize actually, no, it can't because they're a bit more senior or a bit more strategic than they relax. But that is short term thinking in my opinion. So what I've been saying to students and people that are kind of studying marketing, getting into this, it's not a bad career to get into. Don't panic from the point of view that we're getting this technology first. But what you need to be really good at doing is critical thinking problem solving. Yep. You need some strategy frameworks. You need the cause of, so those fundamentals, business fundamentals of marketing become really important. The ability to prompt really effectively, and I, ever ago, still a really good approach to prompting is really rare still. You know, that whole piece had to, and we will do an upcoming episode on some prompting techniques and good prompts and so on as well, just to give you some deeper tips on that. But I think that critical thinking problem solving, understanding, you know, the economics of how the world works, these are kind of core skills and it's always been interesting to be that marketers don't really understand economics. And actually, they really should because they're going to see how does their organization fit into the broader world. How does that fit? So there's those kind of core business skills, I think, really important. So we will do a follow up episode on, yeah, what skills you're reporting in the marketplace right now. But I do think that organizations need to be looking at what they want to do this in the short term, but what will be the long term implications of doing that. And unless we're able to do that, we're going to cause ourselves lots of problems and also like with any of this stuff on the surface, the shiny like, I can do this, you can do that. Test it, refine it. Now, one tip with AI agents, get it to do something, see where it goes wrong, rewrite your prompt, start it again. Okay. Remember every conversation in an AI is in a context window, meaning it's remembering what you've said before in that conversation. So very often when you have a conversation that goes off at a tangent, it gets it wrong, learn from it, improve your prompt, start the conversation again. So kind of go through that process. The other question I've been having is that as these things are more and more powerful, they are using more and more processing power. And actually, there is a carbon impact of using this stuff as well, which there is. And if you do a very simple AI prompt, you know, it's using a sentiment of carbon. If you do a really complex prompt, it's going to use more carbon as well. But what I would say about all of that is that, and I have big debate about this as a really interesting organization recently, is that yes, that does use more carbon. And if you'll just sit there will you nearly playing with it, you're using it for the same time, you sitting on Netflix for eight hours at the weekend and watching Netflix will obliterate that in terms of the amount of carbon you're using. Right. And I've got a chart and I will put the chart into the show notes in terms of like, what does a Google search use, what does an AI query use, what does watching TikTok videos for three hours a day use, what does streaming Netflix use, that video streaming is off the charts comparison. So I think that we've been a little bit unfair in honing in on AI and saying, I use those to publish and be doing this now. There are environment to impact. There are data centers that are needed that's using huge amount of water. There is environmental damage. I'm not denying that. But I think it needs to be a conversation that's held in context as well. If this is something that's of interest to you, please message in and we'll do an episode on it because I've got a ton of research. we've been using for a couple of clients recently. So short term, amazing. Long term, what's the impact going to be? If you haven't played with the AI agent in TechPT, go and do it. It's going to blow your mind a little bit as well. Think about how it's going to impact your organization. We have got a half day masterclass for target internet members coming up on effective use of agents. So if you're not a member, get signed up for that. Also, we've got an update session for our newsletter subscribers and members. So targetinternet.com. For those of us newsletter, you've got those one-hour update sessions every month with me and there are a lot of fun. We're going to really big audience. Everyone's coming together for those. There's a lot of fun as well. So go off and have a play with it. We'll do another session soon, but there's a lot to think about and a lot to think about how it's going to impact your organization, your role and the future of marketing. [BLANK_AUDIO]