Podcast: Are investors underestimating AI?

Podcast transcript

John Przygocki: Welcome to Talking Markets with Franklin Templeton. I'm your host, John Przygocki, director in our global marketing organization. Today I am joined in the studio by Jonathan Curtis, Chief Investment Officer of the Franklin Equity Group. The Franklin Equity Group manages a wide range of equity strategies across both public and private markets, all grounded in the belief that sustained innovation is the key to long-term value creation. Jonathan has over 30 years of experience at the intersection of technology and investing. He started his career as a software and systems engineer, transitioning later into asset management. That background gives him a unique edge in understanding emerging technologies like artificial intelligence, as well as helping him uncover innovation wherever it lies.

Jonathan, welcome to the show.

Jonathan Curtis: Excellent. Thank you, John.

John Przygocki: I really have been looking forward to our conversation today, focused on the next big breakthroughs in artificial intelligence [AI]. So my first question for you is tell us why you're so excited about AI and in particular, generative AI.

Jonathan Curtis: Yeah, well, we think that it is simply a profound opportunity and quite frankly, much bigger than investors appreciate. But at its core, we've really figured out how the brain works, how intelligence works. We don’t have a perfect model for it, but we're clearly in the right zip code for how it works. And when you layer on then something like a chat model, much like ChatGPT has done, we've made it accessible to anyone. So what's so exciting is that it's easy to use and it's essentially intelligence at your keyboard. So what's exciting is that we are really going to start on a very exciting path for meaningfully increasing the supply of innovation in the world. We think it is about to explode.

Now, I think investors get that there's a revenue opportunity in the tech space. That's probably priced in into public markets. What's not priced well in the public markets is all of the efficiency gains that the technology sector is going to get out of artificial intelligence. They are going to be some of the first appliers of artificial intelligence to their business across software development, across marketing, how they run their finance operations. So that's not captured in markets and in valuations.

And then, third, I do not think the broader market is capturing the efficiency gains and the surge in innovation that's going to come as this is applied outside of the technology space. So we think it's profound. We think anybody can use it. And we do not think it is accurately captured in markets. As an investor, all of that says there's something exciting going on.

John Przygocki: That makes a lot of sense. A couple follow-ups there. Can you explain the concept of agentic AI? And then also you mentioned efficiency and productivity. What do you see as potential impacts on business operations and productivity? Maybe in general?

Jonathan Curtis: Well first let's talk with agentic AI. So, probably many of your listeners are familiar with using one of these large language models, a ChatGPT, a Google Gemini. There's many of them out there. They understand how to talk with them and, you know, get a poem out of them or summarize a document or maybe get help in writing an email, or even writing a resume.

In that model, the human being is acting as the agent. They're engaging with the model, and they're driving the model towards some outcome. But agentic AI is really interesting, because like a human being or some animal with intelligence, agentic AI will be able to pursue a goal and navigate the complexities of pursuing that goal. So, intelligence came as a result of ultimately wanting life to be able to pursue a goal and then navigate all of the challenges that the real world throws at it. So, these agents will be able to pursue a goal and navigate curveballs, challenges that come up in the pursuit of that goal and ultimately work around it.

Now, let's bring this a little more to the real world. One of the first killer applications for these large language models (this generative AI idea that you've heard about, ChatGPT) has been in the area of software development. These models are very, very good at writing software and helping software developers be better at writing software. So software developers can literally go into these models, give them a prompt and say, “create an application that does X, Y, or Z,” or “review this section of code. There's a bug in it here somewhere and I can't figure it out.”

And these models then go into effectively agentic mode and work on the task and solve the problem. And you can sometimes see them back up and reverse course and change direction. In that case, they're acting like an agent. They're driving towards some particular goal that has been outlined in the prompt, and then they're working around problems and solving them. And so they're doing what life does. They're doing what intelligent life does. So that's what an agent is.

Now let's answer your question about productivity. And we'll go back to a point I made in my first answer. One of the primary cost inputs in the technology sector is software developers. They're a key part of realizing the digital world, and they're a very expensive part of realizing the digital world.

Some of the leaders of our world's largest technology companies are saying now that they're getting 25% to 30% of the code that their engineers are writing directly out of these models. A large, social networking company, the leader of that said that within the next year, he expects that these models will be as good as any mid-level engineer.

So, it's not about replacing so much those engineers, but making them much more productive. Getting many more lines of code, getting many more applications, getting many more bug fixes out per engineer in one of the highest cost areas of these tech companies is going to be a margin driver. We believe that what's going to happen is these same types of capabilities are going to find their way to all forms of knowledge work, and in fact, we think it's already happening.

And so we're going to see more output per human being and higher margins for companies that are early adopters of this type of technology. So it's going to directly impact both the supply of innovation and ultimately the bottom line of really every company in the world that is staffed heavily with knowledge workers.

John Przygocki: Very interesting. It brings up another question here around physical AI. What are some of the examples of physical AI, and how do you see those potentially recreating or maybe reimagining existing industries or potentially creating entirely new markets?

Jonathan: Yeah. Investors kind of are getting their arms around the generative AI opportunity and how it impacts knowledge workers. I think the physical AI opportunity is further out still, and investors are in the very early stages of figuring that out, as are we. But ultimately, we've known about robots for years. The robots that we've had helping us on our manufacturing lines or vacuuming our homes or delivering mail around on carts and in our office environments, they've all been pretty dumb, right?

They haven't been able to navigate the complex world and come up with the plans to solve a particular goal. But like the straw man from The Wizard of Oz who only wanted a brain, these robots have only wanted a brain, and we now are figuring out how to put brains, if you will, AI models into these robots. So, what that ultimately means is they are going to become agentic, right?

So that goes back to your second question. Now, we already have real world examples of agentic AI robots in our midst. Anybody who owns a Tesla and owns full self-driving or has been to San Francisco and increasingly other parts of the United States and taken a Waymo, they've seen what the future is in transportation and in taxi services. It is agentic vehicles. We're going to start seeing similar types of things in manufacturing. So those dumb robots in a manufacturing line, let's say helping to build a car, will get a brain in them. The vacuum cleaners that we have in our homes will get brains in them. And the full manifestation, the fully exciting opportunity here is ultimately humanoid robots that will be able to go out into the world and interact with the world that has been built for humans and act in an agentic manner.

And that's exciting, because of course, there's all sorts of menial work that I'd love to have a robot do for me: mow my lawn, paint my house, deliver my groceries. But there's also dangerous work that I'd love to have them do for me: go into that nuclear power plant and stop the meltdown, go into a fire and find the source of the fire and deal with that, go into a war setting and fight an enemy. So that opens up some of the opportunity around physical AI.

You know, there are a billion knowledge workers on the planet, 8 billion people. There are many, many more physical laborers in the world. And so the opportunity to augment physical labor with humanoid robots or other specially designed robots that are able to navigate the complex world is an enormous opportunity—multi, multi, trillions of dollars, we believe.

John Przygocki: That's really interesting. I'm actually thinking about what I saw earlier this afternoon. So, in the northeast we're experiencing another heat wave. We've had two here in the last couple of weeks. It must be 95 outside. And one of my neighbors is currently having their roof replaced. So, there are a number of gentlemen up on that roof. And I'm sure that they would rather a robot—

Jonathan Curtis: Yeah, that is that is brutally difficult work and dangerous work. And it's probably the type of work that a robot could do eventually very, very well. So yeah, you've nailed it. It's exactly that type of thing.

John Przygocki: Very interesting. How about AI transforming the health care industry? We've heard a lot about this, but what are your thoughts here?

Jonathan Curtis: Yeah. Listen, the opportunities are endless. But at the core of health care is the word care. It is a doctor engaging with a patient and helping to heal them. And this I think gets at one of the things that I think people are mistaking about this opportunity around generative AI. Doctors go study medicine for, I don't know, 10, 15, 20 years, become doctors to be care providers and to be effective care providers, right? To come up with good solutions that ultimately heal their patients. Not to be somebody who is filling out forms, insurance forms, managing staff, doing their finances. We think that AI is going to do two things, really. It’s going to make the doctors better at coming up with solutions. And we're already seeing studies that show that human doctors, paired with AI models that know a lot about medicine, get to better outcomes for patients. But it will also free up the doctor's time to give them more time to lean into the humanity of their work.

And we know that caring for humans requires another human being. And being a good knowledge worker for that matter, particularly in the type of work we do here in finance and active money management requires a human being. And so I think these models and generative AI is going to free humans up to get out of the drudgery of their work and allow them to lean into the human parts of their work and ultimately allow them to transform their experience as workers. And in health care, I think that's going to result in higher levels of care and more care being provided, and less drudgery work for the doctors.

John Przygocki: Yeah, and I can imagine efficiency as well. I'm just thinking about any of the doctor's offices that I've been in recently where, to your point, there are more people doing the processing of the visit than time that you actually spend, you know, with the particular doctor. Very, very interesting. You mentioned our industry: finance. How about financial services and opportunities for reshaping and what we do?

Jonathan Curtis: Well, financial services is one of the biggest spenders on tech on the planet, maybe outside of the government. So they've always been early adopters and intense users of technology. And why? Because they're loaded with data. There's lots of repetitive processes. So the financial services industry broadly is going to be ripe for finding many, many use cases. In some of the world's largest banks, the leaders of those banks have been talking about finding thousands of use cases and very substantial budgets are being applied to it.

But let's bring it to a firm like Franklin Templeton, a place I know well. We do two things here, maybe three things here. One, we try and generate strong risk-adjusted returns—so have compelling products that serve our clients. We then want to communicate with our clients about the value proposition of what they own and, you know, give them updates on it. And then we want to go out, raise more assets so we can keep serving our clients and having a good business for our shareholders as well.

Across each one of those lanes, there is opportunity. In engaging with our clients, these models are going to help us communicate with our clients in more narrowcast types of way and much more quickly. They will help us be better salespeople.

But let's focus on the alpha generation part of the job. Very similar to a doctor, there's a lot of drudgery work in being an investment analyst. There's lots of documents to read, there are lots of models to be built, and there is value in some of that, the human beings doing some of that. In that struggle, there is insight. But there is much more insight in going out and engaging with the managers of these businesses that we invest in, seeing how they allocate human capital, understanding how their customers engage with their products, what their employees think about working at the companies, how these managers treat one another, and the people on their team. All of that requires human engagement, human touch. So I believe that in my world of active, fundamental equity investing, we will free our analysts up from some of the drudgery part of their job and enable them to spend more time at conferences, spend more time at dinners, spend more time doing the human part of the work, where only the insights that a human being can generate, but which are so essential to understanding value creation—well, we're already seeing it happening, and I think that's going to be a huge lever for my team.

John Przygocki: That's another really interesting example that kind of takes me back. So I joined this industry in 1995 and I could tell you, in the first 10 years of my career, exactly what you just talked about was what we were marketing, communicating and selling—the funds or the products that we were offering. It was all about what the portfolio managers and analysts were doing with the companies that they either owned or were looking at. And to be honest, I haven't heard much of that, you know. I'm not saying it doesn't occur, but it certainly is not part of the communication cycle.

Jonathan Curtis: Well, yeah. So I think that that's a very real part of how it'll make us more productive. Now, I couldn't do my job today if I didn't have some of the fancy, financial tools that I use every day. Bloomberg. FactSet. Excel. All those type of things. So I think these models will sit alongside those tools. None of those tools that I just mentioned generate alpha for me, and I don't anticipate these models will ever generate alpha for me.

That is still the job of the human being. But I couldn't be effective at my job if I didn't have access to these tools, and if I wasn't using them. They will become a central part of our work. Using them as buddies, using them to help us learn new topics, asking them, you know, “My portfolio is structured in a particular way. Here. Look at it. Make a recommendation for me on my next best trade.” But ultimately the human being will always be the agent in that, making those decisions. And what will active managers bring to that? The human context, the human insight, the stuff that will never be in these models.

John Przygocki: Earlier in our conversation, Jonathan, you mentioned the number of knowledge workers that there are in the world. How do you quantify the potential impact that these innovations can have on the knowledge worker, productivity and efficiency?

Jonathan Curtis: Well, first off, we were chatting earlier and I learned you have some kids that are still in high school or maybe just starting college. Some of your children are already in the professional world. I have children that are in—graduated from high school or just entering high school. And ask any one of them. They're using these tools all the time to learn more quickly, help them complete their homework. So, like, they're adding value. And many, many of the people in my organizations are already using these. So, I can just look around and see value being generated in my, both my personal life and in my work life as a direct result of these models.

But how do we size the opportunities is really your question. So there are a billion knowledge workers in the world, earning we estimate on average about $50,000 a year. So that's a $50 trillion knowledge worker services TAM [Total Addressable Market] effectively. So I gave you a stat earlier on how productive these models are making software developers, somewhere between 25 and 30%. So pick a number, right? How much more efficient do you think these models will make a knowledge worker? 5% more productive? 10% more productive? You can very quickly get to, you know—10% more productive for a knowledge worker, that's a $5 trillion opportunity. The builders of these models and the companies that are going to express these models and their existing products, they can charge against that, those productivity gains. And ultimately that is their TAM. So it's enormous. And that's just in knowledge work. That doesn't even get at the physical AI opportunity that we talked about earlier. But I think that's many, many times what we're going to see in knowledge work.

John Przygocki: So how about with that projected growth that you're highlighting (and we've talked about a couple of sectors in particular)—are there other sectors that you expect, to benefit more than the ones that we've highlighted here today?

Jonathan Curtis: Yeah. Well, certainly. And any sector where there's knowledge workers will benefit. And today in a knowledge-worker-driven economy, that's almost every single one of them. So they will all benefit the highly labor-intensive ones, the highly knowledge worker-intensive ones will benefit the most. Health care is heavily knowledge worker-intensive. We talked about that already. Software developer is knowledge worker-intensive.

But there's also transportation. We spoke earlier about full self-driving taxis. That is going to transform those markets in meaningful ways. There was a really interesting article in financial literature today talking about how one of the world's largest e-commerce companies is using robots increasingly, robots with brains in them, to operate their logistics platform and how the efficiency gains have really gone up. So the investor should be looking for knowledge worker-intensive economies, and/or at the transportation sector, but also education. We've already touched on that a little bit with the experiences our kids are definitely having using these models. So we're going to see it in many, many, many places. But that's a quick framework for where the biggest impacts are probably going to be.

John Przygocki: Okay. That makes sense. How about investors? You mentioned earlier that investors may underestimate the potential growth for major innovations and when we think about AI or when you think about AI, do you believe that that is true today?

Jonathan Curtis:  Yeah, I do, and I'll talk a little—the core of our understanding of why this happens. Human beings tend to be very good at linear thinking. They're not very good about thinking about compounding growth. And you see the same thing in investing. Human beings often miss the value of compounding and of long duration stuff. And so these models are ultimately massive compounding machines. The more compute you put into them, the more data you put into them, the algorithmic improvements you put in them, they all have compounding benefits.

Now, what's super interesting right now is most tech investors grew up over the past, let's say, 50 years, where there was another compounding function at play. It was Moore's Law. Every 18 months to two years, semiconductors would get half as expensive and two times as capable. So every 18 months to 24 months, there was a four times sort of natural speed up compounding benefit that was going on. And that powered everything that happened in tech in the past 50 years. It was a really powerful compounding force.

These models are improving at a pace much faster than Moore's Law. And if that continues on, the same thing that we saw with Moore's Law that ultimately drew digital everywhere, right? We went from a handful of mainframes back in the late 50s, early 60s, to a point where I've got an iPhone in my hand here, an iPad in front of me, a PC in front of me. I've got a computer on my finger; I've got a computer on my wrist. Moore's Law made that happen, that powerful compounding force. So now if the same dynamic goes on in these models and they're improving at a pace faster than Moore's Law, then intelligence is going to be everywhere. And ultimately human beings are just linear thinkers. They're terrible about thinking about compounding growth. So I'm confident they're not fully appreciating the opportunity that is really here.

John Przygocki: So let me take a slightly different tack here. I would say that generally speaking here, your perspective on AI has been extremely positive in our conversation. Is there anything that keeps you up at night when you think about the future of artificial intelligence?

Jonathan Curtis: The past 20 years have been super exciting in technology. We have learned how to do scale up cloud computing, and we've ridden the benefits of Moore's Law. And out of that, some of the biggest innovations and most impactful innovations on society have been some of these social network platforms. And they're wonderful. They allow me to connect with old friends and old high school buddies and the like.

But when you wrap an ad-supported business model into them, they can also create powerful and distorting incentives. And so I think we need to be careful as a society about wrapping the wrong business model in with AI, because we may end up in places we don't want to be. And the fact of the matter is, for all this wonderfulness of AI, the companies building these models still don't entirely understand how they operate. And so if we don't fully understand how they operate, they're going to be hard to govern in the early days. And then if you wrap the wrong business model around them, you could get bad outcomes. So that is the biggest thing I worry about.

And there's other stuff I worry about: the energy intensity of these models. I think human beings and these models together, working with humans, will solve some of that. There's certainly stuff for the courts to work out over fair use of data and the like. But where I started, that's my biggest concern.

John Przygocki: Okay. So Jonathan, as we look to move to the conclusion of today's conversation, I'm wondering if there's one thing in particular that you would like our listeners to take away from what we've talked about?

Jonathan Curtis: Yeah, listen, we're at the beginning of, and I mean the very beginning, of something we think that will be very, very profound. It will be positive for the growth of the global economy. It's going to be wonderful for many of the tech companies that have already been identified, both on a revenue front and efficiency front. It's going to be great for knowledge worker- and labor-intensive economies, so it's an exciting time.

I would encourage investors to be thinking long term and to have an allocation that allows them to benefit from what I think is set to come. But also be open to the idea that we'll have a lot of volatility. So there's going to be opportunities to be opportunistic, and there's a lot that we haven't figured out yet.

But the big thing is real. We have figured out how intelligence works and we are now scaling that up. And we're going to thus get a lot more innovation in the years ahead. And that is going to raise the standard of living for human beings and going to generate, we believe, a lot of value for investors.

John Przygocki: With that, Jonathan, thank you for spending your time with us this afternoon and for sharing such great and informative knowledge with us. To all of our listeners, thank you for spending your valuable time with us for today's exciting and informative update with Jonathan Curtis focused on AI. If you'd like to hear more Talking Markets with Franklin Templeton, please visit our archive of previous episodes and subscribe on Apple Podcasts, Google Podcasts, Spotify or any other major podcast provider.

This material reflects the analysis and opinions of the speakers as of July 1, 2025, and may differ from the opinions of portfolio managers, investment teams or platforms at Franklin Templeton. It is intended to be of general interest only and should not be construed as individual investment advice or a recommendation or solicitation to buy, sell or hold any security or to adopt any investment strategy. It does not constitute legal or tax advice. 

The views expressed are those of the speakers and the comments, opinions and analyses are rendered as of the date of this podcast and may change without notice. The information provided in this material is not intended as a complete analysis of every material fact regarding any country, region, market, industry, security or strategy. Statements of fact are from sources considered reliable, but no representation or warranty is made as to their completeness or accuracy.