Speaker 1Hello and welcome to Don't Stop Us Now AI Edition. I'm Claire Hatton.
Speaker 2And I'm Greta Thomas. AI has been described as the most transformative technology since the harnessing of electricity. And we are here to keep you in the loop for what you need to know and do in order to stay relevant in this fast changing world.
Speaker 1Each episode, we bring you leading experts and together we explore how you can stay ahead of the curve and in the know for what skills will be valued in the future, how jobs may change and how industries will evolve. So why not subscribe to stay in the
Speaker 2loop and without further ado, enjoy this week's episode.
Speaker 1Welcome back to Don't Stop Us Now AI Edition. If you've ever wondered what it would actually look like to be a business owner, you're probably wondering what it would look like to be a business owner. And if you've ever wondered what it would actually look like to be a business owner, you're probably wondering what it would actually look like to be a business owner. And if you've ever wondered what it would actually look like to be a business owner, to have a colleague who never sleeps, never calls in sick, and can process thousands of conversations at once. Today's guests can shed a little light on how that's a reality for some right now. Our guest
Speaker 2is Patrick Forth, an executive director at Future Secure AI. Now, Future Secure is a company that's building what you and I might call AI agents, and they call digital co-workers.
Speaker 1Patrick brings real technology transformation expertise to this conversation. He spent more than 20 years at Boston Consulting Group, where he led their global technology, media, and telecoms practice. And he was founding board member of BCG's Digital Ventures, which has helped launch over 200 new businesses.
Speaker 2In our conversation with Patrick today, we explore all kinds of angles relating to agents in the workplace, and cover some of the most pressing questions we're probably all asking about agents right now. Patrick shares real-world examples of agents in the workplace, and he's going to talk about some of the most AI co-workers already in action, why Future Secure is focusing on complex use cases to automate, Patrick's take on the future of white-collar jobs, pretty interesting, he's pretty optimistic, and why he thinks we should treat digital co-workers more like colleagues than software.
Speaker 1If you want a glimpse into the future of your possible workplace and colleagues, then sit back and enjoy this episode with someone who's at the coalface helping companies in the workplace. Choose use cases and deploy digital co-workers right now.
Speaker 2Patrick Forth, welcome to Don't Stop Us Now, AI edition. Thank you. It's really great to have you, and we're super intrigued and fascinated to learn more about what you're doing these days. Before we do, though, a question we ask all of our guests is, imagine you're at a dinner party, and you're sitting next to someone who you've not met before, and they turn to you and they say, so Patrick, what do you do? How do you typically answer that question?
Speaker 3Well, what I do is I work for a company that builds artificial intelligence workers that work alongside people in businesses. Generally, that either intrigues some people or those people who are a little bit technology shy may turn to the person on the other side to talk to.
Speaker 2I'm sure, I'm sure they do. Now, that company that you work for that builds these workers, is called Future Secure AI. Would you like to explain a little bit more to listeners about what Future Secure does?
Speaker 3Yes, sure. It's a company that's been going for about two and a half years. Started in Australia, but we moved the center of gravity now to the US, and we have about 400 workers. Human workers? Well, actually, we have 400 human workers. We have a lot of AI workers as well. So we work with enterprise customers in all sectors. I'm in charge of companies that work in social impact, so for example, areas like disability insurance and aged care. Also, I work with private equity who have typically lots of portfolio companies. I work with technology companies and logistics and transportation companies, and all of these companies have a need to build agents or what we call AI co-workers that can help them get better business outcomes. It's a new and disruptive technology. It's moving very fast. And at Future Secure AI, we like to think that we are creating both a new category of value proposition, as well as a new way of working with customers, which is pretty exciting.
Speaker 1And when you're working with organizations, are there any sort of fundamental, tools that they have to have in place in order for it to be easy to implement co-workers?
Speaker 3You know, a lot of this comes down to, you know, just good change management. And so, you know, it's important to have leaders who are excited by this. And, you know, so we often talk to chief executives and heads of business divisions, and they're often the people who know they need to do something quickly. And, maybe often have also been involved in proofs of concepts or pieces of work that may be going sideways and not generating much value. And so, what we say is, you know, with our particular approach, it's like a rifle shop. We go through a process with the customer to determine where the use cases will have most impact. And we then agree that. Then we move to design the AI co-worker and figure out exactly what they'll do and what they won't do. Then we build it. And then obviously do all the sort of testing and ultimately put it into production. And there's a lot of technical things that need to be done there to make sure that the AI co-worker can access the right data. That data may be sitting in a SQL server, or it could be sitting in a system of record. It could even be sitting in a PDF file that needs to be scanned or whatever. And so, there's quite a lot of thought that needs to be done about the minutiae of that workflow and the data treatment and the use of large language models, for example, for generative AI parts of the task. But that, you know, we handle all the technical things, but we make sure that we are in lockstep with the customer who is both the technical people who are helping us with issues like security and data access, but most importantly, the business owners and the people who will work with these AI co-workers, because they're the people who ultimately are responsible for working with the AI co-worker in order to get the outcome.
Speaker 1Yeah, that makes sense. I guess I was, I was actually, you alluded to the data piece, because, you know, there's obviously a lot of talk about how organizations need to have their data strategy sorted before they actually start to implement. Yeah. AI. And maybe that's changing. I mean, most companies
Speaker 3would, would resonate with some version of the fact that their IT stack is, you know, not perfect and has a lot of legacy components and that a lot of their data is hard to access, maybe has sort of low accuracy and low integrity. And so, no one ever starts from the perfect starting point with either the IT systems and, or, you know, the data. Yeah. Again, why we have a rifle shot approach, what we say is once we've defined the use case that we want to deliver, we then work backwards and say, well, what data do we need in order to solve that? And then we will build the APIs or use the optical character recognition or whatever is necessary in order to extract the data, ingest the data and use the data for the, for the analysis and the tasks that the AI co-worker would perform. So, you can get a result quickly with far from perfect systems and data. And obviously over time, all companies are spending time upgrading their systems and cleaning their data and making it more accessible. But we're very strongly of the view that you don't need to wait for perfection before you start to create these business outcomes.
Speaker 2Could you hazard a guess, you talked about, it made it seem all very frictionless and that, you know, not that long before things get, the co-workers get into full production. How many digital co-workers, very roughly, are fully in production versus being in that test or parallel process, which I imagine is also quite important for some time? Is the world already infiltrated with thousands of digital co-workers or is it hundreds?
Speaker 3Let's be careful with the nomenclature here, because when we talk about an AI co-worker, it's an agentic system. So, sitting underneath that will be maybe one or two hundred agents, meaning the pieces of software that will do one specific task. And so, the word agent has become a little bit abused. And what I tried to describe earlier is that our AI co-workers are at the complex end of the spectrum. And so, they are agentic systems, if you like, with maybe a couple of hundred sort of tasks. And so, they are agentic systems, if you like, with maybe a couple of hundred tasks. And so, they are agentic systems, if you like, with maybe a couple of hundred and that's the thing that takes the building and the orchestration and the customization. So, you know, we're building now, I've ordered hundreds of these AI co-workers. If you go out into the broader market, I'm sure there are.
Speaker 2Oh, yeah. No, it was more about the co-workers.
Speaker 3Exactly. Many, many very simple agents.
Speaker 2Exactly. Yeah. It's just interesting to get a sort of a sense of scale and everything. And for listeners, I think you really touched on a key part there. And I'll just paraphrase it. Feel free to correct me if I get it wrong. But, yeah, the digital co-workers you're referring to that you build are basically a composite of agents that do, for want of a better phrase, a micro task or a very specific task. And how they are sewn or stitched together to do this customized workflow is what you're building there.
Speaker 3Yeah. And it's always easier to use an example rather than to use these words, which sometimes mean different things to different people. So let's imagine we have a task in a business, which is, let's say, rostering for a call center. That's a very tough problem because, first of all, you have to figure out what will be the demand of your call center in terms of the number of incoming calls on Monday morning. And so you can use AI to help you with that demand analysis and to figure out, is it winter? Is it summer? Is it a Monday or a Friday? Have you just launched a new product? Have you got a problem, et cetera, et cetera? And so there's some analysis that can be done around the demand side. Then you have the whole supply side, which is, you know, you may have a lot of agents who work in the call center, but they don't all work nine to five Monday to Friday. And so there's a job to be done to communicate with what may be hundreds, even thousands of people in order to figure out, you know, who's going to come to work on Monday morning. And so in a similar environment, I'm disguising it a little bit, but in a similar environment, we've built an AI coworker that handles that demand side and then an AI coworker that handles the interactions with all of the customer service agents. And so in that particular case, the second AI coworker basically lives on WhatsApp and will then have thousands of conversations with customer service agents. And say, hey, you know, I'm not well today, or my child is sick. I need to take them to the doctor. But I'd love to be doing some more work, and I'm happy to work later shifts because for whatever reason, we're going on holiday. And so all of that amount of data, which it would be pretty much impossible for humans to gather, that amount of data then is pulled in and it's compared with the demand and then optimized so you get the right number of people who want to be able to do that. I like giving examples because it brings to life that you can actually have those win-win-wins, whereas sometimes people think, well, if the AI wins, a human loses.
Speaker 2I think that's right. And I think there's some legitimacy for some people to feel that. And there's certainly a lot of doom and gloom. So I guess that brings me back to you sort of said it's very, very, very seldom. Do you encounter sort of the fear of people having their jobs? But it feels like when we talk to others and when you read research that in reality, there are a lot of entry-level jobs that appear to be headed for extinction in the next few years in particular. And presumably, you're seeing that as well in the work that you're doing.
Speaker 3We're not really. But if I step back, there's no doubt about it. So I think the treatment of this issue of job losses, which is super important, it tends to be a little blank. And the reality is it's very nuanced. So I agree with you that there will be job losses. And in fact, you know, there's data in the U.S. today that points to the fact that, you know, junior positions in call centers and junior positions in software development are already under threat. So we can see that happening. And it's obvious as well that there are job creation opportunities, you know, to build agents, to work on the systems and the sensors. And the orchestration of the sorts of systems that I've described and the many, many applications in the different functions and the different sectors. So but of course, whenever there's job creation and job losses, they never occur at the same time and in the same location and for the same job groups and people of the same age. And so this is one of the, you know, age-old issues of technology disruption in that there is always some destruction and there's always some creation. There's no doubt from what I have seen that there are specific opportunities for people in companies to shed some of that repetitive work and to evolve to more value-added thinking, intuitive work, including supervision of the sorts of AI workers that I've been describing. And so I think we're going to find a new equilibrium between what technology can do and what humans can do. And I think. I think there will also be lots of second-order effects. So, you know, we'll worry about job losses. But actually, you know, if AI reduces costs that they should do, then, you know, a lot of the products that are being produced are going to be cheaper for consumers. And so, you know, this is going to be hard to see through. But I don't think you should stand back and say, you know, this is going to be a bad idea. It will have some challenges along the way for some job groups. There's no doubt.
Speaker 2Yeah, I think that's exactly right. A final question before perhaps we do step back a bit and look more broadly. How do you ensure that your digital co-workers, to use a bit of a dramatic phrase, and then I'll tone it down, don't go rogue? How do you monitor them over time and make sure they haven't blackmailed another agent co-worker to give them access to some special data that if they combine that with their data access would, yeah, make them billionaires overnight?
Speaker 3No, and again, these are great questions over the dinner table for sure. You know, I've become aware of a paradox, which is one of the obvious things that's easy to throw at large language models, and quite reasonably, is that they are a black box. When you understand how they work, it's huge amounts of manipulation of matrices and repetition and interactions. And probability weightings and, you know, by the time they come to an answer, it's very hard to interrogate them. And therefore, there is that very justifiable fear of hallucinations and, you know, large language models coming up with the wrong answer in a very plausible way. The interesting thing about the AI co-workers that I've been describing to you is that they absolutely can be interrogated, and they are very – more so than humans. They have an audit trail, and you can understand their reasoning. And that's because of the fact that they are – they've been designed, and they're not completely – they're not just about the workings of the large language model. So if I give you an example, there's an AI co-worker that is involved in making assessments of the wonderful thing we have in Australia, FBT, food and beverage tax. And it's a – It's a complicated area, and this AI co-worker assesses the particular issue and comes up with a judgment of what the tax should be. And it does it in quite a complicated way that I won't go into. And in the design, we can make sure that that AI co-worker either reports or throws up edge cases and says, you know, a human needs to look at this. And this – The human supervisor was telling me that they can interrogate any of those decisions and understand very transparently how those decisions were made in a way that's actually quite hard to do with other humans. And so there's this paradox that when you build the complexity that I'd be describing, you can then manage the risks very explicitly in terms of what you want to see, audit trail you need, what data you need.
Speaker 1And I think that's incredibly important. you say, what's really interesting is that this is very hard when you have humans doing the work to be able to audit these kinds of decision making. So in some ways, it actually makes it less risky. Not in some ways.
Speaker 3I mean, that is the direct feedback we're getting. And that's often the point. I mean, we're working with financial institutions for whom risk management and regulatory compliance is their bread and butter. So it's built into our DNA and therefore built into the design of our AI co-workers.
Speaker 1Yeah, really important. So one of the things I'm interested in is you were talking about how you were talking to a supervisor of a co-worker. She or he was telling you about how they were able to really, you know, look into the audit trail. Are you finding that supervisors, managers, leaders are needing different skills when they're managing co-workers versus they're managing the humans? You know, I'm sure they are different skills, but what are they and what are you seeing?
Speaker 3No, that's a really good question. I mean, in a way, the answer to that is not really. Meaning that managing an AI co-worker is much more like managing a human than it is like managing an Excel spreadsheet or a tool. And that's why we've given them personas and names. Really, when people have been working with them for some time, they genuinely do treat them like a worker that they never see, potentially a worker in a different country. But you interact with them by email or by WhatsApp or by voice. And, you know, they have personas. They obviously have things that mark them out as being software and that they can work 24-7. And I think that's a really good point. And they have an incredible work rate. But, you know, one of the customers we work with, when they do an organization chart, they put humans and AI co-workers on the same chart, but just in different colors. So you might have someone, you know, a human being that's in charge of financial reconciliation, but that actually has, you know, two AI co-workers working for them.
Speaker 1Yeah. I mean, I would have thought, though, that when you're interacting with a co-worker versus a human, you're interacting with a human being. You would probably emphasize different things. And, you know, you wouldn't need to perhaps have the empathy and the, you know, all of the human side of being a leader that you would need with
Speaker 3the co-worker. It's an interesting issue. I know people who, for example, swear by the fact that if you're more polite to a large language model, you'll get a better answer. Or if they, in the case of a friend of mine, someone who has asked the AI co-worker to treat them very harshly and not, and not praise them and say, what a great question that was. And so, so there is a little bit of that. I think we're at an early stage in this and how it's going to evolve is going to be very interesting, but I suspect the bottom line will be these AI co-workers are doing very valuable jobs. And therefore probably the right culture is to treat them with the respect that you would treat any other worker. As opposed to, you know, the way you'd treat a, I don't know, a piece of hardware or a spreadsheet, which, you know, has no personality. I think the persona is very important here because they're adding value and they're part of the team.
Speaker 1Interesting. I think we can have a whole conversation about this, Patrick.
Speaker 3That's for debate. I think, you know, honestly, we don't know how that's going to evolve. There's a lot of talk about how operating models in companies evolve as there are more agents. And in the case of Future Secure, more AI co-workers and what the limitations are. And obviously the technology is evolving very rapidly. And so that frontier between what an agent can do and what humans can do, I'm sure will change. And that's not a bad thing, I think, as a human, because that challenges me and us more to add more value and go where, you know, software is less able to cope.
Speaker 2Listeners who are in organizations who are maybe mid-sized, you know, so maybe just don't have the bandwidth or the budgets right now, but want to be as AI and agentically ready as possible. What are the markers that make a great use case in terms of these more sophisticated co-workers, would you say? How do they start to think about that?
Speaker 3It's a great question. And to be honest, our co-workers are as, deployable in relatively small organizations as they are in huge ones. So I mentioned earlier that I do work in companies that do social impact. And some of those are not for profits. And some of them have, you know, pretty tight budgets. The good news about them is that you can design them and deploy them firstly quickly. And secondly, it's not a particularly capital intensive endeavor when you think about the large amounts of capital required for big systems and big platforms. And so it is that rifle shot specificity about solving a particular use case. And so medium-sized companies,
Speaker 2absolutely fine. So how does the leader in the medium-sized company, I know if you were, if they were your client, you help them identify a use case, but how do you build that muscle for ideal use case identification?
Speaker 3Yeah. And it's a muscle that needs to be in any company that's going to deploy any digital technology and certainly AI. So you go through a piece of analysis that says, where is the use of either predictive or generative AI going to have the biggest impact in the business financially and strategically? So, you know, it might be a telco that says, we really want to transform the customer experience. And therefore, we're going to focus on that. It may be an airline that's going to be able to do that. It says, we're really going to focus on operational improvements and we can get a lot more mileage out of that.
Speaker 2But are there any specific characteristics of, or maybe you're about to kind of drill down from that very, very high level kind of category?
Speaker 3So the answer is a function of the sector. So I gave you two real examples. And typically speaking, there's more value in all of the main customer processes, if you like, from, you know, from sales and marketing through to customer acquisition and then customer service and payments relative to the support functions like HR and finance. But having said that, it's about 70, 30, 60, 40, depending on which sector. And therefore, you know, there's still 40% of the value in the support group. So I guess my point is, there are definitely higher return places to go. But the reality is that, you know, AI, co-workers are going to find them, find their ways into most parts of organizations. And so, you know, we have a conversation with our customers in order to say, where do you think the value is or where do you know the value is? We then have our own experience from the sorts of thematics where we have been producing these AI co-workers. So all sorts of things to do with financial reconciliation is another area. In companies that deal with lots of humans, the onboarding and bringing new workers up to speed is another area. You know, dealing with contracts and matching contracts with tasks that needed to be done in order to execute the contract is another complicated area that tends to be inefficient with just humans running it. And so we bring those thematics to the party, overlap with either strategic or financial considerations that our customers have done. And we can get pretty quickly to a good place to start. You know, we are always looking for high return on investment that can be built relatively quickly so you get the return quickly. Our teams are really good at that.
Speaker 2Yeah. And the business model, just briefly, you know, is it sort of a client rents a digital worker off you or they own it?
Speaker 3You know, we're not consultants, but we're not just people who are selling software and throwing it over the fence. It's a customized build. And therefore, there's quite a partnership involved in that journey from choose the use case, design it, build it, launch it. And basically, what we do for that is it's a subscription business. And so we charge to build and then to run those AI
Speaker 2co-workers. Yeah, fantastic. You were at BCG, the management consulting firm for 20 plus years, working a lot in. Digital transformation and technology facing client problems for all those years. Just stepping back and using that kind of experience and journey that you've been on is where we all collectively find ourselves today in terms of AI and the emerging agentic and the like technologies. Is this about the same as that kind of heyday of the late 90s? Yeah, it's about the same as that kind of heyday of the late 90s. came crashing down. Is this about the same degree of change on humanity? Because the web did end up being quite transformative, but it took some time. Or is this bigger? Or where do you see it?
Speaker 3Great question. I think the impact that AI is going to have will be bigger than the impact that, let's call it, you know, whether you want to call it internet 1.0, 2.0, had cumulatively. But there are some really strong analogs. So, for example, we will probably have a boom and a bust in the financial markets and maybe a few because human beings tend to over invest and therefore, you know, bubbles happen in the financial markets. But, you know, when all that was happening in the original dot-com era, something like e-retailing and e-commerce was monotonically increasing. And so, whether or not and whatever happens in the future, in the financial markets, the adoption of AI in both consumers and enterprises will grow monotonically. And there's no doubt about that. The speed at which it grows is an interesting one. And it's gated by, you know, some of the capabilities of large corporations and just how good they are at deploying these solutions. And the answer is, you know, that's pretty slow and slower than people would like. BCG did some research recently that showed that only 5% of companies are really getting at scale value out of AI. You know, 35% are definitely on the journey and the rest are either grappling with it and struggling or haven't begun to grapple yet. And so, and interestingly, it's not a function really of sector. It's more about there are some companies who really want to do this and are winning. And there are other companies that, you know, are just laggards. And so, you know, that's a really important dynamic in the answer to your question. And by the way, that happened in the dot-com era as well in the adoption of digital technology. So there are lots of analogs, but my personal conviction is that the impact, if we could stand back in 50 years and have a look at both technology disruptions, I think the impact of AI will be bigger.
Speaker 1I think we are probably on the same page. And that sort of leads me nicely to, you know, sort of wrapping up a bit and our last couple of questions, which is, you know, looking into the future, let's say, let's say three years time, because who knows what's going to happen any, I mean, even three years is just, you know, a very long way out in the world of AI. What most concerns you about AI and the impact?
Speaker 3So I give a lot of sort of discussions with executives. And I always say anyone who tells you that they can see the world three years out, don't trust them. And, you know, never ask and to start with the end state and work back, because anyone who tells you they know the end state, I wouldn't believe them, right? So with that caveat, what worries me? Look, I think there's obviously geopolitics that abound in the infrastructure parts of the AI value chain. And, you know, that is going to probably impede it, you know, being rolled out well and helping humanity. It frustrates me, but I think it's going to probably impede it, you know, being rolled out the speed with which the adoption is taking place, because I believe that in general, it can be a force for good. And therefore, I want to see it adopted faster. Like all things, there's a balance between learning and deploying and making a few mistakes, and then good regulation that makes sure that, you know, that consumers, in particular, are looked after and that, you know, we don't have biases, and hallucinations and jailbreaks and all of the things that we know are possible and that, you know, come under the rubric of, you know, AI that doesn't work well. Probably the thing that worries me the most would be the bad actor dimension, which is, this is a very powerful technology. Everything I've been talking about is about building it in order to help customers, the customer experience. You know, efficiency and risk management in companies. But of course, bad actors will use this for nefarious means. And so that whole area of cybersecurity, I mean, just using AI in order to enhance the threat vectors of cybersecurity is one specific example of that. And that is already worrying me and many other people who are close to that space.
Speaker 1Yeah, I couldn't agree with you more. And, you know, on the flip side, you said you're very excited about some of the things that AI can do. What's the thing that most excites you about? What do you think could be the most sort of groundbreaking that AI technology could help us with?
Speaker 3Yeah. So, I mean, I think there are some places where it really can be revolutionary. The obvious one is in drug discovery. And so what AI can do to speed up drug discovery and therefore the solution of many of these problems. Yeah. So, I mean, I think that's one of the things that AI can do. I think that's one of the things that AI can do to speed up drug discovery and therefore the solution of many of today's, you know, awful illnesses has got to be the biggest one. Not far behind that would be AI's part in helping with energy transformation. So as we move to a more renewable sort of energy system, that places all sorts of incredible demands on transmission and distribution and generation. And generation becomes more distributed and therefore, you know, the whole system design needs to be revamped. And AI's role in that is to be able to do that. And so I think that's one of the things that I think is going to be fundamental.
Speaker 1Yeah. And how we use AI because it obviously consumes a lot of energy too.
Speaker 3Well, there is that. There is that. So it's a complicated world. I mean, I also think that it'll have a profound effect in the fullness of time in autonomous driving. But, you know, that one will probably take a little longer than people think. But, you know, those are three areas where it's fundamentally disruptive.
Speaker 1Yeah, absolutely. Absolutely. Well, it's been so fantastic to talk to you today, Patrick. It's been really fascinating to understand the future secure model and to understand, you know, how co-workers are already being utilised in some organisations. Now, if our listeners were interested to learn more about future secure or indeed yourself, where would they go to find out more?
Speaker 3We have a website. So more than welcome to go to that. And I'm easily findable on LinkedIn.
Speaker 1Fantastic. Well, we'll make sure they're in the show notes. Well, thank you so much again. It's really been a pleasure. And we wish you all the success with future secure.
Speaker 3Great. Thank you for your time.
PodcastTranscript AI