1065. Insights: Is agentic AI the future of investing?

0:14Welcome to Fintech Insider Insights from 11FS. I'm Benjamin Ensor and today we're unpacking a big question. What happens when artificial intelligence starts making investment decisions for us? Investing has always been about judgment. It's about deciding what matters, when to act, and how much risk to take. But that starts to change as we move from a world where technology can help us analyze markets to one where it can act within them. Not just suggesting what to invest in, but deciding what to buy, what to sell, and when to do it. And that starts to create a whole different set of challenges.

0:52Because unlike payments, where an outcome is immediate, investing plays out over time. decisions compound, mistakes aren't always obvious, and accountability gets a lot less clear. So this isn't just about automation, it's about delegating judgment and uncertainty. So that's exactly what we're going to explore today. And I'm delighted that I'm being joined by a panel of three brilliant guests to explore this. So first up, we have Yannick Malik, co-founder and co-chief executive at public.com. Welcome back to the podcast. It's a pleasure to have you with us again. Can you remind our listeners a little bit about Public and a little bit about you?

1:32Yeah, sure. Hey, it's good to be back here. So, Public.com investing app in the U.S. building what we call the first real agentic brokerage, which lets you install AI agents directly within your portfolio in a safe, easy, transparent way to help you carry out anything from research, trade execution, building, implementing strategies, portfolio rebalancing, cash management, monitoring of markets, risk, and much, much more. And so spent, I would say, the first, we launched in 2019, and so we spent sort of the first few years building this, like, multi-assets infrastructure of bonds and equities and crypto and a bunch of other asset classes.

2:24retirement accounts, brokerage accounts. And then I think since 2023, we've been quite deep in building AI, this agentic brokerage model, basically AI within your portfolio that can, starting with research, and then we've recently launched the ability for agents to help with trade execution as well. Which is super, super exciting. I'm also delighted to be joined by Megan K. Wood, founder and chief executive at K. Wood. Megan, welcome back. It would be fantastic if you could also tell our listeners a little bit about you and a little bit about K-Wood, please. Yeah, absolutely. So I'm an American by background, but I've been in England for about a decade now, so working in banking and finance.

3:06So previously, I moved over to help start Starling Bank and then was over at Barclays for a bit working on consumer banking technology. And now K-Wood is an agentic AI firm that builds financial management software for banks and corporates. So, yeah, still very much in the banking, fintech space, but now focusing on AI. Very exciting to have you with us. I'm also delighted to welcome Peter Rossi, who is a technology and AI advisor to private equity and co-founder of Hubble Field. Welcome to the show, Peter. Could you also introduce yourself to our listeners, please? Great. Yeah, thanks. So, yeah, my name is Peter Rossi.

3:43I spent the last 20 plus years in tech, so designing, building and scaling platforms across quite a lot of verticals. Started my career with McLaren in Formula One, traveling with the world with them. I spent 10 years in venture capital-backed tech startups and most recently spent four years in a P8P-backed buy and build in the certification space after selling my SaaS business to them. Now I spend my time split between advising P and VC funds in what's real in AI versus hype, helping their port caves as well. The thing I'm most excited about right now is Hubble's build. It's an Asian native discovery platform for private markets that the team are working on at launching the end of the month.

4:28It makes me slightly nervous in this conversation today because I've got skin in the game about it. So there you go. That's exciting. Okay, well, I love what you said about trying to distinguish the sort of hype from the reality. So why don't we start with the basics? Megan, I think I might come to you first. We've had quantitative investment strategies. We've had machine-driven investing for decades. What's new and different about bringing AI and specifically agentic AI into investments? What does that enable us to do that perhaps we couldn't do before? Yeah, I think it's a great question because to your point, we have had AI and machine learning and banking for a long time.

5:17I think originally, if we were to rewind 10, 15 years ago, it first started coming out really with machine learning applied at transaction monitoring, so to combat fraud. And that was actually very highly effective. I think if we then kind of speed up a little bit, fast forward five years, then we would have the advent of robo-advisory. And that was quite interesting for algorithmic processes underpinning training. I think now if we look at what has shifted, we went from AI that was predominantly machine learning to Gen AI, which is the brilliance of LLMs. And now agentic AI is the execution layer on top of that.

5:52So agentic investing is not just automation. It's intent translation plus continuous action. And that's really what is new. So automation executes a known instruction, whereas agency interprets it, chooses among other paths, and acts over time. And so what we have now is not just the brilliance that we previously had before it, but the ability to enable it to act and deploy and to make decisions, to have nuance and to play a role it didn't previously have, I would say, is the most recent development. them. Indeed. And we started to see a number of sort of firms starting to use agentic AI in various ways.

6:30We've seen firms like Citi building internal AI agents to support wealth managers, taking on tasks like portfolio analysis, market research. But then we've also seen firms like Moomoo letting customers connect their own AI agents directly to trading infrastructure. And of course, Gannick and your whole team at Public introducing these AI-powered portfolio agents sort of directly into the investing experience. Janik, can you tell us a little bit about what you see as the sort of the opportunity? What made you at Publix say, hey, this is something we absolutely want to get behind and offer to our customers?

7:08Yeah, I think in Mega's sort of framework, I think all the machine learning that have been used to sort of combat fraud and these kinds of things haven't been things that really dramatically improved the user experience. They've been sort of like behind the scenes, sort of lower down in the income statement also. They haven't necessarily driven in a very directly dribbled way, a lot of top of funnel growth, but they've helped made, you know, financial technology firms and banks and such way more efficient. And then I think the move to LLMs, and I would agree specifically the shift to reasoning models have really been where you can sort of rebuild your entire user experience around that from a consumer perspective.

7:54And so we, just to give the background, we started in 2023 with like being the first to sort of embed research directly into the investing app experience. So you're on the Apple stock page, you swipe down, you can ask any question you want. It can be like, hey, how many iPhones were shipped this quarter, this, that, and the other. It goes through all the statements. And so that became just like a several, what a magnitude faster way to do research. It also gave us a very interesting viewpoint into what people are actively thinking about, which we necessarily didn't always know when they were just hitting the buy and sell buttons prior to that.

8:25And now it's sort of evolved to having this agentic layer that Megan also refers to, where you are really moving more from placing orders to instructing intent. So, you know, the financial markets have always ultimately come down to you place a market or a limit order at the end of the day. And now the user experience is changing to be one where you can be like, hey, you know, if NVIDIA is down 5 % on the week or the stock becomes cheaper on a PE multiples basis, after an earnings call, start adding to my position. And so you're more instructing agents with these like general objectives, which creates this abstraction layer between the grunt work, if you will, of actually placing orders, moving money around, you know, hey, every time I got money, my checking account, earning zero, move it to my bond portfolio.

9:16or like we've already seen hundreds of thousands of examples of how people sort of help automate a lot of workflows in their portfolio, how they sort of get to a setup where they're just more sophisticated, I suppose, and a little bit more intelligent about their investing. I always think for the everyday investor, you know, people have asked me, by the way, for years and years and years, like, oh, what does it take to be like a great investor? I was like, listen, the same if you want to be in the Champions League or the NBA, you got to spend the time, right? Be the first in the gym and the last guy out.

9:48That's now actually kind of changing. Now you actually can really get a lot of time back. And so I think for everyday investors, especially, which obviously is most of us that don't, aren't day traders sitting in front of a screen 24 seven, you miss a lot of moments in the market. There might be weeks where you're not really following what's happening. The fact that you can instruct agents to do a lot of this stuff on your behalf, I think is a huge game changer to that experience. and it even goes further down into like, you know, we're thinking about things like, okay, you know, as a product consumer company, you look at monthly active users, right?

10:22That's like a pretty standard thing. Is that really the metric to go for anymore? I don't know. So it changes a lot of like things strategically, even how you build product. And that's why I think that move to reasoning models and LM specifically was such a big shift versus just, if you will, a lot of the ML improvements that were great for fraud and everything else, to Megan's point. Thank you. Peter, Yannick was obviously just talking about sort of everyday investors and sort of retail investors. And obviously, it's natural to look at this with a bit of a retail lens. But is this also relevant to institutional investors?

10:56I mean, you were talking in your introduction about private markets and so on. Do we think we're going to see a shift among institutional investors as well? Or do they maybe not need this? Is this an institutional play as well? Or is it just for retail investors? I think it's really an interesting point that Yannick made about the monthly active users thing because is it active users anymore or is it going to be active agents or actors in your platform or active tool calls on your platform? Because I think that whole sort of surface, you know, sort of the entrance to these businesses and platforms is going to change a lot.

11:31I think with regards to sort of who it's going to be and sort of how it might structurally shift, I think there's a lot going on in the market. And I think the thing actually wider than this is around sort of how AI is unleashing a lot of sort of sleeping capability out there that sort of people that might have ended up in careers that they haven't been able to sort of switch into something they might be good at. A really good example of that, a friend of mine is a maths teacher, and he's really very attuned to what's going on in AI. I've been doing a fair amount of mentoring with him. And he messaged me the other day basically saying that he built an algorithmic trading platform that was AI integrated using a parquet database for high speed data streaming off of bet trading markets.

12:21And he's basically, you know, he's a math teacher. Six months ago, he wouldn't have had a capability to do that. And now he's leveraging a stack that wouldn't be out of place with a quant working on an algo. So, yeah, I think it's going to change a lot. I think you're going to see some really interesting stuff going on out there. And some new people emerging doing some things. Yeah, I will say I've, somewhat ironically, by the way, my old company before public, which was actually a lot like called tradable, was sort of in this world of like API algorithmic trading stuff. And the biggest issue we had was like the people with the trading strategies could not code on our API.

12:59So it was such a small overlap. You're looking for like a little sliver of the market. And then now that Venn diagram just have become fully eclipsed. And I'll give you a wild stat, which is we launched our API public like six months ago probably. And we've been the first to do sort of MCP support and open claw scale and perplexity connectors and claw connectors and all that stuff. And now on a daily basis, 25 to 30 % of our volume is now running through API. In the last six months, our volumes overall on the platform have nearly tripled. We're talking tens of billions of dollars. And so it's sort of a little, it's by far the fastest growing part of our business right now.

13:48But basically just having that ability to just go to settings, grab your API key, and then go do stuff, right? Like it's a little bit less slogan. You can just build stuff. And I think that's just become so true for everyone now. And even in our little world of retail trading and investing, we've just seen that firsthand. I mean, the amount of people we have programmatically accessing their accounts now, like we ask what people's job titles are when they sign up. And so I know roughly how many software developers are engineers, and we're well beyond that. So there's a lot of proof that this is really working.

14:24That was actually going to be my next question to you, Megan, because, you know, some of the people that Peter and Yannick are talking about are, as you say, software engineers, math teachers, people with, you know, good head for numbers, good understanding of the technology. Does this start potentially moving into maybe more mainstream, more, I have to say more normal people because I don't want to offend all the, you know, the brilliant software engineers, you know, maths majors listening. But do you think over time this perhaps becomes more mainstream because not everyone's necessarily going to have the confidence to instruct an AI bot to say, well, if volatility in the market changes, do this, or if currencies move like that, hedge that, or whatever.

15:02Some people are just going to want to say, help me retire earlier, or help me afford a yacht, or whatever it is that their goal is. Yeah, I think you're making a really good point, which is it really gives more people confidence to do new things. And so I think on the one hand, it's financial decisions. I think to what Peter Nanak were saying as well, is it's across a whole host of things. It could be trying different skills, or it could be coding. like I was in San Francisco last month and I was like, you hear about vibe coding in London, but everyone's vibe coding in San Francisco. Like I was staying with a friend at a weekend and she just vibe coded something for fun.

15:35And I actually, I didn't use that term for it, but when we were starting KWID, I actually just vibe coded the V1 with the ChatGPT. And it's like, okay, put all these into these repo files in GitHub and I sent it to my CTO. I'm like, okay, try this out. What do you think? And I actually, similarly, even with our first V1 designs, I had ChatGPT write the requirements for it, hire the designer for it, pay the designer for it, give the designer feedback on it. It was a phenomenal experience. But I think effectively, kind of through my experience testing it and what I'm seeing out in the world, is generally speaking, bringing agentic AI on top of the layer of LLNs is fantastic because it's bringing knowledge to bring confidence to a mass of people to help them to do things and think through things that they didn't know how to do before.

16:18I think with financial investing and agentic investing, The reality is that customers don't need to either be very technically adept. They don't need to understand model weights. They don't understand each nuance of the technicality. But they need to understand the mandate. They need to be able to have a conversation with the agentic program that they're using to be able to set the parameters of what the agent can do, know what the agent can do, when it can do it, how to stop it, be able to have a human read journal to, you know, at any given time, understand what's taken place, understand the reversibility of those actions or lack thereof.

16:51That's what people need to understand is just really the capability of the agent. But I will say, I think over time, we've had a lot of conversations on, you know, prompt engineering. You know, you have to be good at writing prompts to interact with agents. And I think that's actually becoming less necessary now because LLMs, at least the major ones, GPT, Claw, Gemini, Grok, and others are so good at understanding nuance, understanding intent,

17:20emotionally relating to users to understand what they're trying to achieve on multiple levels. So it's not just buy a yacht. It might be kind of something else around this kind of goal of financial freedom and a timeline to kind of understanding the fullness of what they're trying to achieve. But then also in moments of crisis, I think what's really interesting is if we were to just rewind us a few years ago when I was at Barclays, the way that it would be kind of presented about is like AI might be able to help you make a brilliant financial plan. It might be able to help you execute that plan and like understand changes and flux and respond to the market even more adept than human advisors can.

17:53But we really hypothesize that it wouldn't be able to help you with fears. And when the market crashes or something goes wrong, that kind of handholding that a human advisor could do would be difficult to replace. But I think what we're increasingly seeing is that the LLMs and agents are also able to have that kind of connection as well, to be able to be available 24-7 and to be able to communicate in an effective way to help not only answer people's questions, but to kind of solve some of their anxieties and stressors as well. So I think now the brilliance of, you know, LLMs combined, particularly with agency, is this enhanced ability for more people to have confidence with investing, but to do all sorts of different things that they didn't do before in a supported way.

18:35So, yeah, I think it's a good question. That's a really, really interesting perspective that AI is starting to be able to understand or at least know how to respond to human psychology. I'm immediately sort of thinking of the things that can go wrong. I mean, let's say, you know, we've got examples of people asking AI to draw them as a doctor and it goes and produces a picture of them as Jesus Christ. you know what happens if you're if people start giving ai instructions that maybe don't make sense i mean you know peter your your teacher friend's obviously very smart but we could imagine other people who might potentially instruct ai to do something that actually probably wasn't smart you know i mean for example might they might say you know put all of my cash into you know a particular stock which obviously just leaves them completely undiversified which generally speaking isn't going to be a great strategy it might be if the stock happens to be apple 10 years ago or whatever but it's often not going to be a great strategy do you think we're going to need to sort of build guardrails around this to protect people either from themselves or from ai doing something slightly different from what they perhaps intended well Well, the way that I tend to sort of tackle this conversation is AI is a tool.

19:59So if you use the analogy that power saws are brilliant pieces of kit that you can use if you're building a house and you need to be able to cut through lots of wood. But also it can chop your hand off if you don't know what you're doing with it and you don't appreciate what it's doing. But as a result of that, we don't ban power saws. right we um people can still buy them and people still make mistakes and people still have to go to hospital and bad things happen but majority of people can use them safely um and i think there's definitely a learning curve element around that um with regards to the guardrails i think that's interesting i think like you know coming back to that point it's a raw tool right you know a very powerful saw that can cut through wood can you know can do lots of things and it can do a lot of damage the um you know typically you would want to use that in the confines of you know building a house with a team of people that build a house and it's it's similar in sort of like the the software development world i think where things are changing and with a lot of the um sort of tech teams i'm working with across all the pe funds is that you you're no longer hiring a development team to build a product you're hiring a development team to build the factory that builds the product.

21:15So that's where all of the time and effort's going into building those guardrails and systems around how it's supposed to work, what does good look like, what's a safe thing to do, what not to do, how I want you to check the things. So that when you sort of release raw material to the floor, to use the sort of the manufacturing phrases, it goes through that factory and it comes out as a quality finished product on the other side, something you'd actually expect it to be. But if you've just got a load of people on the shop floor with a load of hand saws doing things, you know, bad things can happen.

21:47Yannick, I'd love to bring, thank you, Peter. That was really good. Yannick, I'd love to bring you on this because you must have thought quite hard about this. And from what I've seen at public, you've put quite a lot of thought into the sort of guardrails and the consent gates and so on to make sure that your customers are doing things knowingly. Have you seen any customers doing things that were potentially bad for them? Have you sort of had to sort of rethink a little bit your guardrails over the past sort of six months? Yeah, you're right. We definitely have this like brand of trying to be very fiduciary, very responsible, especially maybe not by European standards, but certainly by American standards, I would say.

22:29We've actually seen the opposite. We've seen that people are way better at risk management when they use agents versus when they don't. And I think that's one key thing to think about.

22:42you know, is it that agents will never make a mistake? No, but it's a little bit like the self-driving car movement. We're not coming from a world where no one makes mistakes. We're coming from a world, actually, where resell investors, on average, probably aren't doing risk management as well as they should. And so I'll give you a very concrete example. I tried to set up a strategy that was a Martingale strategy, right? And so for the reader, if you don't know, it's like double if you're in trouble. It's kind of the casual way of saying it, right? So if you really, really believe in a stock, you think it can only go up.

23:18Every time it goes down, you'll just double your position. And then in doing so, you sort of bring your average cost basis down. Okay, so I tried to do that. And it checked me three times. Do you want to stop sort of a maximum stop out here on the strategy? I said, no. It's like, are you sure? Mangeel strategies can run really wild. Like, here is like a scenario that you can end up with." And I said, no, again, it's like, okay, Yannick, I really think you should do this. You know, I see you already have drawn a little bit of your margin line of credit. I really think this is a bad decision. I implore you to reconsider.

23:53And then I cave. But basically, if you take that level of experience versus just having a digital platform, whether it's a brokerage or a robo-advisor, whatever it is, where you don't actually have the ability to even put up these dynamic, intelligent, personalized, just-in-time, in-the-moment guardrails. That does not really exist in any product experience. And so the way that you can nudge people to do risk management has completely changed. Because in the past, the best thing you could do is have a couple of pop-ups come up. It's like, hey, have you read all the disclaimers? Do you realize that this company is X, Y, and Z?

24:36And yeah, let's face it, sometimes people just like hit next, next, next, next, next. But in a chat conversation interface, you kind of read what's coming back, especially if you have to answer it before it helps you execute it, which is the case here. And so we've seen that, yeah, from a product strategy perspective, this completely changes how you can do risk management. And we don't have to live as a self-directed brokerage. We don't have to live in this world anymore where we're either being paternalistic, overprotective, or not having any guardrails at all. And it's just the Wild West. Now, this is a perfect, happy medium where we can just basically harness the models to very heavily over-index and risk management.

25:20But if you really do understand all the risk and you say yes that third time, even after you've been laid out all these different scenarios that could go wrong, yeah, then it's your own risk. And then you're happy to underwrite that risk. and that's something that you go in to eyes wide open. And so the product strategy is incredibly happy with this movement because it just completely changes how you can sort of manage these types of scenarios. That is such an interesting point. Can I add on to that actually? Just a few thoughts on guard routes because I think it's such an interesting conversation because I agree with what you both are saying.

25:56I think the nuance that's really interesting now is how control becomes layered where the investor sets the goals and the mandates. The agent then operates within the mandate. And then the platform enforces the guardrails. But if those layers aren't auditable, then control is more theoretical than real. But what we need to separate out now, I think, is that the problem isn't that an AI might lose money because markets lose money. So the problem is not that, but it's when no one can understand whether the AI lost money when it was doing what it was asked to do fundamentally. Exactly. And that's the real opportunity with guardrails, I think.

26:30Oh, totally. Fantastic. Let's take a quick break here and then we'll come back to explore that point just after the break. So we're going to leave you on a cliffhanger while we have a quick break and we will be back very shortly.