Episode 219: Agentic AI: What to automate, augment and keep human, with Baringa

Speaker A: Foreign.

Speaker B: Hello everyone and welcome to the B2B Marketing Podcast. My name is Kavita Singh. I'm head of Growth Solutions content at B2B Marketing. And today I'm joined by Adam Berthelsen from Baringa who will be speaking at B2B Ignite. We are here today to tease his upcoming session on AI, obviously a very timely topic. So Adam, welcome to the podcast. How are you doing today?

Speaker A: Very well. Thank you very much for having me.

Speaker B: Yeah, thanks for joining. Um, I briefly introduce you, but, um, why don't you introduce yourself and tell us a little bit about your upcoming session.

Speaker A: Yeah, absolutely. So I'm Bethesdon, um, AI workforce transformation lead at Baringa, who are a, uh, global management consultancy that work across sectors, um, including energy, financial services, media, um, life sciences. Ah, I lead our work at Baringa on the human side of agentic AI transformation. So really looking at how organizations identify priority AI transformation opportunities, design roles, collaboration models, looking at the reskilling of workforces and ultimately also the governance of how you actually capture value from AI, not just deploy it. Um, so my job actually involves very often being in the rooms where AI is changing, how work gets done across various sectors and, and looking to bring in lessons, um, how to manage the intersections between people, data and technology for our clients. And most recently I've led transformation programs for sort of law firms looking at adopting digital AI technologies, designed future AI workforces for, uh, a few major government clients, um, led AI upskilling initiatives for media organizations. And I've also looked at kind of an agentic AI workforce redesign for an international technology company. So as you'll tell, I've got a bit of a spread. You know, I'm coming out kind of B2B marketing very much from an outside in perspective, but hopefully can bring a lot of very interesting lessons from various sectors.

Speaker B: It sounds like you've done a lot around AI, so I'm sure we could take this conversation in so many different directions. But I guess you know, in your experience across sectors, which type of tasks are AI systems absorbing first? And what does that pattern sort of tell us about how marketing roles are sort of likely to change in the future?

Speaker A: Yeah, absolutely. And I think just try to be very brief in terms of when we're talking about AI. Of course it's a very, um, uh, kind of a word packed with lots of different definitions. So very briefly, if I just define agentic AI as, um, if you think about AI at one end as something that simply you prompt an AI tool and it responds, you decide what to do with an output. So uh, for example if I use ChatGPT, like Google search, I just put in a request, I get an output. You know that, that sort of one end of the spectrum. At the other end, genting AI solutions are more those sort of agents that autonomously pursue goals that you set using tools of memory across multiple steps or loops, you know, without requiring your input. And so tools like OpenClaw that come out sort of led the way at that end of the spectrum. And of course there's a lot of a big wide middle ground where humans and AI, uh, agents kind of collaborate together in real time where you know, AI tools have access to workspaces and humans steer them. So for instance, when I've uh, been building a history game on Claude code, very much it's a collaborative effort. I really wanted to personalize it. I give it the steers I want to. So my point being that each of these um, different types of AI have their uses and very much obviously not everything should be uh, agentic. Um, but I just want to make clear that there is a big shift at the moment from AI that answers to responses, um, to AI that actually acts um, in terms of your question about what tasks that these systems are kind of absorbing first. I mean there is a pattern that is quite consistent that I'm seeing across every sector. Ah, um, which again probably won't surprise most people listening to this, which is that uh, AI is absorbing anything that's repeatable, high volume and particularly things that are explicit by the sense of um, AI can actually understand the context and have a clear um, kind of view uh, on whether their response, whether the output um, is accurate or not. Right. It can actually be judged um, in a machine led way. And so for law, for example, you know, things like document review and research, you know, synthesis obviously fall into this category. Um, in financial services it's looking at reporting and you know, routine analysis, um, and so on. Right. So I think for marketing, uh, B2B marketing, I imagine you know, that a lot of the uh, media opportunities across this is content creation, copy creation, um, and obviously some personalization. And again customer service is another area that uh, we're seeing a lot of um, early adoption of these sort of agenda capabilities. Um, and I think the thing maybe we can get into a minute I think is around, you know, this is obviously an entry point for a lot of organizations, uh, looking at how we can ah, really automate and you know, massively um, you know, build the amount of um, kind uh, of content that we Create the production value. But there is a constraint to this still. Like it's not just a case that we can implement agentic systems and um, deliver lots of value very quickly. Right?

Speaker B: Yeah, absolutely. I think based on some of the customer conversations we're having, I feel like um, those sort of repeatable tasks that you mentioned. Yeah, that definitely seems in line with what we're hearing in the market right now. Um, I think something else that we talk about is the fact that creativity, you know, especially in B2B marketing, that's not something that AI can kind of replace. But I'm curious, you know, things like creative judgment, taste, brand instinct, those feel like human things. And we talk a lot about that sort of remaining human. Are they actually at risk from AI adoption? Just curious. Obviously you have quite a lot of experience.

Speaker A: Yeah. So it's on a surface level they, in some ways they do, they, they do appear to be at risk. So AI systems actually surprisingly empathetic. Right. You've seen that the growth and explosion of the use of uh, companion agents and chap bots for example. So the assumption that AI couldn't um, provide outputs that uh, um, appear to be creative. And again, I think there was a, uh, book the other day actually released. Wasn't there a uh, piece of fiction that might have won an award that was potentially heavily AI generated? Right. There clearly is um, a capability there. Ah, interestingly, I think the risk here is actually less about um, devaluing those skills, those human traits. And I'll come to these human native capabilities that I'll talk to in a second. Um, but I think the risk is less that these are going to be devalued in the sense that all AI systems are still ultimately producing yesterday's best answer, as it were. So they're obviously trained on historical data produced. If you use them extensively in your business or personally, you're always going to get some sort of form of output, ah, that is not necessarily unique in terms of um, bringing taste forward, so to speak. It's going to uh, ultimately converge, um, into the sort of the best kind of medium over time. So I think there's a piece there in terms of. I think that there is still value for humans to be bringing judgment and taste and that brand instinct that I think AI can't necessarily accurately convey as the world is always changing, it's inherently unpredictable. I think the risk is actually that it doesn't replace these capabilities, but it will um, impact the amount of practice and experience that people and uh, humans can and can sort of take in their day to day to enable them to build and maintain that taste. And for example a couple examples. One is uh, consulting what we do. You know there is a reason ultimately that a lot of consultants have spent a lot of their time um, developing, building um, PowerPoint presentations and Excel spreadsheets. Because it's only through, you know, uh, those hours of actually considering a task and performing it and you know, that sort of osmosis process of being in the meetings, capturing those that you do build that understanding awareness of what works and how you should present content, etc. You can't get that you're not going to build those skills. If you're somebody receiving AI generated beautiful outputs, that just isn't going to happen in the same way. And similarly, in another example from healthcare, from some of the studies out there, a lot of the um, AI assisted diagnoses, for example, they've taken over a lot of routine diagnostic work. On the surface of course, they're fantastic. They're helping with um, improving quality rates and reducing the risk of missing uh, cases. Unfortunately of course human nature is, you know, specialists have often over relied on it and approving these, these um, uh, kind of diagnoses too quickly and there's research showing that actually over a course of just less than 12 months, people's intuition rusts. Right. So the actual ability for people once the AI was taken away and in this clinical study, uh, that was only um, delivered last year, that you know, the kind of the crutches removed and the performance radically reduced and I think that whole mechanism will apply to creative judgment and marketing um, as well. Right. Like if you're using AI too much as a um, you know, cognitively offloading onto these tools, you know, you're only going to find out unfortunately when a high stakes brand decision goes wrong potentially.

Speaker B: Yeah, no, definitely. I think the experience of you know, looking at what insights you have and then coming, you know, coming up with maybe a strategic plan, um, um, 100% simplifying the whole process because you're taking in so many different factors. That feels a lot different I think when you're relying heavily on AI to kind of, you know, almost go through that thought process for you. Um, so yeah, it does definitely feel like it's a, it's a game changer in that sense. So I know you touched on it earlier, but what human capabilities and skills sort of matter more now, not less as AI sort of gets more capable.

Speaker A: Yeah, I think we sort of see is sort of five kind of human native capability Kind of groups. Right. Uh, and these are things that I think even though AI um, capabilities continue to exponentially increase and if you looked at sort of, there's a sort of meter benchmark that shows the exponential increase in the length of tasks that uh, tools can autonomously complete and particularly around coding, um, just showing how far these tools can continue to work to produce high quality outputs. I think they will miss these sort of five things. Right. First is around translation and judgment. So no matter how good our ah, agents and AI tools become, you need to give them intent and give them that intent quite precisely in order to develop something that's actually useful or what people want. So I think there's even more focus on the brief in terms. And there's obviously been a lot of talk about prompt engineering. I actually think prompt engineering um, as a skill set is less important and it will increasingly become less important than again just having the judgment and that experience to know what you want and what the brief should look like in uh, terms of you don't need as much of the uh, understanding of how to give it to the AI, if that makes sense because the AI can interpret how you give it, but you still need to know what you want from the, what you want the AI to do, what it should look like. So that's the first kind of category. Um, the second area that again is just not going to go away anywhere is this around um, accountability and trust and also credibility kind of gets wrapped in here. So ultimately um, we need someone who is ah, a name team behind a lot of high critical output. So whether that again might be in this case a brand campaign, you're not going to necessarily sign off on it. If it's high uh, cost and potentially high impact to your business, you'll need someone who owns a uh. And then similarly of course we might have already the capability for fully autonomous airplanes. I don't think I'm not going to get in one of those without a pilot purely for the reason I have that trust and that there is someone clearly accountable there. So there is that piece too which blends into the third area of embodied presence and behavior change. And again maybe less relevant potentially to some of knowledge work. But there uh, is always going to be a role I think, particularly when we are trying to transform organizations for having um, humans on the ground physically next to each other, being able to help embody um, kind of what needs to happen next. So again I think there's going to be a capability there as well. Um, then fourth, um, again the piece we talked about Erna Kavita, about taste and this idea of provenance, um, that accumulated domain judgment that people build through their work and experience. That's what's going to make your output or your work distinctive rather than purely competent. And actually um, is going to distinguish you from everyone else.

Speaker B: Right.

Speaker A: So I think that becomes even more important as we just discussed. And ah, then finally there's the relationship, um, kind of capability and that need for social connection. There's going to be a lot of clients, whether it's B2B or B2C. Right. That people want to feel like they've got a real relationship, they want to be known, they don't want to just be served. And again, there will be again a spectrum there right there. Sometimes you just want to get your car insurance, for example. You don't necessarily need a relationship around that. But at the same time, if I'm buying a car, perhaps I might actually want a little bit of a relationship with the person I buy it from just to feel like I've got that connection and again blurring into some of those other capabilities, that trust and that embodied presence. So those are the five kind of human capabilities I think that matter even more now than before. And I think that the one thing I'd say against these, that uh, they're not again. The conversation last year or two, which is shifting now, has been very much of AI taking jobs and looking at how do we defend our human capabilities. I think reframing this to look at these as where organizations get a competitive advantage, where everyone else is getting access to the same models, training the same data, producing yesterday's best answer. Uh, I think there is an opportunity to reframe this and really try and focus on some of these areas that uh, a lot of organizations that aren't necessarily in the technology domain already have a lot of these capabilities. Right. Um, and a final point there, even in software engineering, um, although there's been an explosion in Vibe coding and this idea that we can democratize coding and everyone now can be a coder and therefore what happens to software engineers? Um, Vibe. There's evidence that Vibe coding commits, they introduce far more vulnerabilities, you know, risks into the code base, you know, more than nine times the human only rate, you know, you still need expertise in software engineering or marketing to detect the risks and issues that these tools are producing. Right. So yeah, I think there's going to be a big focus on these capabilities in the coming, in the coming years.

Speaker B: I think when you have a certain partnership and you feel really good about it. That's like a feeling that I think a lot of us can relate to walking away and thinking like, oh, this is going to go really well. We're on the same page. And I think, yeah, I don't think that's something that AI can necessarily, necessarily cover. And it's something that I think, um, is super important. We actually, um, we spoke to client side marketers about like what they're looking for in, in agencies and one of their top answers was just having someone who's nearby in terms of location. I think part of it was like they want to be able to go and get a coffee with them. Um, they want to be able to build a relationship, they want to be in the same time zone, like just things that like, they just can't be replaced in that sense. And I think that's really interesting. Um, if someone is listening to this and thinking about how to develop their own team's AI capabilities, you know, what would you tell them to kind of focus on that maybe organizations might be getting wrong today?

Speaker A: Yeah. So particularly some of these human capabilities as well. Right. I think a lot of the, you know, there's been a lot of focus on, on AI training and AI fluency. Right. Across organizations and you know, is one area to build some of these capabilities. And I think that that's necessary but not, not necessarily sufficient. And certainly the discussion we're just having is not where the competitive advantage sits. Right. So again, if you're an agency, as you said, it's great if everyone can use the same AI tools. Well, that doesn't matter if that agency is local or not. Right. In terms of, you know, whether you're, you're not building, you know, on your competitive advantage there. Uh, um, so particularly for knowledge work, some of the capabilities we just talked about. Right. You know, translation, judgment, taste, accountability, that's what's going to make your work just more distinctively valuable rather than just churning it out faster. Um, and so I think what I've seen in some of the work I've done in the global media organization, for example, we've refocused not just on the technical what. Right. But very much on the how and the soft skills around that. Right. Um, and really in particular, seeing a lot of organizations focusing on that adaptability piece too, because again, um, initially, you know, it's great we can train people on the, the specific tools that come out. Now, as we've been seeing every few months, they're changing and as I referenced earlier, the nature of Prompt engineering, again you can develop fantastic prompts, but actually the next tool comes out, you don't even need it because they're so intelligent now working out what your intent is. So I think yes there is some technical training needed there, but I do think um, some of the softer skills, behavior interventions and the culture are just as important. And also of course to promote that engagement and buy in from staff because again I think the other thing we're seeing is that growing capability gap between those who are accelerating, advancing and whether in their personal lives or of course at work playing with more gentic um, solutions and systems versus those who, as I said, maybe some people are still very much on the initial um, kind of stages of simply interacting with AI as I said, very much as they would have done previous digital tools or search engines. Right. So, um, yeah, I think um, that is the key thing I focus on is not really just the technical training but that broader, both kind of individual, um, kind of soft skill capability. But then also at the team level, that culture one, uh, as well.

Speaker B: Yeah, definitely. I think taking a step back is super valuable in that sense. Um, it's funny because I think at Ignite we noticed that uh, the conversation around AI has sort of evolved. A couple years ago, uh, there was a question of whether to actually use AI, uh, which I think is interesting because it shows how fast things have sort of uh, moved with AI. The conversation has kind of shifted to, you know, why the returns are not matching the investment. What is the thing most organizations are moving as they try to move from experimentation to like real strategic value.

Speaker A: Yeah. And to your point, you know there's, there's always researchers in there every, every uh, few months on this. I mean most recently I saw one, I think from an mber, ah study of over 6,000 executives saying 80% of organizations are not reporting meaningful productivity gains. Right. As you, as you say, is a challenge that a lot of our clients are facing as well. Um, and I think what we're seeing a lot is uh, again the challenge that I think two challenges actually. One is that organizations still treat AI as a tool rollout rather than that wider operating model change. And as I just referenced, looking at the wider culture and um, decision flows, accountabilities, etc. There's that challenge. The second is also the kind of the sprawling pilot and not being intentional and decisive enough around actually um, focusing on where are we actually getting value and um, stopping pilots once they've proven whether or not they're successful. Um, and so yeah, I think the Organization that we've seen that are developing or generating almost like a kind of compounding return right. That is not a sort of a one off kind of cost saving exercise which then turns out once you start adding on um, token costs and all the other AI kind of overheads right. That actually doesn't deliver value. The thing we're seeing that they're looking at is very much how do they transform their governance, how are they realigning performance incentives? And again it's fascinating to see how many organizations really struggle with that thinking about how um, they take their performance management across the organization and readaptate AI because obviously it's the classic uh, almost chicken and egg right. Until you've seen the technology working uh at scale are you going to realign incentives around it? But that's a key element that we've seen when that goes well um, and then finally that kind of decision right piece and how you build learning loops around it as well and to improve collective judgment. And I think again there's the tendency of organizations to very much focus on AI almost like as an individual um productivity enabler um when of course a lot most of our work is collective right. It's team based. Um, there's always going to be handoffs and there's going to be bottlenecks we create elsewhere. Ah. And similarly getting everyone to build their own agent for example. I think the early evidence is showing that that doesn't seem to be a very productive case. Right. In terms of a. Agents require careful setup and design but also maintenance and the average individual employee or staff that's not really the best use of their time. Right. So thinking of the best kind of teams I've seen in terms of generating return and changing the ways of working effectively are looking at it at more of a collective level. Okay, what are the kind of the team agents, organizational ones, you know those different levels as well. Absolutely critical I think to this too. Um, and I think finally um, you know actually you know writing down what AI is permitted to do and what requires human approval. Again I think there is almost um m sort of a uh hesitation for organizations to do this. It's interesting. I'm not quite sure exactly why um in terms of why organizations um hesitate to do that but I do think being really explicit to everyone in the team and potentially to customers and to regulators and again we've worked recently with a global technology company looking at some of the compliance uh processes and bringing more agent ways, working across them and trying to engage with your customer, suppliers, regulators on That I think being upfront is going to be crucial. And I think until you've actually as documented kind of what will require human approval, intervention, what can be handled agentically, I just think you're still in trouble for later. Right. So I think that's again a key gap that ultimately then um, handicaps any um, kind of agentic sort of transformation and stops that investment providing returns.

Speaker B: Yeah, it's interesting because you touched on productivity gains which kind of leads into uh, the cost factor. But I think marketing leaders are also under pressure to show AI is saving money. But do you think that's actually the right measure of success at all?

Speaker A: Unfortunately again as you said there's been a lot. The primary focus has naturally lent on cost efficiency with AI because it's an easy uh, way to justify investment. Because obviously such a new capability that does need funding to stand upright, the classic investor save it needs that investment. Um, and obviously as we've discussed at the outset this podcast, the high volume repeatable tasks do lend themselves to some automation and to some cost savings. Um, but I do think it's the wrong measure. Uh, um, because I think the opportunities and the human LED capabilities we discussed before, AI makes possible things that weren't before and enables you to differentiate yourself. And I think that that's important. Um, but I do, but I do think in terms of staying with cost for a second, focusing on cost now, a lot of organizations I've worked with have underestimated, I think the level of additional cost that AI brings. And we're seeing this obviously at the moment play out the last few months where um, there has been a lot of sort of subsidization of AI sort of token costs. So the amount of money it costs any employee or any organization to run a workflow with AI, um, that's increasing and agenda use is very expensive because these tools are uh, running more uh, regularly all the time and running on more complex tasks, spitting out often multiple agents at a time. Um, and as well as those token costs, that infrastructure as well, including human roles, you can't just create an AI agent and just leave it permanently. You will always need someone or some team to manage it, maintain it, make sure it doesn't drift, etc. So those costs I think have been often underestimated and when organizations therefore have tried to use it as cost out. I think that's part of the reason the previous thing about the feeling that productivity hasn't shifted, it comes from, and I think Klarna has also been in the news, hasn't it um, quite recently around they tried to optimize for cost reduction and particularly around their customer service and they've actually now this year have gone back to try and rebuild some of the human layer they dismounted because of issues with quality. The cost out thesis hasn't really ah, landed there. Um, and I think it's also the point we discussed earlier. Cost out is directing investment towards eliminating human work rather than building capabilities that make AI work actually genuinely valuable. Um, so again it's the classic you can produce more and more output that's competent but uh, if it's not distinctive I don't think you're going to compound your value. And again I think um, Uber's coup came out with this as well recently about you know. Yes, they are able now to generate a lot more code in their uh, you know, their software engineering. But actually seeing that return on value in terms of customer features, they've not been able to see that that directly linked. And I think again money, using money as a measure of success there just isn't going to encourage um, you know, encourage that return.

Speaker B: Yeah, Um, I think on sort of productivity it's, I guess it'd be great to know what does productivity actually mean when agents can produce content at scale as well.

Speaker A: Yeah, so exactly. So again sticking with software engineering um because I think uh, the furnace ahead right in terms of agentic kind of um, integration. Right. Um, as I said current coding agents can produce over 17 times as much code. I think some of the research showing um, and therefore you are seeing increases in software release. But as I just said in terms of the Uber example it's hard to see ah and that, that you know, kind of the full. You're not getting m. That that you're not getting 17 times as much useful, valuable software, you know off the back end and I think um, there's the sort of the gap is again the sort of human bottlenecks. Right. Uh, so although at the production level in the middle, so again once you've kind of made your, your design, you've made decision of what the brief should be and again as I said earlier you still need you know, a lot of human input into that, that, that you know, cards that probably shouldn't be autonomous, you might be able to massively um accelerate that middle, you know, productivity level. But in most cases ah, I'd argue you still need a human to review that output. Right. Um, you know, unless it is very um, generic, you know, copy for example that you're quite happy, you know, you are Just going for volume. Um, and kind of as I love the term, you know, AI slop. Right. In essence, if you're happy with that, and that's what your business is, that, that's great. But of course, as we said earlier, that's not going to differentiate yourself. And I think as we're seeing in, you know, as I see in consulting. Right. You know, me and my teams, we can respond on a much, much quicker level to way more proposals and deliver more projects, you know, with some of these AI tools to accelerate us, uh, but in order to provide our clients with, you know, the great experience of working with us, as you said about your, you know, the agencies nearby, like, you know, that's really important, you know, the experience, you know, throughout the projects, but then also fundamentally make sure that what we do is delivering real value. Right. And then that requires our human expertise to be applied to know that what we're recommending is the right thing for our clients, rather than just what Claude says it is, for example. Right. Um, and so I think, you know, that, that, that sort of rule of thumb of if you don't understand, you know, what you produced of AI, right. You put something in, you've not bothered to review it and read it, which, as I said, is the bottleneck. You know, it's AI slop, and ultimately it's going to cost you time, tokens and potentially turmoil. I think if it generates issues, um, you know, and it's probably not going to deliver much value. Right. So going back to, I think what marketing leads probably need to think about tracking, um, and incentivizing is a key question. And I think some of the technology companies recently have tried implementing kind of leaderboards of AI token uses. But of course, that, as we've said, has encouraged perverse behaviors where people will just use AI for the sake of it. Right. Um, so I think reframing in an agentic world where there's going to be metrics like what's the how? And it's going to be specific to, um, different domains, different industries and different organizations. What's the quality of your human judgment per unit of age and output is probably a better measure than the volume of output per headcount. Right? So thinking about the quality of, um, whether it's from the brief, uh, to the sort of the thought leadership piece or whatever, how do you assess that quality, um, of it per output, uh, rather than the volume, um, and finally, on top of that, I'd also point out again back to our cost piece that, you know, those Token costs that are associated. If you have poor brief quality, you're going to have more iterations and, you know, you're going to spend more money getting, you know, that, that production. Right. You know, yes, you produce even more. But as I said, if you're not actually guiding it at the outset and managing it throughout, that. That production is actually said, not. Not necessarily cheaper anyway. It is still often cheaper, um, you know, to, to get a human to do a lot of the work in the production. Um, and one other thing I would say is, well, this is, again, the challenge is this requires people being even more, uh, capable and supported than ever. Right. Because actually supervising agents and doing a lot of this multitasking now, which is what I find in terms of now working, trying to do more with the tools I've got, um, that production layer to manage it. It's really draining. It's hard, it's mentally challenging, it's exhausting. And engineers say that they're knackered by lunchtime. I find that too. I actually find I'm probably more mentally tired in a week than I used to be. Right. So I think again, there's that interesting, um, kind of unexpected thing that we thought that agentic AI would make us all, um, work less and that we'd be outsourcing more. Critical thinking. I think it's the opposite. Right. We're having to do more critical thinking. Um, yes, to produce more. But ultimately I think it will require people at the front end and at the accountability, so judgment end, to make sure that we actually get the returns we want, uh, from AI.

Speaker B: Yeah. It's funny because I think the excitement around agentic AI and how it's sort of shifted to actually we have to think quite critically and even a little bit more strategically. I think it kind of mirrors how we viewed AI a few years ago when there was a lot of excitement around it, maybe a bit of nerves. And now people are realizing it's becoming a lot more complicated and has inevitably led to, you know, sort of generative versus agentic. And I think it's really interesting that the conversation has sort of evolved. I mean, that being said, AI is obviously moving fast and the pressure to deploy it is very real. So I guess how do leaders make smart decisions about what to automate, what to augment, and then also what to keep entirely human? Because I feel like those are the three buckets that we're sort of working with.

Speaker A: Yeah, absolutely. I think. And again, there's always going to be some trial and error, as you said, because As I said, the capabilities, tools keeps changing rapidly. Right. Um, so for now it's certainly the case that you have to experiment now because otherwise you will be um, left behind unfortunately. Um, so I think taking, I guess pulling together some of the threads of what we discussed so far, I think there's probably almost like three categories, right? In terms of automate and automate and then keep human rights. I think automation, as we said, if there is stuff that is explicit, repeatable, high volume and low risk, ultimately to the organization, absolutely, you should look at that. So I think there are some clear signals in the organization to identify there. I think then of course that middle bit of augmentation, anything that clearly will still require some human judgment but will benefit from the sort of scale and the analysis kind of um, features that AI brings. Right. So being able to take much more complex data sets, one government client I worked with, we um, brought in a lot of kind of consultation, um, sort of from external agencies. Right. There's a lot of external spend on those providers because obviously it was a challenge to pull in so much data and analyze it. But with AI they were able to bring it back in. Right. So that there is a, a case there of reducing external spend but actually um, keeping some of that in house. So it's an interesting kind of automate and augment example. But I think that that augmentation piece, as I said, a mix of judgment and analysis. But then within that there are going to be lots of quite a surprising number of tasks that I think you want to keep human in terms of anything that you know, has capability, develops or are ah, capabilities that your organization needs to maintain. Um, you know, as I said earlier about like create you know, that kind of brand distinctiveness, creativity, you know, as I said, you need to keep that kind of practice going. And whether that is a case of um, again different human kind of AI collaboration models, you know we talk about like centaurs and cyborgs, et cetera. It might be a case that uh, um, individuals, you know, kind of, you know, have dedicated, you know, use the AI, you know, to develop some of the, you know, like that thinking, that strategy, that content, but then have certain times every week or every month where they have to go through manually and sort of almost like a training wheel and do it by hand so to speak, to maintaining that capability. Right. So there are ways you can mix it uh, up. But I do think um, that bit about anything that is distinctive to you, anything that requires or needs accountability to your clients and stakeholders, again that's something you really should be trying to keep more human than I think potentially a lot of people are considering at the moment. And as I said that balance is not fixed. Like capabilities are increasing. That might change a bit, but I think that way of approaching it shouldn't change too much as the technical capabilities increase. Um, and then finally I say as well, and again, as I referenced earlier, ah, that need to be explicit and actually document write down what you are happy with AI to do autonomously, what needs approvals, you know, what needs escalation. Again, I think that that shouldn't, ah, take too long because again you don't need to go to the end, you know, too granular with it. But I think being really clear about and rethinking kind of those decision rights and workflows I think is something that's worth doing almost on a blank sheet of paper.

Speaker B: Yeah, I'm curious if you kind of share the same sentiment, but I think we're quite lucky that we have gone through so many years of having to do things manually and have gone through the shift from doing things manually to automating. And I, um. In a way that's so much better than kind of jumping into sort of automating things automatically. And um. Yeah, I don't know. Do you kind of agree with that? Because I think it's nice to know what it feels like to do things manually because you can appreciate sort of the automated tasks.

Speaker A: Yeah, absolutely. And as you said, I think, um, what's fascinating, we're seeing at the moment, the newest models are coming out right again, even for PhD professors, etc. They are struggling to understand the black box kind of concept. Right. They really struggle to understand now how exactly these tools are getting to their answers. Right. And until you've done the work right, you know, as I said, whether it's um, as I said in some of the, like developing an upskilling program, for example, having done that myself conceptually, you know, start to finish, I can use some of these tools to help me, you know, accelerate some of that and to challenge my thinking and to augment and build on, you know, the solutions I develop. But as you said, if I, if I tried to, if I rely on it and you know, do it to do the end to end without never having done that manually, I can't credibly, I think, understand whether it's good or not. And similarly, a lot of the time, both from personally or from ourselves, but also obviously from clients and from our colleagues and then customers, I think people do Often want to, um, not always, but sometimes they want to know how something has been made right. Have that reassurance. As we said, that trust piece comes from, uh, having that reassurance that someone, a human has gone through, you know, the production of this content. So I totally agree, Kavita. I think, um, we are lucky in that sense that, you know, a lot of us. And as I said earlier, I think the, you know, expertise that you've accumulated in organizations is a competitive advantage now. And it's about rethinking, you know, some of the culture as well as the skills and AI architecture and infrastructure around people to maximize and get that value. What this means in the future for, um, you know, for new joiners, for um, more junior employees. It's a massive, you know, it's gonna be a massive challenge, I think, you know, to how we integrate them in a way. And I do suspect that we will see some sort of, um, kind of almost return, I guess, to some, you know, more like apprenticeships and things, I imagine, in terms of, you know, that need to be more explicit actually. Look, we're going to invest in people to build those capabilities that we need to use AI rather than, you know, purely use AI to deliver the work that was done before.

Speaker B: Yeah, absolutely. I mean, I could keep talking to you about this for forever. Uh, but I think we have a couple minutes left. You're obviously speaking at B2B Ignite in July. I guess without giving too much away, what else can people expect from your session and why should they? Make sure they're in the room for the day.

Speaker A: Yeah. So really excited to speak at B2B Ignite. So my session is called Agentic AI and Marketing. Um, Early Lessons from Other Professions. So, um, and the premise, as we've already covered here, is, uh, that marketing is not the first industry to navigate this transition that's underway. Um, software engineering, consulting, law, very much on the agentic curve. And as I said from my, um, experience, I'm hoping to bring around what I've discovered around what happens particularly when you get the human side wrong of some of these, um, transformations. Um, so I'm really trying to bring these signals and examples, real world examples into the room so that marketing leaders can act on them before they arrive. Surprises. So, um, and I really hope people will leave my session with a bit of a shift in the question they ask. So as I said, rather than focusing purely about which tools to adopt around AI, actually let's think more considerately around how we design the human side to make those tools generate lasting value which I think is a bit of a different conversation. Ah and one many people have not had and hopefully having that conversation up front I think will support the return on investment that people are looking for from AI. So if you're invested, if you're a marketing leader investing in AI or thinking seriously about what your future team needs look like, I think this session should be great for you.

Speaker B: Awesome. Well I think that's a great note to leave on and that about wraps up our session. If you like this episode we highly Recommend Attending our B2B Ignite conference which is taking place in London on July 1st. You'll be able to unlock a ton of B2B marketing insights including Adam school session. We will also leave a link in the description for the event information as well as an exclusive discount code which you can use at checkout. Thanks so much Adam and stay tuned for another B2B marketing podcast episode. Thanks so much. Sam.