# 295. AI, strategy, and the future of work: What business leaders need to know

- **Show:** Inside the Strategy Room
- **Published:** 19 Mar 2026
- **Source:** https://omny.fm/shows/inside-the-strategy-room/295-ai-strategy-and-the-future-of-work-what-business-leaders-need-to-know
- **Transcript:** https://podcasts.opennash.com/shows/inside-the-strategy-room/295-ai-strategy-and-the-future-of-work-what-business-leaders-need-to-know-c42716/
- **Provider:** OpenNash Podcasts (machine-generated)

## In brief

Most companies are failing to get real value from AI because they treat it as a sales-and-marketing add-on. Oxford economist Jean-Paul Carvalho argues the real payoff comes from restructuring workflows and cutting bureaucracy. Because AI is a general-purpose, globally scalable technology that automates thinking rather than muscle, he expects it to disrupt business more deeply than earlier technology waves.

The main tension is between short-term efficiency and long-term capability. Studies show AI helps individual workers, and overall employment effects look small so far. But hiring of junior workers is falling, which threatens the pipeline of future senior managers who must oversee AI. Carvalho says the outcome depends on whether AI is built to replace people or work alongside them, and on political choices.

## Key insights

1. **Most AI failures are strategy failures** Carvalho cites an MIT study finding 95% of gen AI implementations failed, and says the cause is strategy and execution, not technology. About half of AI budgets went to sales and marketing, while the bigger value lies in restructuring workflows.
2. **Junior hiring, not firing, is the visible damage** One study found a 16% employment drop for US workers aged 20 to 25 in AI-exposed jobs. Overall employment and wages in firms have barely moved, so the effect shows up as fewer entry-level hires.
3. **Cutting junior hiring breaks the senior pipeline** Humans will need to supervise AI agents for the foreseeable future, and those supervisors normally come from the junior ranks. Firms that stop hiring juniors now will lack experienced overseers later.
4. **Market power can vanish faster with AI** AI can produce cheaper substitutes, and new entrants can scale without the usual hiring limits. Carvalho says established CEOs will need to be nearly as nimble as startups and willing to restructure radically.
5. **AI's current direction favors replacing humans** Model benchmarks usually boast about beating humans at a task, not about making humans plus AI better. Carvalho, drawing on Acemoglu and Johnson, argues that human-complementing AI would be more humane and would keep skills and training alive.

## Questions this episode answers

### Why do so many companies fail to get value from generative AI?

Carvalho cites an MIT study saying 95% of gen AI implementations failed, and he blames strategy and execution rather than the technology. He notes about half of AI budgets went to sales and marketing, when the bigger gains come from restructuring workflows and trimming bureaucracy.

### Is AI actually taking jobs right now?

Not much overall, according to studies Carvalho cites: effects on total employment and wages in the US and Denmark were small. The clear impact is on young workers, with a 16% employment drop for ages 20 to 25 in AI-exposed jobs, mostly because fewer are being hired.

### Why is AI different from earlier technologies like the steam engine?

Carvalho gives four reasons: it automates thinking rather than physical work, it is a general-purpose technology, it can be delivered globally over the cloud, and it can generate new knowledge with national security implications.

### What is the 'pipeline problem' with junior hiring?

If companies stop hiring entry-level staff because AI can do their tasks, they will have no experienced people later to oversee AI agents. Carvalho says firms will have to hire and train juniors despite the short-term incentive not to.

### What is the difference between automation and augmentation in AI?

Automation means AI replaces human work on a task, while augmentation means AI is built to make people more effective. Carvalho says AI development currently leans toward replacement and argues for a human-complementing approach.

## Facts

- **Jean-Paul Carvalho** [0:03:44]: "We were forced to come to grips with it in November 2022, when Chat G P T was released, but still there remained some heads in the sand."
- **Jean-Paul Carvalho** [0:04:25]: "A little while later, in July 2023, Pew did a survey, and in that survey, sixty two of respondents believed that A I would majorly disrupt the workforce, but only 28% believed that their job would be affected."
- **Jean-Paul Carvalho** [0:03:11]: "The world we're in today has really been shaped by the Industrial Revolution. This is something that began around 1760, the ability to manufacture goods cheaply and at scale."
- **Jean-Paul Carvalho** [0:09:52]: "Eric Brynjolfsson and co-authors looked at a staggered rollout of AI voice chat assistants amongst customer service customer support agents in a Fortune 500 firm."
- **Jean-Paul Carvalho** [0:09:52]: "the customer support agents that had access to this AI technology, this this uh c AI systems, resolved 15% more cases per hour."
- **Jean-Paul Carvalho** [0:10:27]: "Jebarian and Henkel look at a field experiment of 70,000 job applicants who are randomly assigned to an AI voice recruiter, human recruiter, and then a third subset is given a choice."
- **Jean-Paul Carvalho** [0:11:21]: "When the individuals are given the choice between the AI recruiter and the human recruiter, 78% choose the AI recruiter, and there's no difference in satisfaction after these interviews between the human and AI."
- **Jean-Paul Carvalho** [0:12:29]: "Daron Acemoglu and co-authors have looked at the aggregate effects of AI adoption in the US, and they find little change at the employee or at the rather at the occupational or industry level in employment or wages."
- **Jean-Paul Carvalho** [0:13:43]: "The industrial software I'm speaking about is really uses real-time sensor data to predict machine failures to optimize maintenance, help workers use the machinery."
- **Jean-Paul Carvalho** [0:14:19]: "Eric Brynjolfsson and co-authors again look at US workers aged twenty to twenty-five, and they find 16% reduction in employment amongst for AI-exposed occupations."
- **Jean-Paul Carvalho** [0:17:14]: "I was speaking to a tech founder not long ago and his new employees are regularly sent deep fake videos of himself asking them to click on links for training but these links are not training links as you can imagine"
- **Jean-Paul Carvalho** [0:19:14]: "There's an MIT study which shows that 95% of gen AI implementations have failed, 5% have been successful adoptions."
- **Jean-Paul Carvalho** [0:19:14]: "About 50% of these AI budgets have been for sales and marketing."
- **Jean-Paul Carvalho** [0:22:09]: "Sun Microsystems stock price went up 6,400% from the summer of ninety four to 2,000, early 2,000. And then it crashed ninety to 95%."
- **Jean-Paul Carvalho** [0:22:51]: "the release of DeepSeek's LLMs, V3 in December 2024, R1, January 2025. These models developed frontier capabilities on a shoestring budget"
- **Jean-Paul Carvalho** [0:22:51]: "supposedly without the advanced NVIDIA GPUs whose export was restricted under the Chips Act."
- **Jean-Paul Carvalho** [0:29:00]: "They make this distinction between a kind of substituting AI technology and human complementing AI technology."
- **Jean-Paul Carvalho** [0:29:26]: "which is really w proposed by Alan Turing, right? How do we develop something that substitutes for human? But there was an alternative line of work developed by Norbert Wiener"

## Transcript

**[0:00:02] Jean-Paul Carvalho:** There's some evidence that the latest, the early trials of gen A I in corporations have mostly failed. There's an M I T study which shows that 95% of gen A I implementations have failed. But that, I don't think is a matter of the technology, it's a matter of strategy and execution. About 50% of these A I budgets have been for sales and marketing. But really, where the value creation can come from is from restructuring the firm, changing the workflows and the processes, and slimming down the bureaucracy inside every large organization.

**[0:00:41] Jean-Paul Carvalho:** This is where huge value can be created.

**[0:00:45] Sean Brown:** From McKinsey and Company, I'm Sean Brown, and welcome to Inside the Strategy Room. While artificial intelligence is moving rapidly from experimentation to execution, many companies still struggle to capture real value from the technology. To explore how C E O s can more successfully scale A I, as well as the implications of the technology for the future of work and organizations, we're pleased to share this conversation with Professor Jean-Paul Carvalho. Jean-Paul is a professor of political economy at the University of Oxford's Department of Economics and director of Oxford Elevate, the department's executive education profile, which includes the Economics for Business Leaders program now in its seventy fourth year.

**[0:01:30] Sean Brown:** Jean-Paul is joined by Robin Nuttall, an expert partner in our London office and a leader of our regulatory strategy and environmental, social and governance work. Robin is also a business fellow at the University of Oxford's economics department. Together, Jean-Paul and Robin will explore what makes A I different from past waves of technology. What the latest data tells us about its current and future impact on labour markets, how junior hiring is now under pressure and the longer term implications for talent pipelines.

**[0:02:02] Sean Brown:** And finally, what CEOs can do to scale A I more successfully, not through pilots, but through true organisational transformation. We hope you enjoy the discussion. Now here's Robin to start us off. Jean-Paul, it's terrific to have you here today. And of course, our topic is artificial intelligence and what are the implications for the nature of work, for the organization and for the business leader. But before we dive into the content, I'd love to understand how did you personally get excited about the topic of uh of A I?

**[0:02:38] Jean-Paul Carvalho:** First of all, thanks for having me, Robin. This is fantastic. How did I get excited about A I? Interested in it? It's apparent that we've been living through a technological revolution for some time now. And as an economist, I'm naturally interested in it because technology is the main engine of economic change. What I find is that the people that I know are shocked by the pace of change when it comes to A I today, and they're somewhat terrified by it.

**[0:03:11] Jean-Paul Carvalho:** And why shouldn't they be? I think most of us, and most of the listeners of this podcast, will have grown up in a very stable and prosperous environment. The world we're in today has really been shaped by the Industrial Revolution. This is something that began around 1760, the ability to manufacture goods cheaply and at scale. This is the industrial revolution, and it leads to cars, suburbs, advertising, consumerism, keeping up with the Jones's, and much of what we associate with modern life.

**[0:03:44] Jean-Paul Carvalho:** But now we find ourselves in a time of change. Until relatively recently, what would you advise your kids? People would advise their children to train for professions like medicine, law, accounting. These are professions that are formalized in the nineteenth century and date back much further. So we've lived through an unusually stable environment, and that is changing now, and it's taken a while for people to come to grips with it. We were forced to come to grips with it in November 2022, when Chat G P T was released, but still there remained some heads in the sand.

**[0:04:25] Jean-Paul Carvalho:** A little while later, in July 2023, Pew did a survey, and in that survey, sixty two of respondents believed that A I would majorly disrupt the workforce, but only 28% believed that their job would be affected. So this is the kind of fear and trepidation that we encounter when it comes to A I. But there's a way to understand it. Now, I'd like to go back a little bit further than that. Actually, in reality, a lot of the changes predate November 2022.

**[0:04:57] Jean-Paul Carvalho:** The changes began some time ago. In 2012, I gave a wedding speech in which I proposed that A I and robotics would replace marriage. Now, the lesson here is that you never invite an economist to give a speech at your wedding. You never do. This is tongue-in-cheek, of course, but nevertheless, it was already apparent then. I was living in California at the time. The driverless cars were being trialed on the roads. AlexNet was out. So there was a transition already occurring from symbolic AI to the deep neural nets that we see today.

**[0:05:36] Jean-Paul Carvalho:** You had software engineers training AI agents that would code. You had AI based online job vacancies skyrocketing around 2015, 2016, and by 2018, I was incorporating it in my courses, my graduate course in political economy. And today, there's no escaping it. I am at the moment, in addition to being professor of political economy at Oxford, director of Oxford Elevate, which is the executive education program run by the Department of Economics. It's probably the oldest executive education program in the world, potentially per perhaps, but as far as we know, it's run for over seventy years.

**[0:06:21] Jean-Paul Carvalho:** And the executives and the firms that we interact with all want to know about A I. This is front and center for them. And so this is a great time to speak about it. And so, Jean-Paul, an initial provocation would be, as you say, there is a long history of automation. It's at least since the beginning of the Industrial Revolution with mechanization and the steam engine. You could argue it goes back 2,000 years to watermills and and windmills.

**[0:07:00] Jean-Paul Carvalho:** Do you think the AI revolution is novel or just another milestone in the long history of automation? I think there is a combination of novel factors that could mean that the consequences of the AI revolution are for far more wide-reaching and profound. We don't know yet, but here are the here's here are the factors. First of all, AI automates cognitive tasks, not physical tasks. Cognitive tasks are what humans are good at. Physical tasks are what they're not really good at.

**[0:07:21] Jean-Paul Carvalho:** A silverback gorilla that shares 98% of our DNA can tear us limb from limb. But we're the only species that has cumulative culture that can build knowledge generation after generation on top of each other. And so it is our unique cognitive skills that have set us apart from everything else on the planet, and that's potentially not the case anymore. Secondly, AI is a general purpose technology. No such general technology was available in the industrial revolution.

**[0:07:54] Jean-Paul Carvalho:** The industrial revolution automated very specific things. Thirdly, AI is globally scalable. It doesn't matter where the companies are really are located, that you can access these AI services on the cloud or remotely, wherever you are, and the companies can be concentrated in a very small parts of the world, Silicon Valley, China, et cetera, and they can service the whole world. And finally, AI will generate new knowledge, new capabilities, both offensive and defensive, and really, as such have national security implications, of the sort that nuclear energy did, but nuclear energy is a very specific technology, not a general technology.

**[0:08:42] Jean-Paul Carvalho:** And so AI, this com these combination of factors really makes AI, I think, unique, or at least rare, and make the implications possibly quite different from previous technological revolutions. So I think if we then dig in a bit more into AI, Jean-Paul, and the way that we think about it, just very simply, is the kind of three waves of AI, machine learning and predictive technologies and so on. Second of all, the generative AI wave, and then most recently the agentic wave.

**[0:08:50] Jean-Paul Carvalho:** You could arguably have a fourth wave of physical AI with robotics and so on. But if you if you take these waves of AI, they are obviously starting to impact the labour markets, the talent that the the company needs, the the shifts that we're seeing within labour markets, wage compression. Could you share what you're seeing in terms of the latest empirical evidence on how is AI impacting the the labour markets? So it's early days. We have to rely on evidence that's just coming out and some of which is unpublished.

**[0:09:52] Jean-Paul Carvalho:** What we see first of all is that it's clear that AI systems can improve individual productivity. This is at the individual level. For example, Eric Brynjolfsson and co-authors looked at a staggered rollout of AI voice chat assistants amongst customer service customer support agents in a Fortune 500 firm. And what they found was that the customer support agents that had access to this AI technology, this this uh c AI systems, resolved 15% more cases per hour.

**[0:10:27] Jean-Paul Carvalho:** We see similar improvements in productivity when it comes to coding, various forms of software development, et cetera. But as you mention, as AI agents, which is the next stage advance, the technology becomes far more substitutable for human labor. And really, at the individual level, the task of the individual is to make human plus AI a better value proposition than just AI. And there's already evidence that we have full substitutability in certain tasks. For example, Jebarian and Henkel look at a field experiment of 70,000 job applicants who are randomly assigned to an AI voice recruiter, human recruiter, and then a third subset is given a choice.

**[0:11:21] Jean-Paul Carvalho:** The AI recruiter does better than the human recruiters in terms of job offers, job starts, and thirty day retention. When the individuals are given the choice between the AI recruiter and the human recruiter, 78% choose the AI recruiter, and there's no difference in satisfaction after these interviews between the human and AI. So it seems that there's a high level of substitutability, but this is at the individual level. If you're talking about at the firm or industry level, it's more complicated because there are offsetting effects.

**[0:11:54] Jean-Paul Carvalho:** And there are several effects, competing effects that are going to determine the overall aggregate effect on employment and wages. Number one, how substitutable is the technology for human labour? Two, how much of a productivity boost will firms receive? Because if they're more productive, they produce more, they're going to hire more human labour, even if there's some substitution. That's the productivity effect. And finally, how quickly will new tasks be generated that employ displaced human labour, in which humans also outcompete the AI?

**[0:12:29] Jean-Paul Carvalho:** It's too early to tell about the third effect, but we can talk about the first two effects. Okay, the evidence suggests that this substitution and productivity effect roughly offsets each other. So Daron Acemoglu and co-authors have looked at the aggregate effects of AI adoption in the US, and they find little change at the employee or at the rather at the occupational or industry level in employment or wages. They do find reduction in non-AI related hiring, but that's offset by other types of hiring.

**[0:13:06] Jean-Paul Carvalho:** In Denmark, where you have very, very rich administrative labour data, women in Vestager. Look at the changes in hours worked and earnings in firms with the widespread adoption of gen AI. And what they find is very little change. But what they do find is changes in occupational switching and organizational restructuring. And so that's really what's going on. There's a there's a lot of shift that's going on. There's a big shift that's going on within organizations. And you can see that in work, De Souza's work on adoption of industrial software in Brazil.

**[0:13:43] Jean-Paul Carvalho:** The industrial software I'm speaking about is really uses real-time sensor data to predict machine failures to optimize maintenance, help workers use the machinery. These were tasks that were formerly done by white-collar workers in factory offices. So what you find after the adoption of the industrial software is reduction in hiring of these workers, but an increase in hiring of the manual workers because the machines can be operated more continuously and more efficiently. And, of course, wages go down because of the composition effect, but the results are really mixed.

**[0:14:19] Jean-Paul Carvalho:** Where we have really clear results, and you've alluded to it already, Robin, is in terms of junior versus senior hiring. And that is the real difference. Eric Brynjolfsson and co-authors again look at US workers aged twenty to twenty-five, and they find 16% reduction in employment amongst for AI-exposed occupations. That's a large drop. They find it especially in areas and occupations that in which AI substitutes for human labour, not complements human labour. And it's really about not about firing, but it's about reduction in hiring of junior workers.

**[0:15:07] Jean-Paul Carvalho:** And this has been confirmed by follow-up studies using different measures of AI exposure, occupations and firms, et cetera. So this is really the main trend that's going on, and this is a big challenge to how we train workers, how we educate workers at our stage of the process, and and how CEOs manage firms. And more recently, we've seen various scenarios being painted where the dynamics that you've just described, senior junior, the AI producer versus the AI user, has the consequence of compressing wages amongst white collar and then having a an impact on spending and therefore growth.

**[0:15:57] Jean-Paul Carvalho:** Do you do you think those kind of scenarios are likely or or or credible? The main issue I see is that in these firms, humans are going to be in the loop for the foreseeable future. However, how do you get the humans that oversee the decisions and the work of the AI agents who are senior employees when there are no where there are limited number of new employees being hired? You're breaking the pipeline. And so firms will have to hire people and train them, despite the short-term incentives to not hire them, if they are to get the right senior management to really oversee the work of the AI agents within their corporation in the future.

**[0:16:38] Jean-Paul Carvalho:** And that's something that managers, CEOs have have not had to think about to date, but this is a clear pipeline challenge that they will have to grapple with going forward. It's one of many challenges that they face. And so you mentioned the managers and the and the CEOs. I'd like to pivot to that topic now. What are the implications of some of the dynamics that you've described for the CEO, for business leaders? How should they prepare for these disruptions?

**[0:17:14] Jean-Paul Carvalho:** How should they think about the value chains, industry disruptions? They have to cope with massive disruption, I think. Firstly, there's issues of trust in organizations that haven't arisen before. I was speaking to a tech founder not long ago and his new employees are regularly sent deep fake videos of himself asking them to click on links for training but these links are not training links as you can imagine so that shifts things to more things to in-person interaction and very trusted systems the trust is also if you have a divisional manager now with a lot of AI agents working under them before maybe they were managing a thousand employees now it could be as if they're managing a million how do

**[0:17:49] Jean-Paul Carvalho:** they cope with that You don't want to wake up in the morning and your AI agents have signed a million pages of contracts that you have to review and unwind. You have to have trusted systems to deal with the new with the new technology. So trust is going to be a big issue. Secondly, I think market disruption, market power. Market power can disappear faster than in earlier times. Because firstly, AI produces cheaper substitute goods and services that can outcompete you.

**[0:18:19] Jean-Paul Carvalho:** We're seeing this now with enterprise software. And the market really thinking through which firms will be outcompeted by substitutes generated by AI. But even if substitute goods and services aren't generated by AI systems they can still enable disruption within the value chain itself And the reason is that new entrants into the industry can scale much more quickly than they could in the past because they don't face the same labour constraints on scaling And what's going to happen is that CEOs of established companies are going to have to be almost as nimble and innovative as startups They're going to have to be willing to radically restructure Now there's some evidence of course that the latest the early trials of gen AI

**[0:19:14] Jean-Paul Carvalho:** in corporations have mostly failed right There's an MIT study which shows that 95% of gen AI implementations have failed, 5% have been successful adoptions. But that, I don't think is a matter of the technology, it's a matter of strategy and execution. About 50% of these AI budgets have been for sales and marketing. But really, where the value creation can come from is from restructuring the firm, changing the workflows and the processes, and slimming down the bureaucracy inside every large organization.

**[0:19:51] Jean-Paul Carvalho:** This is where huge value can be created. As we know, in bureaucracies, there are incentives that are not great, let's say. If there's no profit motive in a particular line, how do you get a promotion? How do you make more money? Well, you have to hire more workers under you. And that creates potentially maladaptive bureaucracies that grow too large or grow inefficient in various ways. AI doesn't have those incentives. And so there are organizations that will be radically restructured using AI technology and will create huge amounts of value creation.

**[0:20:32] Jean-Paul Carvalho:** Hedge funds today are trying to pinpoint the companies that are best placed to do this, or the startups that are best placed to do this. And healthcare, for example, is in the front of the line. So these are some of the big challenges, generalized disruption that CEOs will have to face going forward. So then is one implication of what you're saying, JP, that being able to scale AI as a large incumbent enterprise, be able to scale AI will be a new source of competitive advantage.

**[0:20:52] Jean-Paul Carvalho:** Because there may be some deployments of AI, which at the smaller level are relatively commoditized. The standard application of a large language model, but in terms of scaling and getting real value and excitement out of AI, that that is more of a moat, if you will, for for an organization and and and uniqueness. Is that is that how you would see it? I think that's one area which we can clearly see where there'll be value creation, I think.

**[0:21:35] Jean-Paul Carvalho:** It's the where will the profit lie in the AI value chain. That is a kind of 1 trillion dollar question that is front and foremost of in the minds of every investment and hedge fund manager. Where will it be? If we let's if we could go back, Robin, to an earlier dot com boom and the internet revolution, what we saw is that where was the value created? People thought initially that it was a picks and shovels play, that you wanted to own the servers and the routers and companies like Sun Microsystems.

**[0:22:09] Jean-Paul Carvalho:** Sun Microsystems stock price went up 6,400% from the summer of ninety four to 2,000, early 2,000. And then it crashed ninety to 95%. You know, after the dot com crash in march 2,000. What emerged from the ashes of that crash were the companies that have shaped our world today, the big platform plays, the that benefited, that built market power through network externalities and various lock in effects. So it was thought that the AI labs would have enjoy such market power that they would have a large mode.

**[0:22:51] Jean-Paul Carvalho:** It was thought that because it required such large capex and various algorithmic secret sources to training these models. However, that was somewhat blown up with the release of DeepSeek's LLMs, V3 in December 2024, R1, January 2025. These models developed frontier capabilities on a shoestring budget, and supposedly without the advanced NVIDIA GPUs whose export was restricted under the Chips Act. And so it was thought that the AI labs did not have the moat that we originally assigned to them, and that really affects their ability to monetize their services, and that really affects valuations.

**[0:23:43] Jean-Paul Carvalho:** However, as we progress toward AGI, artificial general intelligence, it could be, again, that massive capex is required, and various algorithmic secrets are required to train these highly advanced models. And so the mode could reappear. We don't know. It could be that the platforms really benefit from this AI revolution as they did the internet revolution. That could be the existing digital ecosystems, which we're all familiar with, which could be supercharged by AI. Or it could be new platforms that disrupt the existing digital ecosystems.

**[0:24:23] Jean-Paul Carvalho:** For example, we have the fear that LLMs will take people out of the existing digital ad ecosystems. And so it could be new platforms that are the 1 trillion dollar companies of tomorrow that replace them. But this is a little bit unclear. What is clear is what you said that huge efficiency gains will be made by scaling AI within the companies. Some companies may be better placed to do it than others. Some companies may be more willing than others.

**[0:24:53] Jean-Paul Carvalho:** It's really going to depend on management and how they make this transition. And but that is going to be a huge source of value creation. Secondly, on a slightly old fashioned note, I think that commodities, basic commodities such as copper, silver, uranium, et cetera, antimony, tin, all kinds of things, will benefit considerably from the coming AI data center and infrastructure build out. It's not clear where all of these raw materials are going to come from. So much is required for this build out, but only price can really solve the imbalance at the moment between supply and demand.

**[0:25:29] Jean-Paul Carvalho:** So I think there are going to be huge huge gains in that relatively old fashioned sector as well. So those are two things. And I think, John Paul, what you say is is being realized at the moment if you look at the commodity pricing and markets and copper and and silver in particular and others as well. But if you dig into the scaling as a source of competitive advantage, which is going to be of interest to a lot of our our listeners who work in large organizations, these could be private sector organizations, these could be governmental organizations.

**[0:25:38] Jean-Paul Carvalho:** There's what do you see as the key elements of the playbook to scale AI successfully and avoid the the kind of plethora of pilots that that we sometimes see? That's a difficult question. It's going to depend on the individual companies, individual industries. In healthcare, there's a huge amount of back office work that and and depends on the context. US is very different to the UK, for example. US is extremely complicated system, and there are new, very nimble healthcare providers that can and well placed to really simplify that whole process.

**[0:26:39] Jean-Paul Carvalho:** And that's going to take that's going to cut down on a huge amount of cost. In education, for example, you see this in healthcare and other industries. A huge fraction of the increase in the wage bill, sometimes over 100% of the increase in in the wage bill is as due to administration. And that has ballooned since the early twenty tens and even before. Finding ways to really trim that down and eliminate all the inefficiencies that crop up in large organizations, that's going to be the key.

**[0:27:16] Jean-Paul Carvalho:** What is the the silver bullet for that? That is a difficult question to answer in general. I think it depends on the specific industry. One choice is facing business leaders at the moment is, you know, automation versus augmentation, just to pose them as two sort of polar extremes for the for the discussion, if you will.

**[0:28:42] Jean-Paul Carvalho:** So automation is obviously, as you've described with many examples, mechanizing, automating tasks, which may be routine or or actually more cognitive, as you say, JP. And that obviously has its immediate attraction to organizations. It drives efficiencies, it drives speed. But then you have the notion of augmentation, which would be that I will actually invest to enhance the effectiveness that emerges from AI. So, for example, automation might help give a sales colleague information more quickly and and at lower cost.

**[0:28:52] Jean-Paul Carvalho:** But then I'll invest in a skill set for that sales, the remaining sales colleagues, to improve their effectiveness and their output. How would you see these these forces playing playing through the automation versus augmentation? Do you think that augmentation is is a sort of a false hope, if you will, and labour augmentation is a false hope, and automation is just too alluring for organizations? Or or do you think there is a prospect of of of a kind of augmentation wave, if you will?

**[0:29:00] Jean-Paul Carvalho:** I think much of it depends on an earlier stage on how the technology is developed. So Aj Mogle and Johnson look in their book about which deals with automation and also AI. They make this distinction between a kind of substituting AI technology and human complementing AI technology. And they distinguish between two paths that the technology could have taken. The current path is really of technological development around AI is really about substituting for human labour. If you look at the release of any AI model, the benchmarks are beats 99% of humans in task X.

**[0:29:26] Jean-Paul Carvalho:** It is not we it made humans faster, 99% faster at this particular task, or it made humans plus AI 20% better than just AI. Those are not the benchmarks that these models are being trained to achieve. And so the direction has taken a a a substituting is a kind of taking a substituting direction, which is really w proposed by Alan Turing, right? How do we develop something that substitutes for human? But there was an alternative line of work developed by Norbert Wiener, which was about how do we develop technology that best complements human productivity?

**[0:30:05] Jean-Paul Carvalho:** And that's what Asmogle and Johnson advocate, and I think that's a very, very important point, and I echo that. Not only do you get a much more humane type of organization, you also solve some of these problems of keeping training humans to be in the loop at a senior stage. And moreover, at a much earlier stage at the level of education, problems for firms can already be created at that level. Because think about the incentives for investment in skills and human capital.

**[0:30:37] Jean-Paul Carvalho:** In the past, fifteen years ago, if I were learning how to code, I would have 99% chance that I would beat a computer, an AI system. Today, I would have very little chance. So what is my incentive to invest in those skills? And those skills can be then affect the accumulation of other skills through the economy that are required for various tasks. Moreover, what happens if others stop investing in these skills? How do I learn from people?

**[0:31:06] Jean-Paul Carvalho:** A lot of our knowledge is tacit and gained in person. But if others are investing in the skills and I can't gain the skills. What if there's a big employment shock? Where do I get my role models from? How do I acquire my skills of punctuality, diligence, grit, all of these things? And so at earlier stage of human skill formation and human productivity, the complementary approach to AI, the human complementary approach to AI, which is the human oriented AI, is going to be hugely important.

**[0:31:28] Jean-Paul Carvalho:** And I think, JP, that moves us nicely onto closing thoughts. You've covered the gamut of the labor model, the organization, the value chain, the implications for the business leaders, and then augmentation versus pure automation. I guess in closing, as you look at the development of AI, are you fundamentally optimistic or pessimistic? And and why would that be? I see a range of scenarios. I'm not sure how I would assign a probability distribution to those different scenarios.

**[0:32:07] Jean-Paul Carvalho:** So I'm not sure if I'm I see different optimistic and pessimistic scenarios. Here the optimistic scenario. The optimistic scenario is, and we have to ask ourselves, why are we doing this, right? Why as a society? Why does this make sense? The optimistic scenario is that AI frees humans from mundane work and unlocks human creativity and our ability to devote our time and energies to things of real value to us. And that could be work, but it could be many other things.

**[0:32:40] Jean-Paul Carvalho:** So we've liberated from work, from mundane work at least. That is what's what John Maynard Keynes thought would happen. That is what a lot of people aspire to, and it is possible that that occurs, but it depends on how political institutions adapt to the AI revolution. If political institutions remain inclusive, the egalitarian norms remain in place, then the productivity boom that could be generated from AI will be channeled into a state in which individuals will be able to live with minimal work, dedicating their lives to things that they truly value.

**[0:33:06] Jean-Paul Carvalho:** But if political institutions don't adapt in a kind of egalitarian way, let's say, minimally egalitarian way at least, then it could go in a different direction. But the optimistic thing is that there is a positive path forward, and which way it goes is really up to us. JP, thanks for joining us today inside the strategy room. It's been terrific to have you, and I'll definitely invite you to my cousin's wedding that's coming up to give the wedding address.

**[0:33:15] Jean-Paul Carvalho:** Um, in a way. I don't, I don't recommend it. Great that you can join us, and thanks for thanks for being with us today. Thanks so much, Robin. With thanks to Jean-Paul and Robin for sharing their insights with us, and thank you to all of our listeners for joining us today. We hope you enjoyed the conversation and we welcome your ratings and reviews on any podcast player. You can also reach us via email at itsr at mckinsey dot com, which stands for inside the strategy room.

**[0:33:25] Jean-Paul Carvalho:** If you'd like to access more of our insights on driving organizational change, we've included our most recent articles for you to explore in the show notes for this episode. If you'd like to subscribe to our weekly series, you can follow us on any podcast player, where you can also access our entire library of nearly 300 episodes. We've also organized our podcast library across nine major themes on our inside the strategy room collection page at mckinsey dot com slash itsr, where you can also access written transcripts of each episode.

**[0:33:38] Jean-Paul Carvalho:** Finally, if you'd like to automatically receive our latest publications and insights, we encourage you to sign up for email alerts on our insights page at mckinsey dot com slash scf, or connect with us on LinkedIn at our McKinsey Strategy and Corporate Finance Practice page, which we've also included links to in the show notes. Thanks again for listening. We look forward to having you join us again next week inside the strategy room.