Matheson in the Loop podcast – episode one part one and part two

This is Matheson in the Loop, your front row seat to technologies, ideas and regulatory developments shaping the future of business with insights from Matheson and voices from across the industry.

Hi, I’m Marie McGinley, Head of Technology and Innovation Group at Matheson. Today, we’re tackling one of the most consequential shifts in the technology landscape. The move from AI as a tool we use to AI as an agent that acts. We’ll talk governance and what it means to manage a workforce that increasingly includes AI agents alongside people. To help us get into all of that, I’m delighted to be joined by Andy Wu, Associate Professor of Business Administration from Harvard. Andy, welcome to Matheson in the Loop.

Hi, Marie. Thank you for having me.

So let’s start with governance and we’ll dive straight in because it’s a word that everyone uses and the thing almost nobody agrees on the meaning of. So beyond policies and the committees and what does good AI governance actually look like in practice and practically speaking. How do you build that oversight in without slowing innovation down?

So this is a really tough question that boards and executives all over are wrestling with. And I’ll just start off by saying that actually I’m a bit uncomfortable with the word governance because it almost feels like that the intent is to prevent people from using AI in your organization. And let me just start off by saying that the challenge here is that there is a fundamental tension. at a very deep level between innovation and security, and that every amount of security and protection you add to the organization prevents the possibility of innovation specifically around new use cases of AI being discovered in your organization.

And so to give you like just two thought experiments on that, one thought experiment would be, you know, how many different AI apps do you let your people use and how many different… pieces of proprietary data do you let those people use now if you have two employees and two different apps and two pieces of data then you have two times two times two eight possibilities of innovation now companies will naturally resist doing that because they want to prevent the wrong employee from getting this data they shouldn’t have or using an app that might be dangerous and then you end up with perhaps only one instance of innovation instead of eight

To give you another example of this tension, so one thing I’ve advocated for is that the most rich source of data that companies should be capturing is the content that is happening inside their meetings. So I’ve actually encouraged companies to record every single meeting in their company and save that transcript for future use for AI training and inference. Now, the typical pushback I get on that suggestion is, hey, Professor Wu, in our meetings, the things we say, They’re so sensitive. They’re so controversial. We couldn’t possibly record them.

I understand that. But let me make the following point. That meeting that you are most uncomfortable about recording is the meeting that would have been most valuable had you recorded it. Because that’s the meeting where you really said the truth. You really said what you believe, what the organization is really about.

Now, that trade-off between innovation and security is something every company faces now. And my point is that For an established enterprise, the fundamental threat of generative AI is that it lowers the barrier to entry in a lot of industries. So everything about this favors small companies. And this governance question also favors new entrants in small companies because a new entrant is going to have intentionally terrible governance as a rational strategy to compete against a larger company that cares about the governance. And it’s rational also for the larger company to care.

Now, that said, the general way I would think about this is that I think as an executive, it is about understanding what are the acceptable risks you are willing to take in this space and then narrowly defining a set of rules and procedures around the very few things that you care about.

So take the example of a pharmaceutical company. There are certain pieces of data, let’s say patient records from a clinical trial. that they have a legal and moral responsibility to protect and keep secure. I understand that. However, most data inside the company is not that. And what ends up happening in most companies is data gets protected at the sort of lowest common denominator, where the pharma companies actually will treat every piece of data like it is patient data, but it actually isn’t. And so we want to narrowly define that rule, say, around data, around just that protected data.

Yeah, that’s a very good point. And that’s something that we’re seeing from a client perspective that a lot of organizations are struggling with. And it’s that balance of innovation, advancement, and as you say, security. And I suppose that naturally can… shifts us to the next point and that being AI agents. And essentially we’re seeing them increasingly move from assisting us to acting on our behalf. And I suppose this is relevant in the context of that security discussion. So how do governance frameworks need to evolve in your view to address that shift from delegating tasks to delegating actual decisions?

Yeah, and this is going to be a huge step for organizations. And you’re exactly right that this really cuts at the core of that innovation versus security trade-off. Let me just start off by saying that in some sense, I’m kind of an AI agent skeptic where the technology is unbelievable. Like I’ve been playing around with this stuff. I’m sure you have too. Unbelievable. However, I am skeptical that pretty much any company I’ve ever interacted with is actually capable of letting the agents run rampant. It’s fully technically possible. In many ways, I think they should. But to do so would be a complete departure from the cybersecurity and governance regimes of the past.

And so to give you like one example, in principle, you could have an AI agent that automatically makes jobs postings and then screens resumes and then, you know, submits job offers, right? In order to do so, you’d actually have to connect that AI agent to your payroll system, to your financial billing system. to like your whatever your job posting board is, all that stuff. And that is an unprecedented security risk. In some sense, you know, we take that risk all the time by delegating these tasks to employees or contractors. But this is a whole new regime that we have never been in before.

And my point is that if you actually don’t connect the AI agent and give it the ability to act across those systems, say to make that final job offer, you basically don’t have an agent. Like there’s actually kind of no point. And so in that sense, I’m actually a skeptic of it in that I think this is going to take actually a long time to manifest.

Most likely what is going to happen here is that there’s going to be a set of new entrant companies that are going to take tremendous risks and gamble on the agents. And they will simultaneously prove that it works and it’s a miracle and that it can lead to absolute catastrophe. And you will see both happen in the marketplace at the same time. And then. incumbent organizations are basically going to sit around and watch this and then read the tea leaves however they want to read the tea leaves. You can interpret both those stories however you want, right? Because they’re happening simultaneously.

Now that said, for the agents and such, I do think it’s important for companies to gamble here. Like in that, there are, again, say, patient records at a pharma company. We don’t want the agent running rampant on the patient records. Absolutely. But let’s say you’re doing, you know, accounting in the… company. In some sense, like the marginal details of whether this receipt reimbursement gets filed or not, like I’m okay letting an AI agent run rampant on like employee travel reimbursements. Like let’s just reimburse all of them, whatever.

And I suppose when we’re discussing the whole element of AI agents and kind of that subject of control that you just mentioned, I want to put the kill switch question to you directly because it’s gaining real momentum in the boardroom, particularly around boardroom conversations and particularly what we’re seeing in practice. So are organizations focusing on the wrong problem here? If an organization’s AI strategy hinges on kill switches, is that a sign that governance has already failed? Or do we think kill switches simply become a standard control for agentic AI in the same way an emergency stop is standard in other critical systems?

Now, when we talk about the possibility of a kill switch, That assumes that prior to needing the kill switch, you have already integrated the agent or the AI with the various systems inside your company. So you’ve already sort of taken the gamble in advance. Now, in that sense, I respect it in that, yeah, I’d rather you take the gamble of connecting it and then have the kill switch later.

That said, as far as managing the kill switch, should you want something like that? I think the way to think about it would be the way the, like say the military thinks about like what they call the kill chain or the chain of command for a kinetic action. Like in that sense, like I think for very critical decisions of a company, like we could imagine that sequence of approvals happening.

Now, the tension, of course, is that the more items you build into what I’m calling the kill chain, the more people that are involved in that, you’re actually giving up the whole point of having the AI agent in the first place. Like every step of human approval in that process is slowing your organization down, in which case you might have just done it the old school way anyways.

And looking at that, Andy, and I suppose this, begs the next question around governance as a principle in itself. And I think, Kenneth, your point is well made at the start. I’m not sure governance is the correct term when it comes to AI. And what we’re seeing is some organizations are getting AI adoption right. Others are struggling with that trust and compliance piece. So back to the point where, as you say, things are being slowed down quite significantly. So in your experience, is there a difference between them in terms of the governance piece? And what have you seen in practice that’s worked well? And what do you think will work well for those successful organizations?

In the short-term sense, it’s very much the case that for pretty much any Fortune 500 enterprise, they’re adopting AI very slowly for all these concerns that we’ve talked about. And I can’t say that they’re wrong. Like, in many ways, I see the rationality of it, but I do think it is short-sighted. Like, I can see why that’s the right decision now. I don’t think it’s going to be the right decision in five years.

Alternatively, imagine you are a, like a… student just graduating from Harvard Business School and you are starting your own company, a one-person company. I wish our listeners could see what this company is like. It is unbelievable. They are taking unbelievable gambles. They have all their data systems connected. They have AI agents running rampant, doing all the marketing, writing all the legal contracts. It is unbelievable. Like, you would find it terrifying. But actually for them, I think it makes sense. I think that craziness actually is rational for a new entrant. You don’t have that much to lose, right? Like, what data are you going to lose?

Now, in the longer term, I think this is going to be very important an issue for large organizations to overcome in order for them to be competitive against the threat of new entrants. Now, in particular, what I want to emphasize here is that in the age of AI, the only advantage that an incumbent large organization has against the new entrant is data. You have more data. But nearly every big company I know today actually prevents themselves from using their own data for anything productive with AI, for all the reasons we talked about.

And I do think that the only way to survive in the onslaught here is actually to allow more of your employees to use more AI tools, to use more of your company’s data. And that’s going to allow efficiencies and productivity both within functions. But what we have yet to explore yet is the cross-functional and cross-business unit synergy possibilities. And I’ll give you sort of two trivial examples of this.

So one example would be, imagine you have a multinational enterprise, and in one part of the company, you manufacture air conditioners, and the other part, you manufacture cars. Now, those are typically kept separate, and there’s separate data sets and AI for both those divisions. But in principle, in theory, there is some value of using the air conditioning data to improve car manufacturing. like in theory. And we actually don’t know yet because no one has actually attempted that because no one has combined those data sets.

The other possibility is a cross-functional integration of data. And I’ll give you a very intentionally dumb example, but I think it illustrates what the possibilities are. So there was one company I was working with where they were considering combining their HR data with their sort of financial revenue data. And There are some traditional job categories, say like a salesperson, where that is already combined, where in order to do performance compensation for salespeople, you have to sort of combine their HR records with their financial records to pay them, commensurate with their performance. And so you actually, as a company, have the benefit in those situations of knowing exactly how valuable that salesperson is to the company.

Now, we don’t have that insight for any other job category in the company. For instance, this particular company has a receptionist at their front desk in their corporate headquarters. If you, in theory, combine all these datasets, you can force the AI to give you an estimate of the revenue contribution of this receptionist who sits at the front desk. Now, by every measure, that number is a hallucination. However, most CEOs I know would be really curious to know what that number is. Like, otherwise, we never really thought about, is this the right set of people or the right choices we’re making in the organization if we can’t link it to financial performance?

Now, fortunately, I think in law firms, you guys have tremendous tracking already of everyone’s contribution. So you guys don’t have to worry about it so much. But this is, you know, you can imagine for plenty of other companies, they don’t track the hours.

Absolutely.

Stay informed. Stay in the Matheson loop.

So let’s talk about what this means for how organizations actually work day to day. Do you see AI agents becoming more like digital workers rather than digital tools? And if so, what has to change first organizationally and culturally? What management practices do you think simply won’t survive that shift?

So technology. changes slowly over time. It feels fast, but over the arc of history, there’s patterns. And one of the things that’s going to happen here is that by the time we have digital workers, they will still feel like digital tools to us, even though they are basically digital workers. Frankly, in my view of the world, like ERP systems today or CRM systems are already digital workers that humans used to do. But we think of them as tools that are part of our workflow. And I think the agents will feel like just tools part of our workflow.

At some level, There will always be human, like overseeing or using AI. That said, the question is not whether there’ll be a human. The question is which human and how many. And I think there’s going to be a very, very dramatic shift in what the workforce and the skill set of the workforce is going to have to look like in the future. In particular, AI allows for a tremendous shift in who is doing what job.

And so I… often give the example of Photoshop. So Photoshop is a specialized software for editing photos. And I don’t know if you’ve ever tried to use it, but I’ve tried and it’s really hard and I have no idea how to use it. Really confusing, weird buttons, like whatever. And so that we have experts in our society who are really good at using that software that we call graphic designers.

Now, in the future, what you have to assume is that there will be something like Photoshop, but instead of clicking buttons, you will just talk to it or chat with it and it will… it 10% lighter or make it sepia tone. That actually fundamentally transforms who uses Photoshop. And so one possibility is maybe the person you have doing graphic design right now, say designing a logo, maybe they just keep using this new Photoshop that you talk to. Maybe another world is the middle manager that manages that person can now use Photoshop instead of that original person. And another world is maybe the CEO just uses Photoshop directly. hey, come up with some logo ideas, and then it generates it, and then you go from there.

And that’s going to be a huge shift because we have a different reporting structure in companies for who is going to do what job and who can do what job. I think this is really, really big. Look, the big academic question around AI is always, is AI a complement or a substitute to humans? And the answer I will give is it is a complement to some people and is a substitute for other people. And this is going to be very dramatic. It’s going to be very, very big.

And I’ll just highlight, I think, one phrase I hear a lot because I think this phrase is very instructive. So one thing I hear CEOs saying all the time, and I’m sure our listeners have heard this too, is CEOs will tell their people, you should use AI so you can spend less time on the low-value work you don’t want to do and spend more time on the high-value work that you want to do. CEOs say that all over the place. I think that statement is flawed in so many ways. I’ll just walk through some of the problems with it.

First, that it’s not clear that there is necessarily, quote, high value work for that person to do. And even if they had that work available to them, it’s not clear they would want to do it in the sense that high value work in our framework is work that is, quote, requires tacit knowledge, which is to say uses your brain and has a high cost of error, meaning that there are high stakes to it. In my view, that is the definition of a terrible job. If you spend your whole day using your brain and making high-stakes decisions, oh my goodness, that would be a terrible, terrible thing. Like, so it’s not clear people would want to do that.

The other sort of paradox of this is we might spend more time doing low-value work. And I’ll explain it this way. So at Harvard Business School, we used to have, decades ago, one assistant for every professor. And so way back when, when you wanted to reimburse your travel expenses, you’d kind of dump all the receipts on the assistant and it would magically get reimbursed. Now what we have is a regime where there is now one assistant for every four professors. And what ends up happening is actually we have software for doing travel reimbursements and professors actually spend more time doing travel reimbursements than they did four decades ago.

And this is related to my Photoshop example. Yeah, now the CEO might be designing logos that they didn’t have to before, right? And that’s actually going to add more low value work. And these like tensions actually in it, I think, is what makes the transformation so difficult. There’s just a fundamental rebalancing of work as we have, quote, digital workers or digital tools that is going to reshape organizations.

And Andy, just on that point, do you think then that we’ll see a reshaping of, let’s say, the structural workforce within organizations as a result of the use of AI agents and then separately, do you think there will be an evolution whereby we’ll actually see, and back to your point of the CEO doing the Photoshop work or the development of those logos, in relation to that particular example? Do you think over time we will start to see a shift back to realizing that there is a reallocation of work that really should not have been, that was considered efficiency in the first instance, but in reality wasn’t? And we might find ourselves backtracking somewhat over the next couple of years?

You are going to see all of those happening simultaneously because… companies, as we talked about, are under utilizing AI and new entrants are over utilizing AI. And you’re going to see it happen both directions where we will discover in the big companies that they have forfeited many opportunities for AI and efficiencies and innovation. And in the small companies, yeah, they’re going to do some real gambles on AI and things that shouldn’t be AI. And we will find out the hard way with those guys.

In that mix of things, the reason that it’s difficult for both sides, both for the big company and the small company, is that up until the current moment thus far, we actually don’t have a true sense of what it means to be truly human. Like we actually haven’t defined exactly, like in an organization of however many people, what are the things that everyone does that is actually human, that requires human judgment?

The point is that… think every company has some piece that requires human judgment. Every employee of every company has something that requires human judgment. But it is not their whole day. It is far from their whole day. If I had to throw a number out, I think for nearly everyone, including CEOs, it is actually less than 25% of their day. And we have to pinpoint exactly what that thing is. And until we do… Well, big companies are going to get scared and overestimate that number. And then small companies are going to be too aggressive and underestimate that number. And we will have a clash of various successes and failures in the coming years.

Yeah, I think you’re right. I think we will have to see that rebalancing of the roles in the future. And I think you’re right in terms of the under and over investment because we’re seeing it across the board.

Yeah. Yeah. So then given what I said. Take like a, you know, a legal contract, right? Like what are the parts of a contract that you think truly require human judgment versus I imagine there’s plenty of it that you think doesn’t, right?

Of course. And they’re probably the boilerplate provisions that you would typically see in every contract. But if you take a step back and it’s back to kind of one of the points we talked about in our competitive advantage discussion, we are now becoming… engaged by clients early on in commercial negotiations or commercial discussions with counterparties be it in relation to a commercial relationship or an M&A transaction and as part of that we’re working with our clients hand in glove in coming up with a commercial structure that works for both parties so when it comes to let’s say that human element to a contract and the thinking piece it’s ensuring that the commercial agreement that’s reached between the parties are ultimately reflected in the drafting so let’s say when you look at the operative provisions within a contract be it the services, the types of services, the standards of services, the service levels, the service credits, the liability provisions, the indemnities, the IP protections, all of those elements probably involve a lot of human element because you’re part of these commercial discussions and they’re not necessarily a lift and shift.

you need to have there is that core reflection of the commercial agreement to achieve the commercial and beneficial outcome for both parties because otherwise you’re going to have one party feeling that they are not being heard in the discussion and that’s counterproductive for everyone concerned so from that perspective I think those core elements of contractual negotiation very much require the foresight strategic direction and ensuring then that what is agreed commercially is reflected in the drafting.

Yeah. And to my earlier point, what you’re going to see in the coming decade is that you will defend that human judgment fiercely and there will be new law firms that are going to gamble and AI that part. And we will have a fierce battle of, you know, whatever is going to happen on this.

Absolutely. So do you think the winners of the next decade will be those that build the best AI systems or those that are best at coordinating humans and AI agents working together more effectively than their competitors?

Yeah, absolutely the second point. I think it is about coordinating the interaction between the people in an organization with the technology that the organization provides those people. This is going to be a… a monumental effort because companies are very, very far from the point where that is anywhere near seamless.

And so as we pointed out earlier, in many ways, the general technology here is kind of generic and available to everybody. But what isn’t generic is the amount of data the company has and the integration of that data. When I say integration, I mean, how easy is it for the employees to access the data to do their work?

And so the typical large enterprise is data scattered across business units, across geographies, across applications, across functions. And it’s really not possible today for employees to use all that data. And if you can set up a system that allows those employees to access any piece of data in the company, that would be a tremendous opportunity. And at least for the five-year horizon, it would be a tremendous differentiating factor. If you had that in your company, you’d be way ahead of your competitors for the next five years.

Absolutely. And I think it’s exciting and scary prospect all in one go. Andy, thank you.

Thanks for listening to Matheson in the Loop. If you’d like to continue the conversation, please get in touch with a member of our technology and innovation group. We’d be delighted to hear from you. Join us again soon for another episode of Matheson in the Loop. Stay informed. Stay in the Matheson Loop.

This is Matheson in the Loop, your front row seat to technologies, ideas and regulatory developments shaping the future of business with insights from Matheson and voices from across the industry.

Hi everyone, I’m Marie McGinley, Head of Technology and Innovation Group at Matheson. Today, we’re tackling one of the most consequential shifts in the technology landscape. The move from AI as a tool we use to AI as an agent that acts. We’ll talk kill switches, intellectual property, competitive advantage, and everything in between. To help get into all of this, I’m delighted to be joined by Andy Wu, Associate Professor of Business Administration from Harvard. Thanks again for joining us, Andy.

So I want to shift now to intellectual property because it’s an area we spend a lot of time on with clients. How would you characterize the current state of IP protection in AI, particularly around copyright trade secrets and data licensing? Where does it break down most? And how do the varying regimes globally impact that breakdown?

So we are in an unprecedented era of… a horrifically weak intellectual property regime. Now, I teach about innovation and technology strategy, and we are very proud of the benefits of providing intellectual property. Unfortunately, for a lot of structural reasons and historical reasons, I think this current era is ushering in something very, very different. And I think companies will need to adapt and play very differently around intellectual property on the assumption that in the future, IP protection will be at least as weak as it is now, if not even weaker in the future. And in that sense, I think it’s really critical for companies to accept that IP might be weak in the future, and that we need to find different ways of operating in different business models around IP rather than specifically monetizing the IP itself.

And so here’s a, I’ll give you my pithy phrase for it, I’m over trivializing it, but my… My pithy phrase would be that copyright is dead, but trademark will live forever. And what I mean by that is that, let’s say you are a media business and you have some sort of media franchise. In principle, every time you release an episode, like that specific copyrighted film is gone. Like you have to assume everyone has it already. Now, that said, you do own the trademark and branding of that franchise. That is something you can protect in court and people can respect.

There are other ways of monetizing and building a business around that that won’t fundamentally rely on licensing and generating revenue off the copyright. And so to give you one example, let’s say you own some TV show franchise. Maybe you can’t make money on the TV show. But certainly you own the brand and perhaps you can build a theme park or a cruise ship or like some sort of clothing or things like that. And I think companies are getting to shift that direction of finding different ways of monetizing IP.

I think that makes sense because I think ultimately, as you correctly point out, and there is a distinction, particularly with copyright between, let’s say, the US and Europe, for example, just in itself. So from that perspective, that will. require a significant shift in itself, because particularly given the fact that we don’t have a registered copyright register, which is that fundamental distinction.

So I suppose if we build on that, and as you mentioned, much of the AI industry has been built on IP with unresolved and contested rights, does that change how you think about competitive advantage? first to assemble and deploy no matter more than first to invent?

Absolutely. So I think there’s two ways of approaching and solving this fundamental issue. The first is a logic that all IP will eventually be stolen, but we’ve got to be on the treadmill and running faster than everybody else and generating new IP that will not be presently in that current moment available to anybody else. on the assumption it’s going to be gone tomorrow. But if we can keep generating future IP, it will still be valuable in the future for that temporary moment that it is new.

One of the fundamental limitations of AI, of course, is because it’s based on historical data, it’s actually not very good at understanding real-time information or trends. And so in that sense, today’s data is always worth geometrically more than yesterday’s data. And so if you can be on the realm of like staying ahead on AI and generating data or what have you and generating new intellectual property and generate it faster and faster, then I think that can be an advantage.

That said, I hope it sounds just like it does to me. That sounds really exhausting, right? That sounds like a terrible, terrible thing. There are some businesses in a position to do that. So to give you one example, the social media business Reddit, they generate the equivalent of a Wikipedia every two weeks. Now, that’s a company I believe, okay, there actually is some data there and they have real-time data. But for the rest of us, it might be difficult to do that.

And so another way to approach this is to accept that, you know, whatever this IP was in the past, that this IP is fragile and that we need a new business around it instead of trying to compete here. And the analogy I like to use is a gold rush. And so… AI race is often likened to a gold rush and the AI model builders, they are what I would call the diggers of gold. Now, in a gold rush, you don’t have to dig for gold. There are other options. You could also sell shovels or you could also make jewelry, which is to say you can be a supplier to the gold diggers or you can be a complement to the gold diggers. And in either case, the weaker the intellectual property of this hypothetical gold is. it’s actually better for the shovel sellers and the jewelry makers. It means the gold is really cheap. And there’s a lot of gold diggers.

So then in that sense, like I would think about ways to be a supplier into the industry that is becoming transformed and weakened by AI or a compliment. If we generalize it more broadly, let’s take a publishing business or a media business, right? In those lines of business, the gold that you traditionally have is the content you generate. As they always say, content is king. The content is no longer king. But the content does create value. Mickey Mouse does still mean something, right? You can’t capture value from digging for the gold. In which case, you’d want to find ways to, quote, sell the shovels or make jewelry. Hence, you know. the theme parks and cruise ships and whatever ancillary businesses exist from that IP.

So I think on that side, Andy, we’ll probably see a huge diversification of organizations as a result of AI possibly to try and maximize that competitive advantage. So if we bring things closer to home for you, Andy, you’re at Hartford, an institution that’s refined how it teaches over centuries. What has Harvard’s approach been to integrating AI into that? And are there lessons any organization can learn from Harvard’s experience?

So let me just first be transparent and say that I think the current AI moment is the biggest threat to higher education that we have faced in the last, probably Harvard’s entire history. And there are many challenges with this, but one of the deepest issues is that, at least in the short term, we have lost the ability to assign homework. We actually can’t make students do homework anymore because any problem set in any essay, the students can use AI to do, or any case study, they can have AI summarize the case study for class, right? That’s something we have to overcome. I think we will be able to overcome that.

But more generally, as we look at the educational enterprise, like I’ll highlight a couple opportunities. First, in terms of the lives that we are training students for, it is obviously the case that our students in the future are going to be actively using AI all throughout their jobs. And so we have built in a lot of curriculum to teach students how to not only use the various AI tools productively and safely with the proper governance, as we’ve talked about, but also to get much deeper in the tech stack.

What is really interesting in the current era. So I’ve been teaching technology strategy for a decade. And historically, the class I teach is a popular class, but kind of niche. Like it is a very specific set of people, usually with a computer science background that want to go into technology. It has gotten so broad now. The class is ridiculously popular because actually now everyone sees the importance of understanding much deeper into the technology stack. And we have students now that, you know, come from investment banking and now they want to work in semiconductors, which is unheard of, right? But they’re seeing that opportunity and we want to make that available to them, especially as, at least for the business school side, the goal of Harvard Business School, you know. the way I would frame it, has always been to train CEOs. We want as many of our students to be big company CEOs as possible.

Now, that role is transforming fundamentally, and we want to make sure they have the skill set for that job. And to give you one analogy, if you go back 100 years ago, the traditional path to CEO would be the head of a business unit. Then sometime after that, it became a chief operating officer person. And then sometime after that, it became more of a chief financial officer kind of person. My view in the future is that we are going to have to draw a lot more CEOs from the ranks of CTOs for all of the reasons that we have talked about in the call so far. That, in fact, the understanding of your company’s data and the technology architecture and the training you give your own people on how to use your technology, I think, is the skill set that will distinguish a lot of large enterprises. And we’re going to need CEOs with that skill set. And I’m proud to say that we’ve added courses and course sessions that will leave students with that experience.

Now, more broadly for the university as a whole, the other big challenge is on the research front in that it is true that AI can actually do a significant amount of the academic research that we have been publishing in the past. Now, we like to think, and I do believe that there is like some core component of human insight that goes into every research paper. You know, I have to believe that. That said, though, the AI can do a lot of the paper, actually. And so we’ve certainly got to adapt that. And one of the challenges there for the academic journals is how do you review papers for journals when there is such a deluge of AI-generated papers being submitted to all the journals.

Now, that said, I’m really proud of the work we’ve done at Harvard. In particular, I think deans and university presidents really understand how critical this moment is. And so they’re trying all sorts of things. We’ve got different professors in every class running every experiment of how to administer different things. and how to run different experiments on how to use AI to teach. In our course strategy course, we have, for example, a new like AI bot where we’ve loaded all the course content into an AI bot and the students can like ask questions and stuff. So that’s been a tremendous resource for the students. Other instructors have developed like AI case materials where like instead of reading a story about one company, you actually have a conversation with the hypothetical CEO of the company and then use it that. method to like learn about the business and make decisions.

That said, I tend to push back a little bit. So I’m actually what’s funny is I teach the class at HBS for the most technologically interested students. And I try to use like no AI whatsoever in my class, kind of a Neanderthal in that dimension. And the joke I often make about this is that, look, my students got to want to work in the technology industry. And in the technology industry, the Goal is to get everyone else out there like using technology and you extract every minute of time and every dollar out of them through their use of technology. Right. And we are not going to do that to ourselves here. Like we are going to do it the old school way in order to learn how to do that effectively.

So just lastly, in relation to that, and I suppose most organizations are basically building on and deploying. the same foundation models. So if the technology is the same for everyone, and we all have access to similar gen AI capabilities, where will competitive advantage actually come from?

And so as you point out, the challenge here is that everyone has access to basically the same frontier AI technology. And how do you differentiate yourself and build competitive advantage in that world where the… this magical thing is the same for everybody. And I’ll go back to the same answer. You got to sell shovels and make jewelry. So it’s not just the AI companies and media companies. We’re all looking for ways to sell shovels and make jewelry.

And I’ll be more precise here for like most enterprises about what it really means to make jewelry. So yes, the AI model is the same for everybody. But what is distinct about your company is the integration between that AI model and your own company’s data and that data then with the specific applications that are relevant for your business and the interface then between the applications and your people, which is to say training. And so the differentiation comes not from the AI model and it doesn’t come from either the application either because actually front-end application development is kind of trivial these days. It is about the integration of the model with your data. with your applications, with your people.

And in that sense, I think it’s going to be really important for companies in the modern era to maintain a fairly robust IT team that actually can do that integration work internally. Now, there will also be service providers providing that service. There’s a phrase that kids throw around a lot these days. It’s a Ford deployed engineer. That’s just what the kids call IT consulting. But that will also be really important. But it’s worth recognizing what all these for deployed engineers from all over the place, Anthropic or Deloitte or whatever, what they’re doing. What they are doing is integrating the generic AI model with your data, with your applications. And that is hard, hard work.

Linking back to the governance question, it’s actually, it’s not a technical problem. It’s actually a huge like political problem inside companies. That’s going to be important. And then the other side of this is that you can also backwards integrate because app development is trivial and you can do the integration yourself. despite being painful, like companies are gonna have to bring that in house and do a lot of that. And I think that’s going to really define and distinguish companies going forward.

And there’s some companies in the dot com era that have already gone through the transformation. So if you look at, say, in retailing, like. Walmart, as they expanded into e-commerce from brick and mortar, had to build like a very complicated, sophisticated data integration effort between the e-commerce and the brick and mortar retailing to get to like the omni-channel dream. And so I think more and more companies are going to have to do that.

Agree. I think, Andy, you very much hit the nail on the head on that one. Diversification is going to be key to that. competitive advantage. And I think it’s something that a lot of organizations going to have to look at and particularly kind of look at their own organizations in the next 12 to 18 months and see where their future lies, particularly with these advancements.

I also want to be clear that, you know, despite me being categorized as an AI person, I am very pro-human. And I think all of our organizations, we have to die on the hill that humans are important. If not, then we’re all kind of dead in the water. We have to believe that humans are important, but I think we have to be increasingly precise and proud and defensive about the very specific places that we think humans are important.

And so… I’m encouraging companies to do is have active meetings every month where they redefine and restate like what is the critical thing our organization does that is so important that we are going to dedicate extensive human time to this. And the other things in the organization, less important, perhaps we can AI more of it.

And so like to give you one example, take a pharmaceutical company. One pharma company told me they thought their core value and their core mission is to develop and commercialize new drugs. I think that is a phenomenal thing for the company and a phenomenal thing for the world. Now, if you actually look at that company and you ask them, what percent of the time, like human hours in the company, are actually spent developing new drugs? Actually shockingly small. Like less than 5% of time is really spent developing new drugs. The rest of it is a variety of administrative functions to support the people who are developing the drugs.

Now, once you have this conversation about what is important, Then we can decide where we really want to put humans and drug development. We can, in some sense, be wasteful with humans. It doesn’t have to be maximally efficient. We could just dump humans at it because it is so important to us. However, we really got to look at the other 95% of the company. Frankly, we’ve made a decision here. It’s actually not as important. No, the most important thing is the new chemical compounds.

Andy, this has been a fantastic discussion. Your perspectives are hugely fascinating. Your insights have been… phenomenal. So thank you very much.

And thank you for having me.

You’re listening to Matheson in the Loop, a podcast from Matheson.

What really stands out in this series for me is that AI cannot be viewed simply as another technology implementation. It raises more fundamental questions about where value sits within an organization. how decisions are made, and which activities genuinely require human judgment.

It was also particularly interesting to hear Andy’s perspective from the US. In Europe, much of the discussion around AI naturally focuses on regulation, governance, and risk management. Andy brought a different perspective, one centered on innovation, competitiveness, and organizational adaptation. That contrast is a useful reminder that AI is not simply a technology issue. It’s fundamentally a leadership and business transformation challenge.

From a legal and strategic perspective, the challenge is not to eliminate risk entirely. It is to identify the risks that truly matter, put proportionate safeguards around them and still leave enough room for innovation. The organizations that succeed will not necessarily be those with access to the most advanced AI models. They will be those that understand their own data, integrate technology effectively into their operations and remain clear about where people add distinctive value.

Perhaps that is the central reflection from today. AI may change the tools we use and the way work is allocated, but sound judgment, clear accountability and a strong understanding of an organization’s purpose will remain essential. The technology will continue to evolve, but those fundamentals are unlikely to change.

Thanks for listening to Matheson in the Loop. If you’d like to continue the conversation, please get in touch with a member of our technology and innovation group. We’d be delighted to hear from you. Join us again soon for another episode of Matheson in the Loop. Stay informed. Stay in the Matheson Loop.