AIと戦略：ビジネスの優位性を再定義する――マイケル・デイヴィス氏との対話

Welcome to the Pulse of AI. Welcome back. I'm Jason Stoughton, and this is the Pulse of AI podcast, Season 8, Episode 173. I just want to take a second and express how much I have really enjoyed this season so far. I've had lots of really interesting guests on, including founders, AI scientists, investors, AI thought leaders, and AI practitioners. There have been so many great insights into what is really going on in AI, both at the innovation edge and also around deployment on the enterprise side. So keep tuning in. There's much more to come. In the meantime, on this podcast, I'm joined by Michael Davies, who is a globally recognized thought leader in business strategy, innovation and leadership. He is a senior lecturer at MIT around innovation and ethics in AI. And he also created the Business of AI program at London Business School. And for 15 years taught the new technologies venture program there. He runs a boutique technology consulting firm, helping business leaders from some of the largest and most innovative companies create value by harnessing and humanizing tech.

I hope you enjoyed the conversation. Let's get to it. Welcome to the show. Good morning, actually. Just good afternoon. Well, it's morning for me. There are some great topics that I want to get into. and I'm really excited about this conversation. Before we do that, you know, tell my listeners a little bit about yourself. You know, what's been your journey? You have a really interesting background. You know, how did you get into what you were doing? What's your passion? You know, just sort of what drives you? That's a good question. There's a couple of, you know, sometimes in your career, there's a couple of really critical turning points. For me, one of them was actually coming out of high school. I thought I was going to get out of Cambridge and be a mathematician. And I took a year off. And in the course of the year off, I actually said, you know, I want to do stuff that's more real world. So I was an engineer. And in the course of that, I got interested in, there's a lot of dissonance.

I got very interested in AI and robotics. And that's germane, because that's how I spent the first decade of my career. In sort of, well, if we were coming up to an AI winter, I guess it would be the AI fall. Like, we've gone past the summer, and it's getting a little bit colder. So at the end of that decade, I moved out of the field of AI into tech and into telecoms. And then about 20 years later, I was drawn back into it because I'd stayed current in the field. I had friends who were working in it, unsurprisingly. And some of the technologies that we first started to use in that first decade of my career were bearing fruit. and shortly after the first deep learning paper came out so actually before the 2015 one through some of the connections at MIT where I was teaching technology and strategy Tom Seroides came to us and said to me hey we think this

new AI thing might be a big deal now they were a really good candidate for this because in those days in particular knowing your data was absolutely critical. And Thomson Reuters is really all about the data. I mean, they are the canonical source of data in all sorts of subject areas. So much more than most companies, they have a lot of really, really well-organized data. And they would have discussions at board level where they would talk about ontologies and taxonomies and stuff like that. So we sort of struck by this, so I dug into it. And I can wrote a presentation as part of a board presentation. what's now 13 years ago we're going anything that can be automated will be automated and there will be an awful lot of things that are going to be able to be automated so made some early calls connections led to a lot of work with Swiss Re which we're starting to see the fruits of now in terms of things like dealing with natural

catastrophes and so on and that and that led me to working on it increasingly over the following years, teaching about it at London Business School, teaching about it at MIT, advising clients on it. And then there's been a couple of other inflection points. The first was sceptical about LLMs, and then, I know that's important to get out there, because it's like, oh yeah, better marketing copy, I was really hard to convince me of the value of that. Software, yes, important, but it's only a small part of the economy. then a couple of years ago I was struck by some real world applications the whole transforming knowledge work in businesses and that was sort of a huge first step and then I think we're seeing over, I was talking to some people about this at the weekend literally in the course of the last

several months we've hit an inflection point in terms of the usefulness of these technologies and that's one thread like this is the big deal the second thread which we can come back to was oh wow these are very powerful technologies but they have some inherent characteristics that aren't very nice and I got really interested in how you mitigate the risks that are associated with them and how you build them in ways that are responsible and ethical and humane so those two threads sort of came together it's not the first time we've had this in human history by the way, as a student of the history of technology, the very first training with passengers on it resulted in a fatality when it was first demonstration and people talk about going back to why can't we all fly like we did in the

1950s and it was such a wonderful experience. But you will give or take 44,000 times as likely to die. So I take those as the hey, let's actually make it happen a little bit more quickly in terms of taking a powerful technology and harnessing it in ways where we mitigate the risks and we're conscious of the downsides. So that's what it would have been here. What a great it's interesting how they converge but before we get into that, I want to talk about both those. I want to talk about safety, responsibility, and AI. I want to talk about AI transformation and the enterprise, sort of your two areas, and where they converge, because there's a lot of convergence there. Especially now when you're talking about agents, you know, and responsibility, you know, and what are companies going to do? It's a discussion we've had. You know, it's interesting, but before we get to that, you know, I just find that, you know, a lot of people, and everybody talks about jobs and work and, you know, I'm not a doomsday in that at all. But I find it interesting the difference between, and if you just look at your career path,

and just sort of as a side note here, the difference between the way the Brits do sort of growth in education and the way Americans do is that I think it's really, really healthy to sort of take that gap year, which is common in, you know, you said you took a year, but that's fairly common in England, right? And here it's just not. And I have a daughter who just entered freshman year, and I kept trying to tell her, I'm like, you should take a gap year because it's like what happens, right? You come out and you take a year and you're like, well, you know, I thought I was going to be interested in this, but, you know, I found a passion or as I thought about it and had some time to think about it. You know, just as a comment on that and different cultures. The deal my parents did for me, which I actually did the same one for my daughters, was we'll buy you a round-the-world flexible ticket, right? And we'll give you enough money for the first place you're going to stay for a couple of weeks. Beyond that, it's up to you. Yes, we will bail you out in an emergency, but off you go.

I love it. Me and my wife, we're big travelers and pre-kids. We did almost two years after the dot-com bus. We did almost a two-year around-the-world trip, and we did 38 countries. I'm a big proponent of that, and I tried to do strategies like that with my daughter, but you get so bought into this, oh, I want to go. My friends are going, and, you know, in England it's easier because everybody's kind of doing it or a lot of people are doing it, and it's one of those things. Anyway, back to this. So let's start with sort of AI safety, responsibility, and AI. And I want to sort of set this down. Before we get into this, before we really dive deep into this, I find this, you know, we're here, you know, a couple months after, you know, the Chachi BT, you know, hiking face thing. I mean, and then we have Anthropoc coming out and how we can't control our agents. This is crazy. And so we're kind of in that framework.

We're talking within that, right? And when people 50 years from now find this somewhere on the way back machine on the Internet and this conversation. But what I find frustrating about it, I guess, is that you've been in AI for a long time. I've been in AI for a long time. And these discussions have been going on since basically the beginning. This is stuff that's been talked about, consciousness in AI, you know, controlling AI, you know, responsibility around AI. And sort of the press, you know, the political powers that be are jumping into this as if nobody's ever thought of this before, right? And we rushed through these, you know, sort of, you know, cycles of media and hype. And, you know, I think now you're starting to have saner voices come into the discussion. But, you know, how would you, I mean, and let's start with sort of the AI companies themselves, right? The people who are creating, you know, these language models.

And, you know, when we're talking about language models, you know, that's only a part of AI, as we know, which I still don't think most people in Washington, D.C. understand, that there's different variations of AI, right? So let's talk about the language models themselves. How do you frame it, right? When you're teaching a class right now on this at MIT, how do you frame the problem? Well, I frame it this way. There's two classes of people here, two classes of actors. it's the people who are creating the underlying models the foundation models and that's one group of people and I have some thoughts about that but they're important companies and what they do will matter but most of us are concerned with how we make use of those foundation models so let me deal with the first group first I have relatively strong views

on what they should be doing. And the question here is, which I sort of find myself going back to, is what is the underlying culture? What is the underlying motivation? And I'll pick up a couple of examples who I would be more concerned about, and then I'll talk about what that means for you as somebody who's trying to harness AI. If you look at Facebook, let's be candid, and I'm out there having said this in public, Facebook doesn't have a strong moral compass. And for what it's worth, frankly Mark Zuckerberg's changes with the wind in terms of what he's advocating. The Analytica scandals, a whole variety of other things. And their response to this was to create an oversight board. Now I'm an engineer, and I worked in manufacturing, and I'm involved in designing a lot of products. this is trying to bake it in after the event

like oh we're going to do product safety how? well we're going to have a safety oversight board and if stuff blows up they'll tell us we shouldn't be having the stuff blow up and what got me was this is just not going to work as an approach so I think about two things and I also think about the same two things which is when it comes to being a user and one is governance right? is there the right leadership at the top? So you ask yourself about what's the history, what's the culture of the organization. Some organizations are deeply, profoundly ethical in terms of what they do. And others are, hmm, how should we put it, more instrumental. Sam Altman, an open AI. Charming, charismatic, brilliant. Saw an opportunity for world domination, went to the dark side. and he now needs to make an enormous amount of money to achieve whatever it is he's decided that he really wants to go to.

Look, he's already very rich, so this is more about power. He's already got more money than he can conceivably spend in his entire life. And that means, oh, if I've got an enormously smart, enormously charismatic person who needs to make a great deal of money, I find that a relatively scary prospect. on the flip side I will have been a supporter of Anthropics openness, their transparency none of us are perfect but for example in that hugging face incident, so the one in which opening eye training model escaped and attacked hugging face that happened on the Wednesday morning of the class when I was teaching at MIT so it was like I said I'm not going to teach this we're going to talk about this issue. Wow. And it was striking, because everybody was concerned about what OpenAI did, and they got more and more concerned about the news, as the news came out about how they knew that the safeguards were compromised.

In fact, I will attribute this to someone else, but somebody saying, oh, they knew that might happen, and they would have been happy if it did. So there was certainly a case we made that this was a great piece of marketing, for the power of the money. flip side, anthropic two ways you can think about how you make a decision what about the rules or what about the consequences this is what ethicists call a deontological or consequentialist perspective on that day, anthropic's rules said we won't let you use our models to attack somebody else even if it's in self-defense so hugging face had to turn to a Chinese model about which I'm always nervous so if I take a step back the concerns I have different levels of concern for different actors there's a spectrum between ones who I are really concerned about the most extreme of which is some of the Chinese

models I don't think you can rely on Facebook or OpenAI themselves they have strong motivations to behave in certain ways I'll put a lot of weight in favor of Anthropic, and we're doing some work in public media using AI, and we're deliberately going to run a little bit behind the frontier so that we can use open source models and be completely transparent about what we're doing. So from that landscape, that's on the foundation side. Do you want to ask me? I mean, really quick on that, though. I mean, to your point, I mean, and as more news comes out, I mean, OpenAI seems to become almost an international cyber hacking company. I mean, it's, you know, I mean, the amount, I mean, at some point there has to be, you know, you know. They've gone beyond the criminal charges. I mean, you know, when you have to have responsibility, but it gets to this point really fast, right?

And I don't want to take us in too far of a different direction, but within this, within what you're saying, part of the problem is them trying to absolve themselves of responsibility by attributing agency, almost consciousness, to these agents. And that scaring the public which is also irresponsible Yes I mean it absolutely irresponsible And it just completely not true There no basis to it And some people have to step back and say, there's no agency. Now, when you train it, you could train it to do something, and then it takes actions, and it seems like it has agency, right, you know, and human, but, yeah, I don't, thoughts on that. It was fascinating. I did a bunch of research in two areas. One was, how do people think about this? And it turns out there's a phenomenon called blame discount,

which is, oh, if an AI is doing stuff that's good, you credit it to the AI. If it's doing something bad, you blame the people who created it, which reflects, I think, the right way to think about this. So, and actually, he's now here. one of my grad students from MIT is now doing a degree in AI and law at Harvard Law School and he and I have been talking a lot about this because back from that we said there is always a human in the loop whether or not you like it you cannot absolve yourself of responsibilities at least as far as I'm concerned to the open AI question I think they've gone beyond being careless to being culpably negligent yeah that's my point once it was an accident twice it's a coincidence three times it's a pattern and we're now looking at scores of such incidents so at that level that's negligence now I'm not a lawyer so I'm not going to argue

about whether or not it's criminal but there's lawsuits being filed in the last couple of days on this issue specific and that's the way I've come to think about it there is always a human in the loop So we have like a 12-step program for harnessing it safely. Originally, step one was, oh, make sure there's a human loop. And then actually became step minus one or step zero, because we just start with there is always a human in the loop. And step one is start by thinking about when and where is the humans in the loop. Yeah. You know, I think a lot of times in Silicon Valley or the tech industry, you know, I sort of use Silicon Valley in terms writ large, right? But oftentimes we use all this, you know, we use different acronyms and we try to confuse people. People get really confused. But, you know, when you boil things down, you still exist within the cultural legal reality that we all live in.

So, for instance, a different thing, but more on the predictive analytics side of AI, for instance, you know, we talk about how great it is that you can go into a store and get personalized pricing on goods. Right. And that sounds great. But when you really break it down and what people are starting to realize, which I talked about in my podcast for hours, I mean, that's just basic old school discrimination. that's you walk in my store and I'm like you know I give different pricing to African Americans in my store or to women or to you know men that's how I do it and it's not legal right so you can couch in these terms of can I just pick up that question it's not legal right certain cases it's not legal but it depends a lot of where you are in the world and and I got into looking at this question about price discrimination first degree price discrimination

is where you'll pay the same price right? Second degree is where I give you different prices for objectively reasonable criteria student discounts third degree price discrimination is I discriminate because your name's Jason in some parts of the world third degree price discrimination is completely legal now I was working with Vodafone a giant telecoms company and their head of strategy was India and so he was doing a lot of travel back and forth between somewhere in the world mostly London and India and he used to sit there with his fancy Mac laptop on the corporate network and he'd pull up an old no name computer and run Chrome and do it in India and he'd look at the flights and he'd look at how big the difference was.

And he would typically pay 25% to 30% less for buying it in rupees at the Indian end of the transaction. Which brought that home to me. Any culture, and talking about travel, any culture where you go in and it's a barter culture at the store level is going to be just ruin a toy at its core. Yeah, you walk into a bazaar, and I'm going to get a different price as a tourist than somebody who's going in there for, you know, who's a local, that's for sure. Yes, yes. So you were asking the question about how to think about this. Yeah. As a user, there's two things you can do if you're harnessing it. One is vote with your wallet, right? Make your choice. All of your choices should be informed by assessments of risks and ethics. Because you've got to take that into consideration. So, be careful what you choose to buy.

That's the first one. And the second one is, at the end of the day, and I'm going to come back to this thing when you think about strategy, you're not differentiated by the tools you use that are built by other people. They're never going to be a source of competitive advantage. and at the end of the day it's you own how you use them and I'm a big advocate that businesses should be thoughtful about this for themselves I don't think we can rely on governments all of the time because A, they're not necessarily well motivated, sorry to be honest about that, and B this is hard enough for people who really understand the technologies there's very very few governments I can count them on the fingers of one hand who have the competence at the policy level to understand what the things to do are. So it's incumbent upon you as a user, a harneter of technology, to be thinking about this as well. Because that's where the overwhelming amount of usage and value creation will go.

So that's a good segue, actually, into sort of this other work that you do, and we'll return to the responsibility part as well, and sort of the use of agents and that kind of stuff. So right now, you know, the AI transformation and AI deployment and, you know, really the growth. If you're in the technology industry, you're hoping for, you know, mass adoption at the enterprise level, right? You know, up and down, you know, businesses up and down the supply chain. And a lot of CEOs and C-suites today are, you know, trying to harness AI. But, you know, if you look at sort of the data behind it, Um, you know, there's, there's this, there's this problem with, um, really extracting a lot of value at this point from AI, um, you know, and for a lot of reasons. And, you know, we can talk about, you know, pilots that go from the pilot stage and they never get deployed. We can talk about, you know, AI deployments that fail and don't live up to expectations.

We can, you know, there's sort of a wide range of outcomes, and a lot of them are not extremely positive. You know, and part of that, you know, comes down to, you know, actually I think a large part of it comes down to, you know, not understanding sort of the transformative ability and opportunity around AI, right? And they think of it as just another SaaS software. They think about it as, you know, just a sort of another IT initiative as opposed to rethinking their business. They don't get organizational leadership alignment around it. And there's this disconnect between what people like you who are out there talking about this all the time, what you imagine, you know, AI can do for a business, right? and know, you know, because you know the technology, and what CEOs and other C-suite leaders are thinking.

So, like, how do you, you know, from your purge, how are you seeing what's going on, you know, sort of the issues, the problems, the opportunities, you know? I'm going to talk about a real experience we just had. we're working with a client who's trying to radically transform how you build single family starter homes and we had a quarterly meeting two or three weeks ago and these are incredibly good engineers but they were still they'd been using open AI and ChatGPT to some extent we had a discussion there about what you can go do and challenge them to some extent to go away and

take on some of the challenges we'd identified and I was having dinner with the CEO and founder Sunday night in California and he said it's the most radical transformation he's seen in an organization in its entire life, because in the course of two weeks, the way that everything was getting done. And it was this, oh, you don't know how powerful it can be until you've tried it, until you've tried it in certain ways. So I think there's three parts to how you do this. The first is, you've got to experience it yourself. So if you're, and that changes people's perceptions. I'm going to draw an analogy with a smartphone. and I knew how many people, so many people who said, oh, I won't need a smartphone, right? And then they saw someone using one and they got infected and it just spreads like an epidemic. It's actually called in the academic literature epidemic adoption where all you need is the awareness of the power.

So you've got to get over that hump. You've got to get people aware of it. In my own organisation, we've had a couple of people who were deeply sceptical for a long while and then somebody sat down with them and literally started walking them through what they could do and was like, aha, those moments when you first picked up an iPhone and went, wow, that's really cool. And that's the sort of the groundwork. There's then some stuff which you look at and you go, well, we have this horrible problem, right? This takes a huge amount of time and cost and that's the low-hanging fruit. And in attacking the low-hanging fruit, you will get benefits. and the barrier to adoption for that has plummeted. I went back to actually one of the presentations for Vodafone about sort of what the limitations were and some more recent work just a couple of years ago which said look, you know, the minimum cost of taking on certain challenges is still in the tens of thousands of dollars but it's not anymore.

It's in the thousands of dollars or the hundreds of dollars because the cost of inference and the ease of use has accelerated dramatically. I mean, it's falling at, I can't remember, was it 48% or 68% per quarter? So the unit cost is just hurtling down. There's a great report out from EPOC AI. So the second thing is those low-hanging pain points, attacking those. But the really big payoff, beyond the significant payoffs from these two things, is when you reinvent the way you do business. and you know the lessons we learned from steam going to electricity from punch cards going to proper IT were that it's not until you look at how the technology changes how tasks get done and you reorganize around what the technology can do

and that led me to thinking well if that's the case but in order to reinvent it you've got to think about what your business strategy is and I'm going to say something which I think is really important no one is going to be able to differentiate their business or have sustainable advantage merely because they're using AI, that's not going to help you doesn't matter how good it is, make slightly better choices about it this as an input is available to all of your competitors you win because you use it better it's the choices you make and that brings you back to a question about strategy which says well if a radical new strategy technology has come along which is extremely powerful I should actually probably be taking a step back and thinking about so what could we do differently is there ways in which we can transform our products and services are there ways in which we can

change the customer experience fundamentally and how would that affect our business? We're working with a chip company and it's about how they do that, how you, in their case as a chip company, how you integrate AI into the products and services. So, very, very interesting times. It's really interesting. You just touch on, you have such great insights and you touch on a bunch of great stuff And I love the electricity thing because, you know, again, because people have been talking about, I mean, we've been talking about this for so long in the AI community, right, that we assume people understand these analogies and these historical references, right? We're like, oh, you know, like electricity. But what people don't realize is that when we say that, I think a lot of CEOs say, well, yeah, yeah, yeah. So in fact, in your factory, once you just sort of put light bulbs in, then you could work longer. and so you got additional productivity. And it's like, no, no, no, no.

The way the factory floor was laid out became fundamentally different. That was the way steam did it. I'm not a manufacturing person, but the way steam did it, it was centered around certain areas. I don't know. You had a great big engine and a whole bunch of rods and belts and clutches and all of that. And that means all the machines had to be lined up and made out. Oh, electricity? We can have smaller ones. we can turn them off really easily we can orient them in different ways we can have several engines and they can be at right angles to one another and then at some point I went oh we're still designing factories for steam and then some people said well if there wasn't any steam how would I go do it now that took decades people debate but we're talking 50 or 60 years I think the situation here is similar in terms of analogy but the pace is different so fast so if you're yeah go ahead sorry

I was just going to say there's a quote from what's his name William Gibson which is the future is already here it's just not yet evenly distributed I keep that in my mind every day and when I get outside the bubble that is MIT or the west coast or London where there's a widespread awareness of this the future is here and what I see being done in the businesses who are advanced and have been out there for a while is incredibly powerful incredibly powerful real world results in terms of productivity and responsiveness and all of those sorts of things observable benefits what the piece about getting from here to there clear strategy understanding of the power and the limitations of the technology good work to manage the risks

But it's, although it might seem, if you haven't experienced this, rocket science, we're now far enough down it that there are some really good guidelines about how to do this well. You know, it's interesting. So when you touch on that, because, you know, And I think it's important because at the moment we're in right now, right, you know, if you're a CEO of a Fortune 500 company, right, or in the season, I mean, you've been. You've been. I wish I was. They get paid a lot of money, I gather. They do. They do. But, you know, they've been, they go to a million conferences. They have McKinsey and other consultants, you know, walking out of their office every day. It's just they've been talking about this for 10 years. Right. But it's interesting because what I keep coming back to and part of it's just the bubble we live in in life is that in theory and sort of the optimist side of AI is that once you put this sort of intelligence layer that's available to everybody.

Right. You can you can tie in. There's millions of agencies. Anybody can access it. you know, price is compressed, fairly cheap, you know, more commoditized than it is today, but we're getting that way, that it allows the individual, like ultra small companies, but, you know, especially mid-sized companies to compete against much larger companies if they harness this properly, right? You know, it isn't that large companies are going to win because somehow they have better data and, you know, that may be true, but a lot of times, a lot of data that people, you know, rely on anyway is unstructured data that's, you know, being created all the time. So, you know. Well, I actually think the smaller firms have the advantage here. So I'm sitting in, actually, on a beautiful evening in Napa Valley, California on Tuesday night and I sit down, I run a boutique consulting firm and I sit down next to a senior partner from one of the big incumbent consulting firms.

And he's talking about what they're going to go do. And I said, this is one of those moments when scale is actually an encumbrance rather than a source of advantage. It's competitive disadvantage. Because you've got to deal with all the coordination problems, whereas the optimum sort of scale is if you're less than 100 or 150 people, you're like 20 or something people then we can harness it, we're big enough to be able to have some embedded expertise ok, we're a little bit skewed in our particular case, because I've got three guys on staff who have PhDs in AI and we're in Kendall Square so, at the heart of MIT but, leaving that putting that aside putting that aside, this construction company, right turned on a dime and literally over the course of three or four days changed how they were running the business

now you can't do that if you have 100,000 employees you should be trying to move more quickly but the little companies can go do it so one of the questions that we were asking was like who will be the big deployers of teams of agents and doing recursive self-improvement I think that the really big businesses will actually be laggards and what will happen is in a lot of times is smaller businesses will figure out how to do it and then come to that and go hey you know this problem you've got we've been solving this for a bunch of people the way we do it is with this and they'll go oh we could do that ourselves well you might but it will take you a lot longer and that you will see a huge surge of entrepreneurial growth around smaller businesses which are just big enough to be able to harness it effectively, but small enough to be able to move quickly. The mid-market.

Well, it's like, you know, it's a small end of the mid-market, right? I completely agree with you. If the thesis is true, right, that X amount of knowledge work will be done by agents, by white agents, by whatever that is, you know. Everybody can argue over the timeline and, you know, how much and all this other kind of stuff. But let's say it's 20% or 30%, you know, sort of at the low end, right? Well, as a CEO or whoever, you can see that as a cost-cutting measure, right, and say, oh, I'm going to do this and, you know, save money. I'm going to fire people and, you know, get headcount savings. But if you look at it from a small incumbent in the space, basically what you're doing is for very close to free or cheap, if you will, you can hire and create thousands of employees, right? And so a lot of it is, and consultants don't make nothing against consultants,

but consultants don't make money saying this. But the real question becomes when, you know, we talk about, you know, when's the first one-person billion-dollar company or whatever, right? But it really is saying if you could start an insurance company your financial services company today, right, from scratch, just you and I sitting here, with what we know, how would you design it, right? How would you harness that? And you should be able to scale almost infinitely, right, because, you know. Well, certainly extraordinary. Well, there's some things for which, sorry, so there's two things I want to qualify that with. there's a bunch of stuff where human judgment and creativity still has a vital role to play there's a bunch of activities where actually you have to have people you need face to face stuff but we are thinking about this in the context of scaling a construction business the core who are developing the technology

you can keep really small that's low tens of people but if you're going to roll up nationwide you are going to need boots on the ground but you need fewer of them and their time can mostly be spent doing face to face stuff which people are still not as better at oddly enough while augmenting their capabilities and automating the tough stuff the way I've started to think about it is there's a couple of pieces well there's the sort of what do you do operationally we immediately work on ways to build broader understanding about what is possible take the low hanging fruit and then you want to be thinking about reinventing but that means you've got to do some work to think through how are we going to use this to create more value than our competitors how are we going to reinvent what we do because it's got to be aligned with the economics of your business

and the strategy now you may be able to change those very significantly that's certainly happened with the steam to electricity change, but it requires that tariff of this is what we're valuable, this is how we differentiate, this is why people choose us, and I would be coming at this very strongly, not from the cutting cost stage, because I don't think that's going to enable you to win someone else will cut deeper to the bone and do a better job of it but if you come up with ways of making people's lives better, saving them time. Yeah. Which is one of the insights to do. Go ahead. And just redesigning people's role within the company. I always use an example like a lot of us do. I end up working from home a lot. You're supposed to have me. And then it's easier. You run late. You hop on Zoom. And so besides this kind of interaction

I end up walking to my local store to buy dinner every night. It's actually human connection for me. So I know the checkers, you know, one of those things. You know, if they automate that, you know, just the checkout process, they should still have, they should, instead of just firing those people, right, and eliminating the cost, they should redeploy them, you know, as greeters, help you think. Put them out there in the aisles of the store to coach you on the stuff. Exactly. you know it's like one of the I will confess to shopping at Whole Foods because there's a bunch in my part of the world one of the things about them is no two of them are organized exactly the same and I'm walking and going okay I was making shrimp tacos for my lord and it's like why are the tacos nowhere near the salsa nowhere near I'm keeping asking where do I find this right yeah So I think there continues to be a vital role.

We're looking at fundamentally different business structures. We've been through this. As we looked at what we're doing in terms of admin, we used to have multiple executive assistants or personal assistants. And we now have one person for those sorts of roles who covers a very broad range of things that she does. She was actually, she was my EA at one point, and she went off and did some interesting other jobs and came back and worked for us again. But a role which is sort of like, you know, broadly supportive business operations, whatever that takes, and a lot of it about, oh, how are we going to harness this? So I like how we have a checklist that here are things which, as a small company we have not yet automated, these are the skills we're going to create to do a whole variety of stuff and I was doing some accounting looking at the use of my time

over the last year or so like hours on different things, which by the way I could only do because I could use Claude to go mine every Slack message I'd sent every email I'd sent or received, every meeting I had attended every change I'd made to a document. That consumed some tokens. At the end of it, and you could see the pattern, that the amount of back office admin work has come down significantly. Tremendously. So if you were, like, you were. So we've talked about the big companies, right? We've talked about the small companies. Most of the companies sit in sort of this mid-market phase, right? They have 200 employees, 300 employees, 400 employees. They're 300 million, 500 million, whatever, right? If I was to drop you today into one of these companies, right, what would be the first thing you'd want to do to get that company on a successful path to AI transformation?

Well, am I allowed to start with it depends? Of course. It always depends. First of all, it depends. And that's actually where I begin. first I want to get a really good understanding of the reality of what the use of AI was and the awareness of and the understanding of it at the coalface and the reason is that provides a foundation for what you can build on and what you need to overcome and then I'd actually kick off oh I think three things in parallel and you go, well why three things? Well, because assuming there's at least some pockets of this so I'm going to, because you know, we'll assume there's some people in there who even if it's just because they've been doing it personally and they haven't yet been allowed to bring it to work so you've got to find those people I'd work on creating a broader understanding about what you can do with people

this widespread, but even before that I'd go, what's our governance mechanism? what policies and harnesses and real-world practices are we good in place to manage risk and ensure that we do the right thing. And I do that right at the outset, because I think that's really important, so things don't go wrong. Secondly, I work on creating broad awareness and understanding. The future's already here, it's just not yet evenly distributed. Try and bring people up to speed. Thirdly, find some that people already know enough about this and say, just go after these things that everybody hates. Most businesses have a process which they hate. In our case, it used to be invoicing. We've transformed that process, right? So it's basically press a button and stuff happens. And the last thing, which I would begin work on as soon as you can, was take a step back and think about strategy and reinvention.

there is the potential for the business to perform in radically different ways and create tons of new value so don't just cut the costs, the real competitive advantage the real value creation, the real things that transform the economy will come from the growth side and it's go hunting for those now, if you're really clear about your strategy and it remains valid with the advent of AI then that's easier to do but it may mean you need to question some of your strategic assumptions so strategy and reinvention so governance broadly awareness hits and costs and you know you say that's a lot to do yes you sequence it a little bit you overlap but that's my that would be my road map what do you think like you work with a lot of companies right how many companies do that effectively effectively a minority which part do they fall down on most

out of those three the first the one they fall down depending on their size, the bigger they are the more likely they are to be too conservative in governance, they try and eliminate all risk as you get to mid-market and smaller that's less of an issue the second one they fall down is just failure to think about reinvention, failure to think about creating new value so much of the discussion out there is about automation rather than augmentation and about saving money one of the things that led to a set of charts actually which is about, was a really annoying use of the phrase digital transformation McKinsey was talking about it and then when you read what they had under the basket of digital transformation it was, oh it's business

as usual, we're just going to use digital technology to save a bit of time and save a bit of money that's not transformation transformation transformation comes about when you build genuinely new products and services when you build new businesses so it's either slowing it because of the full horse movie AI apocalypse. IT won't let you have the computers. Legal won't let you sign that contract. AHR won't let you hire the right people. And budgeting won't sign off on your financial projections. Well, you've got to guarantee this level of return. You can't guarantee the return. You can't go do it. Get past that, particularly in a big business. Put in place appropriate governance. And embed it in what you're going to go do. and then don't just do the cost saving. Start thinking about creating new value. My motivation my whole career has been this deep belief in the power of technological innovation. We're sitting here doing this because of technological innovation.

We need more of it. We can do better on that, but do it in ways that are responsible and ethical and humane. You know, it's interesting because, you know, and I love how you lay that out obviously from your deep experience But you know and I think you touched on something there which I think is really important in this discussion especially for business leaders who are listening to this is that it you can help me frame this right You'll do a better job of this than I will. Part of the problem that I see at companies, right, is this idea of automation cost savings versus augmentation. So what you end up with is sort of two things, unclear strategy at the C-suite level, right? And that's not communicated properly throughout the organization. And this sense of, and out of that, you know,

unless your whole message is we're going to automate everybody and, you know, start looking for a job. And out of that becomes this fear among the employees in the organization. And unlike other quote unquote digital transformations or whatever people call it in the past, you know, you know, transformation, employees are the ones that are going to drive this because to build, agents, agentic workflows, you have to have the people on the ground who understand, you know, who basically put together those workflows. Because half the time, as you know, when you go in and ask somebody, hey, what's the workflow? They're like, well, what I do is a little bit on the computer, and then I yell, hey, Jason, make sure you do, you know, it becomes a very personal workflow within the organization. And that's really where the power is, unless you're going to do something like automate invoicing, right, and some of these other things. But to get to this sort of agentic workflow, which will allow your employees to move up the value chain into higher value work,

you need them to come aboard. And that's where I see, I mean, there's stats that like, and there was a stat, I forget, maybe PwC, where like 30% of employees are actually actively sabotaging their AI efforts within an organization. how would you how do they do it how does this change well first of all let me can I take this in two parts over the course of the last two or three years we found ourselves leaning far more towards augmentation than just straightforward automation and if anything we think that there is too much automation in some cases how can that be well let me just pick the sort of the classical customer service problem I break it into three things it's like one if you're going to automate something it should be superlative right but it should also stop and go

hey I don't know how to cope that I'm going to hand you as quickly as possible rather than trapping you in this conversation where you should be the moment you say agent it props you to an agent but that agent's capabilities are significantly enhanced. T-Mobile have done some great work in this area where they actually reorganized groups of agents serving groups of customers. So you're much more likely to talk to somebody for the second time or third time or at least somebody in the same team. So more augmentation and then free them up to make human judgment about just the exceptions that they can create these moments of magic. And the irony is, I actually think that these capabilities made people relatively speaking more important and the statistics say that if you actually make a job more skilled well aggregate demand for it goes down somewhat but wages go up and people go oh so I could actually because you're doing something more valuable

but if you maintain the same level of employment because people worry about whether it's me or someone else that there's a real leadership challenge, you've got to grow the business. You've got to create more value. So these pieces come together. People become more important. Augment rather than just automate in many cases. And if you have a clear strategy about how you're actually going to grow the business, create more possibility for work, I've been impressed, and this is a theme, the wise companies are putting great effort into educating all of their employees, which empowers them, whether or not they work for you. There's some broad gains to this. And working really hard to keep the people they've already got and to empower them. Because they're the ones with the domain knowledge. They know how your business works. They know what your customers worry about. Now, the second order implication of this,

which we've already cognizant of, and this is what worries me from a broader societal perspective is that reduces liquidity in the labour market there's less turnover of jobs now what has happened historically is there's been a period of transition before you recover that because then new businesses emerge and that's where the new jobs come from now those new businesses can be grown inside income of businesses so things that we can do to enable that new growth I think are really important initiatives. Unfortunately enough to live in Massachusetts in this context, and some people may think that's a terrible price to bear, but at least we're putting a great deal of effort into the work, manage people through that transition, and let's try and keep it as short as possible. Long run, medium for long run, I think we're all way ahead of the game. Now let's mitigate the challenges of making that transition. You know, it's interesting. Two points on that. First is that,

You know, there's also a reason, not just, you know, for some altruistic, you know, reason to keep your employees and augment them because it's nice to do for, you know, society, which is, you know, great. But also because as you reimagine your business, right, as, you know, AI, you know, makes some things so easy that they become, they're no longer a competitive advantage, right? you know, oftentimes it's going to be your frontline workers. It's going to be the people out in the field who are going to be the ones that are going to, you know, uncover those opportunities for your business, right? And conversely, you know, part of, I think, you know, back to the liability and back to the, you know, sort of, you know, problems that you have with, you know, open AI and some of these language models, you know, I can imagine, and it's happened so fast that I haven't had a lot of conversations with CEOs about it yet, that CEOs right now are saying, well, what about if agents in my company break free?

Where is the human in the loop here? So it might almost, you know, require, you know, CEOs might be rethinking, you know, their complete automation around agentic AI for some of these tasks and keeping humans in the loop. And the third point is, back to the society thing, is that I graduated in 93. week. And it was before, obviously, Netscape went public, right? And we used the internet a little bit for kind of research in college, right? There's no interface. It's kind of hard to do. And you went and talked to experts and people and counselors. You go and get informational interviews with senior leaders in different industries and ask them, you know, what should I do? Not one of them would say, well, there's going to be this thing really fast called the Internet, and that's going to create millions of unbelievably interesting, well-being jobs

that you can't imagine, right, like what we're doing right here, right? You know, and social media, you couldn't have said, hey, there's going to be a thing called, you know, cloud architecture. And so there'll be, you know, software, you know, cloud architects. And I mean, just you look at some of the best jobs out there and nobody could have ever imagined because you couldn't imagine the Internet the way it turned out. You couldn't imagine mobile phones, you know, just the whole thing isn't there. So I think I think part of it is just letting it play out. And I think people will find these jobs that I talk all the time and people are like, what about the jobs? I'm like, well, you know, they'll be there and they're like, what are they going to be? I'm like, I have no idea. And nobody in pre-Netscape had any idea. Yeah. Yeah. Nobody. And the overwhelming evidence. And I'm going to come back to myself in this case is that this is called the lump of labor fallacy.

and it's the idea that there's a finite amount of work that can be done in the economy and therefore if we automate something there's less total work well the same technology that enables you to automate that creates new possibilities and it's the seizing those possibilities and mitigating the transition the critical things to growth so I'm a long run optimist I have some real worries in the short run about some of the big actors here and their morality. I know it's going to be hard, but I've literally decided to dedicate a significant chunk of my life for the next three years into doing this. There will be, while both doing it, and actually doing all the research and other stuff. So there's going to be a doctorate on this sometime in about 2029. That is fantastic. I absolutely love that. You know, I am back on the construction thing really quick, and then I'll let you go here in a minute.

So my grandfather was a professor, and he, you know, back in the 50s, I mean, like 40s, you know, all the way through until he retired. But he was a big believer in that there needs to be this innovation within construction and housing, right? So he'd invest in, like, lustron homes. I mean, you kind of go back through these, you know, old inventions, and nobody has cracked the code on that. You know, here in the Bay Area, we have, you know, oh, we're going to do, you know, prefab and all this kind of stuff. And if we can figure that out, that would be one of the biggest innovations, I think, in our society. I mean, we're almost up to here in the Bay Area, we're at like $1,200 a square foot just to build, right? I mean, how do you build a home that's affordable for somebody? And that's not land. That's $1,200 per square foot just to build a home, you know?

And it's like. Well, we're using AI as part of the process. Yes. I can't say much more about how. Yeah, sure. The goal is to reduce the cost of the vertical bit, right, by more than 50%. While actually also making it more robust and higher quality and more resilient. some other things such as making it work where it draws less than 100 amps so that things like hookups become trivial below 100 amps you're not going to impose new loads on transformers there's a set of things that go with that and doing it in ways that for instance keep the total cost of ownership down because it's more insurable and it requires less maintenance the US

in particular has sort of, because of a bunch of policy issues, iterated its way into building stick-built homes at scale on the outside of cities in the south of the US. Spool. With all of the undesirable characteristics that go with it. Argyle, infill, at half the cost. Love it. with better quality and easily deployment. I mean, if you could do that, that would be a game changer. There's absolutely no doubt about it. Well, hey, before I let you go... That was exactly the conversation I was having on Monday morning with the CEO, where we were putting numbers on how many homes you have to build. And this is a 10-year horizon. The answer's a lot in the US. A lot. There are examples of this having been done in other parts of the world on a smaller scale.

So it can be done. Here in the U.S., we reckon it will take a decade. I mean, it's just, it's such a, you know, for so many people think it's an intractable problem, but I just don't believe that. I think there's, you know, people like you are working on it. I think it'll be, it's a game changer because I think a lot of people, when they say they're worried about jobs, what they're really worried about is cost of living and housing is a key component to cost of living, right? And if homes were cheap or cheaper, right, way more affordable, that would take a lot of pressure off of people, and I think you'd end up with a lot more positive. The numbers look like you can probably, if you can do that, you can take 10 to 15 percentage points of people's budget out, and that changes people's lives. and the knock-on effects of this are huge because if you can create single family starter homes it increases the quality and reduces the price of rental homes

because this existing in the market means the renters have to up their game so it's not just that it doesn't affect luxury homes much because that's a different market and there are some places in the world where it's just hard. I mean, I live in Cambridge, Massachusetts, which is extremely dense. And we were looking at some cities in California, and they have thousands of suitable buildable lots. I then went to look for my hometown. Fifteen. There's only fifteen available lots in the whole of Cambridge, and they're very expensive. So, it's going to be... You've got to do something different. just north of San Francisco, and there's just no available lunch. There's nothing. So that won't work here, but, you know, you know. But there is info. Like, we're trying to push ADUs and stuff like that for people's properties,

which, you know, it's kind of in, if you will. So, hey, before I let you go here, because I know we're running up against time, give me some last thoughts. Give the audience some last thoughts. Like, what do you want people to take away from this conversation, you know, both in terms of responsibility in AI and in terms of people who are, you know, looking to deploy AI in their company to sort of transform it. I'd start with that one at the beginning. The future is already here. It's just not yet evenly distributed. These are technologies which are already mature enough and powerful enough that there is no question that you can, should, and must be harnessing them. but it's about the flip side of these very flouser technologies it's about the people so this is people will become more important, they'll become more of a bottleneck that's actually the thing, unleashing human potentials so that people aren't wasting their time on work that can be done more easily by other people

and the third thing would be the oh this is an Alan Kay quote the best way to predict the future is to invent it really think about what you can do to make life better for your customers and at the same time make it better for your employees because you can do both of these things at the same time when they're done well and that's what we need, we need more value to be created we need to be using these not just to automate work but to create more better work I believe that's doable in fact I know that's doable it's doable at scale already well thanks for joining me today thank you, thank you so much for the invitation