How Generative AI Changes Productivity

Most enterprise AI stalls at pilot stage. Publicis Sapien is already powering Fortune 100 companies today. Publicis Sapient AI that's built to deliver. Learn more at PublicisApient.com.

Karim, you've been experiencing some deja vu lately, huh?

Oh, yeah. 30 years of deja vu.

Yeah, tell us about it.

Yeah, so I still remember the first time I saw the web browser. This is almost 30 years now. And I was working at General Electric at that time, and I was at a radiology conference, and this doctor pulls me and said, hey, Kareem, look at this. I'm like, what am I looking at? Oh, it's a coffee pot. This is a coffee pot at Oxford University. I'm like, okay. He goes, yeah, this is the web browser now. You can browse anywhere and get any information you want. And it didn't dawn on me till later that what he had showed me was that the cost of information dissemination was going to go to zero. That a coffee pot put aside by lazy researchers who didn't want to cross the lab to look at if the coffee was full or not now had a global distribution of its image, of its video live feed. And there's a straight line between the browser being invented, Google being invented, and the rest is history. And I get the same feels today. We've been talking a lot about AI for 20 years, Amy, but until ChatGPT came out, nobody knew what it was. And then it's like, oh, my God. And it's no surprise that 100 million users started using ChatGPT in three months, which is a ridiculous number. And they all had the same deja vu moment that I had. Like, oh, now I understand what this AI thing is, and it's cute. Right now I can get a recipe. I can get a trip planned. I can. I can get a Kendrick Lamar version of all my papers. I've done that. I've taken our pieces and put them into ChatGPT and have a Kendrick Lamar rap about our article on competing the Age of AI. It's great, it's cute, but we are just getting our heads around what this means for us.

So for you, Karim, it was about the dissemination of information with the browser.

And then what that enabled.

Yeah, yeah, and spoken like a true economist. For me, it was when I saw my first hyperlink and I had this mind blown moment where I suddenly realized we could make connections across ideas, arenas, et cetera, with no friction whatsoever. I felt my head expanding with that. But I kind of wonder, what is the analog right now? I mean, what is it that you saw with generative AI?

I would say the cost of cognition, the cost of creativity is going to massively decline with generative AI.

Wow, okay. That's a lot to wrap your head around.

And I'm still like, whenever I say that, I'm like, I don't know if I'm right or not. And this is where I'm like, stop being my rigorous academic Today. Lots of people are getting the same feels about this and trying to make sense of it. And so I think my aha. Is cost of cognition, cost of creativity is dropping. And this has profound implications for us.

Right. Some of them kind of frightening if you carry it through. So we'll get into that. Welcome to How Generative AI Changes Everything, a special series of the HBR IdeaCast. Through conversations with experts, we're exploring how this new technology will change workforce productivity, creating creativity and innovation. And we're asking how leaders should adopt generative AI in their organizations and what it means for their strategy. This week's episode, how Generative AI Changes Productivity. I'm Amy Bernstein, editor of Harvard Business Review and your host for this episode. And I'm talking to Karim Lakhani. He's a professor at Harvard Business School and has been researching generative AI. He also co wrote the book Competing in the Age of AI. Karim, welcome. I'm so excited to be having this conversation with you.

So glad to be here, Amy. Looking forward to it.

Let's talk about what this is, because something big happened last November.

Yes.

Something happened with AI. It became generative. What happened?

So I just want to put it in historical context. So I think there have been waves of AI. All the AI plumbing that has existed prior to this moment in November have been fast learners, right? AI algorithms, machine learning algorithms that can make predictions really well, pattern recognition really well, automate processes really well, Netflix, Google, Amazon, Baidu, you name it, all were based on these technologies. Generative AI took all these things, and what did we have? We had massive data infrastructure in place. We had massive compute becoming available in place. And they said instead of just doing a prediction about the next sales thing or the next HR person you want to hire, let's make predictions based on all of human text, all of human graphics that are out there and get these systems to train themselves such that they can predict the next word. That's basically what's going on. The generative AI systems are basically taking a corpus of text, right? Jack and Jill went up the blank, right? And now it's read everything on the Internet. It has availability of all the text in the world and it makes a prediction of what the next word is going to be. Hill. That's all it's doing. In its most basic sense, that's all it's doing. And the same thing happens with all these image generation systems. It's basically predicting what the next pixel should be. That's it. But now take this and do it at scale. You fed it the whole world's information off the Internet, good and bad, right? The whole world's images, good and bad. And then it's just predicting the next pixel, the next word. So then what does it do? Jack and Jill went up the hill, right? And then you say, what's the next word? Two. Right? So inside it goes back and says, what should the next word be? Fetch a pail of water. And behind it are systems that are voting and auto tuning themselves to sort of say, what is the probability of the next word that needs to be. By the way, I'm abstracting out a ton of complexity, ton of math, ton of compute and so forth, but that's essentially what's going on. But what happened was that these systems, these large language models started, these generative AI systems started to get developed basically five, seven years ago. And then people said, well, how do we make them better? How do we make them better? And then the big aha that OpenAI did is actually, let's actually bring in some human feedback. But let's say when it generates a sentence, is this any good or not? Which is basically what they called reinforcement learning. With human feedback, you create a reward system that says the human gives you feedback and that says, be more like the human. And that was the training. And that combination of all this data, all this compute, this next word prediction, this next image prediction basically has given us these systems coming through. So in one sense, it's like this dead simple. If you understand the history of AI, you understand what's been going on, but there's been this magical moment and literally it feels like magic. I was working with somebody who's from Europe, and we're working on a case that we're doing, and I said, oh, let's go do this case together, right? And we put the text in and all of a sudden it was like generating all this text for us. He's, like, stunned. I'm stunned, like, oh, this is kind of cool. And. And the more I probe it, the more it responds to my prodding. It has no sentience. It's literally statistics at scale giving us the next Word. And with human feedback, it just keeps getting better and better.

And so now what we have, even in what feels like its infancy.

Yes. It still is.

We have a tool that can pass the medical boards, it can pass the bar exam, it can apply for a job and get shortlisted.

Yes.

And I'll just keep saying this is in its infancy. Right. So it's amazing. Terrifying.

Yes. I'm not terrified. I'm optimistic.

Okay.

You know me.

Yeah. You are an optimist. And I'm not gonna ask the obvious question about is it gonna be able to edit Harvard Business Review in six months, but.

Well, look, Amy, many people say, are machines gonna replace humans? Right, Right. And many computer scientists have said, machines aren't going to replace humans. Humans with machines will replace humans without machines. I think now it's true.

Okay.

And I love sort of this copilot metaphor that Microsoft is using. Right. So now you have this copilot that knows everything. But remember, when you're the captain and you're the copilot, it's your responsibility to keep the copilot in check.

Right.

And know both that it's superpowers and it has extraordinary superpowers, but also known as limitations. And that's where humans will matter. And my asterisk to this, machines won't replace humans. Humans with machines will replace humans without machines is now like. And maybe we'll need fewer of them. So I think this is where it's going to get interesting for many of us, and that has huge implications for everything we do. Huge implications.

Well, and you know, what you've described in your book, in your articles, and what I think you're getting at now is that it can make our jobs easier. Right. It can help us do what we do better as well.

Yes.

So talk a little bit about how you see that happening in the near future.

Yeah, yeah. My sense is that this transformation moment has arrived way sooner than any of us anticipated. And even my colleagues in computer science at Harvard, they're also stunned at the rate. So there's like this inflection point we have reached with these technologies. And so fundamentally, for me, this provides you with the Iron man suit. Right. Those that, like Marvel Comics, it provides you with these superpowers. And the question becomes, how do you use it? Where do you use it? And how does your work change? How does your organization change when this becomes available? And already anecdotally, you're seeing some very interesting things where lots of the folks that were looking at AI thought people with low skills are going to get hurt the most. It looks the opposite may be true, at least on the left tail. So if you imagine a normal distribution, a Gaussian distribution of skills, there's people on the right tail which are really high skilled and people on the left tail which are really low skill. Well, what the technology does is that all of a sudden now you become as good as the AI, which is remarkable. And already a bunch of studies have come out with which have shown for writing skills and for customer service skills and so on and so forth. The introduction of these technologies, everybody gets as good as the AI, which in itself is like mind blowing.

So it just brings up the quality level.

Brings up the quality level, right. And then the question that we are all racing to understand is what's going to happen to the right tail? How will they use it? There are some early hints that we don't know. There's a postdoctoral fellow working with me, Dr. Fabrizio Delacqua, he's actually done a full on experiment that people fall asleep at the wheel with good AI, with really good AI experts.

What does that mean?

What that means is that if you have really good AI, it's really good at solving all your problems. So then you don't pay attention. And when the edge case comes, you don't notice it. So overall your performance drops when edge cases come. And hence this thing about you are the captain, you have a copilot, the copilot is really good, but it's not flawless and you got to be paying attention to what's going on. And so if the right tail gets these technologies and they aren't paying attention, it could lead to worse performance, for example. But again, we don't know yet because we're just at the beginning stages. Our training systems haven't changed. How we teach hasn't changed already. Today, Amy, I'm putting my cases into OpenAI and getting perfect, perfect student comments back. I put in there saying, imagine you were an investment banking student at the Harvard Business School. Read this case and give me five comments that will make me look very smart. Boom. Then I can probe each of those comments, go deeper, go broader. Right now, if every single student of mine has these superpowers where they may not even read the case, if I put myself back 25, 30 years, I was pretty lazy as a student not to denigrate them. When you have this tool available, you might just rely on it. If that happens, how do we teach? What should the curriculum look like? So for us, it's a crisis moment. For me personally, it's a crisis moment as to what I'm gonna teach, how I'm gonna teach in the fall.

Okay, so I'm going to say I'm scared when you say that.

Yes. I think it's amazing.

It's amazing. It makes me. I mean, I think it's another conversation, you know, what is education? What is learning? What is thinking?

Huge.

How are you using?

As much as I can.

Yeah, you use ChatGPT every single day. How do you use it today, by the way?

It's not just chatgpt now. It's such a big explosion. Right. So you can go back and forth between different systems and try that. And then I also use Bing. Bings. And like, I'm using Bing. I know I'm using Bing.

I have it on my phone.

I know, I know I'm using Bing. Oh, my God. But Bing chat is great. And so today, what did I do? So I've been super crazy busy and I've been struggling with this piece of writing for an academic journal, My co author and me going back and forth. So I put three pages of text in and said, can you please clean this up for me? That's all I said. And it clarified. It cleaned up. And I went back and forth, oh, do this, do that. And what would take me if I hired a copy editor? What would take me? Weeks of iteration with the copy editor. I was getting it done instantaneously. Okay, it definitely hallucinates, right?

Oh, yeah.

Yeah.

So say what you mean.

Oh, yeah, hallucination. So, so basically, look, it's a generative model, right? So like the. Again, like the Jack and Jill went up the hill. Right. To fetch a pail of water. It could also say, Jack and Jill went up the hill to fetch a pail of oil. Right.

And then credit Shakespeare.

And credit Shakespeare. Right. Because it's just. It's a statistical model that's generating things. Now, I think over time, the hallucinations will go down. You know, you'll connect it better to more trustworthy sources and so on and so forth along the way. But hallucination. So again, like, you cannot let it autopilot. You can't. You can't do autopilot. You can't do autopilot.

That's the copilot.

That's the copilot.

So, you know, you talked about the edge examples.

Yes.

Let's talk about a professional.

Yes.

How is an accountant using this? How is a lawyer.

Or should use it?

Or should use it? A consultant, someone doing what we used to think of as Office work.

Yes, yes. Lots of us do office work, desk work, knowledge work. To the accountant. You have a massive spreadsheet. Put it in, ask it to explain what the data mean. Ask it to do analysis for you, conversationally, they'll do it for you. You have a spreadsheet in which the columns, the data are missing, the columns are mislabeled. Ask it to go figure it out what might be, it'll do it for you. You are a legal assistant. You want to think about case law. I'm sure right now there are companies that are going to introduce. You have access to all the legal documents in the world. What should the case law be on this before, it would take you hours, if not days of searching. Now available to you instantly customize it to your particular circumstance.

Right.

All the paralegals can be better than lawyers today.

Wow.

And so that's the thing, colleague of mine, lots of us are busy with lots of demand on our time, so he has to send out declines a lot. Now he's automated the process where the decline comes in. He presses a button, it drives an automation on the back end of Zapier, which then goes into OpenAI, generates a very nice decline, brings it back into Gmail, and then it sends it out. And the responses he's gotten from the. This is so funny, from the people that he's declined are the best responses ever. Because like, it's like a sweet response and typically I'm like very curt. I'm sorry, too busy, go away.

Yeah, right.

This is like, you know, you're asking me about these things, here's some references and so forth. I'm sorry, I can't help you. So they come back and say, oh, thank you so much, you've really helped me, even though I know you're so busy.

Wow. So it helps us even be nicer.

Yes, yes. And so these are all these examples. And what's interesting, Amy, is that right now, again, we're still experimenting, we're still learning, but the speed by which this stuff is being integrated together is incredible. And that's why individuals will be like the first ones to go adapt this.

So how do organizations, how will organizational leaders even know?

They won't look, I think from an individual worker productivity story. This is phenomenal.

Coming up after the break, we're going to lay out how generative AI is going to change productivity at scale across a workforce and across industries. Stay with us.

Most AI delivers promises. Publicis Sapiens just delivers enterprise ready AI that works. Not pilots, not prototypes. Modernization up to 75% faster. Cost cut in half billions unlocked in new revenue trusted by Fortune 100 companies because quite simply, what they build works. Publicis sapient AI that is built to deliver. Learn more at PublicisCapital.com.

Welcome back to how generative AI changes productivity. I'm Amy Bernstein. So right now, you know, generative AI is sort of experimental in the enterprise. People play with it. You play with it, I play with it. But productivity is obviously a huge opportunity here. So how is this going to change the way we use AI in the workplace?

Well, I mean, I think it's already changed it. Right, because people are now using AI for mundane tasks like responding to emails, doing spreadsheet analysis, doing textual analysis, doing summaries. You can. This is a new feature. I do a lot of zoom calls. You record the zoom call, you put the transcript in the zoom call. It'll create a beautiful summary for you of the entire conversation. You can ask it. Can you generate for me the to do list? Can you then also create the OKRs from this meeting?

Oh, my gosh.

So that's like now, this is not science fiction. This is not like six months from now. Today you can do that.

Wow. So let's go through a lightning round of sort of enterprise functions.

Yes.

Can generative AI help with taking customer service calls?

Yes, already. There's a study that just came out that showed you can imagine the augmentation scenario and the automation scenario. The augmentation scenario shows that in fact, the low end of the customer services get way better. So now you would not have the variance that you often see in customer service. Massive. Now you can also imagine an automation scenario where the first level triage is happening through generative AI. You have literally a chatbot that is trying to answer as many right now. You can actually give it your entire manual and then it'll just do that for you. Just on this example, I was talking to some folks who run submarines, okay. And they're like, I love a generative AI system that has all the submarine manuals ingested. So now I have the full knowledge base about the submarine manuals and live data. Then I can just be sitting there saying, hey, what's going on with you sub today? That's the same as the customer service. Imagine customer service for the submarine where you're actually talking to the live submarine with all this stuff coming in. Statistics at scale, doing this for us.

I mean, a synthesis.

Yes. Yeah, yes.

All right. Streamlining administrative tasks. Yes.

Talked about, of course. Yeah.

Scheduling and task management.

Yes.

Where do you see that?

Well, again, like the calendar synchronization, like we sent a doodle poll to everybody and then like, what, what, what? Now it'll just literally look at your data and look at everybody's calendars and says, here's the three dates that make sense. Go.

Wow. Okay, how about identify talent gaps and train employees?

Well, what you have to do is now break it up into components. Right. That's like an HR person would sort of think about or manager would think about. Look, one thing is I'm hearing already is that this raises a very interesting horror story, ethical dilemma and so forth. Do you work with Chad GPT to create your employee reviews?

I mean, it's out there.

It's out there and people are doing it, right?

Yeah. I mean, feeding chatgpt any piece of information you don't want out in the wild is kind of crazy.

Well, no, but now, you know, look, they announced private mode. So. And this is like the rate of change is so fast. Private mode is available. So you just use the large language model, but it doesn't go back into the training data set.

Right.

And OpenAI just announced like OpenAI for businesses coming soon. So that again.

Right. And they'll be gated model models, lots.

Of those kind of things. But here's the question. If I have a team and I'm their mentor and their supervisor, what is my role? And then you could almost imagine a thought partnership experiment to say, I'm trying to build this type of a team. What kind of talent do I need? Look at the world, tell me what talent I need. Does this talent exist in my company? If you have your HR data locked into it, then you go, okay, what, what skills do they need? Do these skills exist?

Right. I mean, really, strategic HR leaders already have strategic talent plans, right? What skills they need on board? They have audits of the skills they have on board. They know what the gaps are, right?

Yeah, but those are all handcrafted, take thousands of hours of work and feedback. Now, some of these systems are sophisticated, so they have all this data around and is labeled. Now imagine if you've done that heavy lifting and that data in your HR system exists now. All of a sudden you can be actually, you could be one of the best employers because you could be thinking about where the trajectory is. You might identify gaps, you might get suggestions for new training.

So we know it can help with communications of all sorts.

Yeah, marketing. Marketing has completely changed now. Oh. Like I was talking to CEO of a large real estate development company and they're like I'm having second thoughts about how big my marketing organization is because all of our press releases are now being done through this thing. And they're really good.

And they're really good and they're seeing results.

Yeah, yeah.

I can just imagine, you know, a clothing retailer, you know, sending me pitches with people who look like me, who are shaped like me in clothes that look good.

Yes, exactly.

That would work.

That would work, right? That would work. Another mind blowing company these days is Shein, right out of China. Shein is blowing my mind. They have 6,000 SKUs per day that they launch.

Wow.

They can scale production from 100 items in a factory to 100,000 in a week. All AI driven, all digital. Right. All based on consumer marketing, consumer transponding and being able to do that at scale.

So you just mentioned factory. How can generative AI increase efficiency, productivity and effect?

So just imagine again for quality control, for detecting errors. It can do that on the fly with you, it can help you. But then also I would say that the whole process automation side of things can be greatly enhanced. And I think the main thing to think about is people often say, oh, there's digital transformation and then there's AI. I go, no, it's the same spectrum. The scenario we talked about, the submarine, automated customer service for the submarine, about how it's feeling today. That can also be in the factory if you have the data streams from your devices available on the fly to then interact with these models. So the digitization imperative increases even more. So if your factories aren't connected, if your devices aren't connected, you won't be able to benefit from it. So you got to build the AI factory, you got to build the data pipelines to then add this layer on top of yourself.

So it's imperative. Yeah, what did I leave off? How else can generative AI improve firm productivity?

Look, I mean, I think the way I would sort of think about is that we can find many use cases across the board. But I think the question I ask myself, which is how does the nature of the firm change given these capabilities? What does the firm look like, which is we can add pockets of generative AI across the enterprise and do this and do that and do that. I think my worry is then we're going to be back to the Barnes and Noble and the Internet. Right. So if you think about it, Barnes and Noble also had a barnesandnoble.com, bN.com was set up and they tried to patch the Internet on top of their existing business and ultimately it was not successful.

All right. We are already seeing companies struggling with integrating AI into their businesses. We know it's hard.

How does generative AI it makes it even harder change. Because again, what's going to happen is that there'll be lots of use cases and we'll get the bumps in marketing, we'll get the bumps in manufacturing, we'll get the bumps in legal services and so forth. But that will be the incomplete story, because what's going to happen is just as Amazon emerged as a new model to how you organize your firm, that the opportunity is for executives to not just be satisfied with the bump or put it aside on this one part and not bring it in, but to rethink processes, rethink their business model and their operating model. I think that's the big, big imperative that lies in front of us.

And the promise there is efficiency, efficiency.

Productivity, better customer service, better profitability, you name it. Right. And again, what have we seen? Pockets of Ant group out of China and financial services, Shein now and Temu for fast fashion, fast retail, Moderna as an example for pharmaceuticals. And lots of pharma companies are trying to catch up and build their models.

These are all companies that were born digital.

Yes.

What if I'm gm?

GM is on a. On a road to be digital. And they have to do more. Ford has to do more. You know, Disney is a great example of somebody of people that are trying to do rethinking and really push and they'll have to do more as well.

Right. But the generative AI is not just a wrinkle. I mean, this is. They have to rethink how they're rethinking.

Yeah. I mean, all the creative aspects of it, all the marketing aspects of it, all the experiences that they do, all of them.

And you've in your book made a very compelling case that companies must embrace AI and they have to do it at the IT architecture level and at the organizational architecture level. But given how easy it is for the individual to use generative AI.

Yes.

Is the argument still the same?

I think so. Even more so, I think, you know, there's a rough rule of 30, 70, 30% is tech, 70% is organizational. And so again, maybe we'll get a bigger bump, we don't know yet, in the productivity statistics of people using this in the firms with their existing jobs. I'm just worried that we're going to have lots of Barnes and Nobles around, which will have a website. They will do E commerce. But the guts of it are still the physical stores. And so ultimately we'll suffer and then we'll have the Kmarts and the Sears disappear along the way. They all had websites, they all had E Commerce. They spent hundreds of millions of dollars on it. But they didn't rethink.

Right?

And so I think the question becomes, can you rethink, can you rethink processes?

It cannot be a bolt on.

No, no, no. And it requires this top down understanding. Leaders have to learn and embrace and not just for themselves, but for their whole organization. There's a big learning mandate here for boards for the C suite and for the rest of the organization.

Karim, I just learned that Microsoft is going to integrate Copilot everywhere, everywhere.

Office, PowerPoint, Excel, Word, you name it, BI, their business intelligence suite, all that kind of stuff.

So we're all going to be able to experience the productivity boost of generative AI on an individual by individual level. But the promise is really scale. How do we move from oh my goodness, it is so much easier to make a deck now to organizations achieving unimagined scale thanks to generative AI.

Look, I think the question now becomes, if it's so easy to do knowledge work with this copilot, what more can you do? If something took you two weeks and now it takes you two hours, are we in the beach for the rest of the time or are we going to just be able to do a whole lot more? And so I think that's going to be the really interesting question. I'm excited to see where I am still scared, but also optimistic about what this means for us. Because when photography got invented, there's a direct line between photography getting invented and the modern art. A moment arriving because still life was no longer the thing that people cared about. And the European art scene changed and we're now in this world of modernism. So I don't know yet, Amy, how to think about when accountants get all this productivity boost. What will they do? Right. Right again, the answer could be twofold. One answer is, well, we don't need as many accountants, right? So let's fire them all. But I think that would be the wrong answer. Some companies will do that and I think that'll be the wrong answer. Now if you say if my accountants have the superpower, what more could they be doing? What additional higher level stuff could be doing? And we see parallels of this. Remember when the ATMs got introduced, we thought all the bank tellers would be fired and what happened instead is they became financial advisors.

Right?

So I don't know what the accounting financial advisor equivalent is going to be, but. But it's going to emerge. And that's the exploration that I think we should all be open to. And leaders need to be thinking that way. You can do the cut, cut, cut model, but that won't make it an attractive place for you to work. If you now say enablement model, reinvention model. That's the exciting story, because you've removed.

The boundaries on human capability in a big way. And in terms of leadership and management, to me, this is also mind blowing because on the one hand, it suggests that the humanity of leadership becomes all the more important. Right. But what can you enable? What can you make happen as a leader with this tool that you weren't able to do a year ago in terms of inspiring and motivating your team, but also in terms of accomplishment?

Accomplishment. And, you know, look, I think it goes back to. I think one of your cover stories was the purpose Driven Fern. I think we're gonna be back to purpose.

Yeah.

Like, what is your mission? Because now you can accomplish your mission. What is your worthy mission that's gonna attract people to work with you.

Right. And then once you've integrated these capabilities, how do you accomplishment in a way that is doing well and doing good?

Exactly, exactly.

So we've talked sort of vaguely about the opportunities. How do I, as an organizational leader, start thinking constructively about those opportunities tomorrow? What do I do? How do I identify them?

Yeah, I would say it's learning and doing. Learning is immersion for yourself in the tools, and the learning is through doing as well. So, like, practice, you know, read, take courses.

But not just you individually. You have to organize your team to do it.

Exactly, exactly. Start to do stuff and start to reimagine what work looks like in your functions and your organizations. I think that's the biggest imperative.

And, you know, we've hinted at it, and a lot of folks have remarked on this, that, you know, it's part of the hype cycle, which is the horror piece, which is, is this gonna eliminate a lot of jobs? What do you. How do we workforce now?

Yeah, yeah. Thank God I'm not a macroeconomist.

Yeah, yeah.

What I would sort of say is, is there going to be transition? Is there going to be upheaval? 100%.

Yeah.

100%. Will there be misguided regulations? Will government intervene clumsily? 100%. This. It's going to happen. Right, right. This will go in fits and starts. And I can imagine that a set of leaders will do the efficiency story and will get super lean and a set of leaders will say, I've got all these people, what can I do with them? And what more can I do? So either will be in a world of efficiency and scarcity or one of abundance. I've got all this stuff, all these great ideas, all these amazing options. What more could I do?

So maybe it's a growth story versus an efficiency story.

Yeah, but I think we'll see both. I think both will show up and it'll depend on the leadership moves. But, but, but I still think my big worry is, are the Barnes and Nobles gonna survive? Will this just be bolt on or will we do radical transformation as we need to? I just heard this thing. It's crazy. Somebody was in a board meeting and there was some discussion about pricing. Another board member was on ChatGPT, asking it, feeding it some questions and then using that in the board meeting back.

Wow.

Right? So if board members are doing this, this is like a legit big firm.

Wow.

Right? And so the CEOs have to take on the learning, take on the doing, embracing it, not fighting it. Also, don't be stuck in the cross section of capabilities today. Right. Even the 3.5 to 4.0 on the OpenAI models, massive improvement in performance. And what's happening, Amy, which is so interesting, is that it's not just one company. There are hundreds of companies trying to do this right. So we will just see combinatorial explosion of new ideas and new ways to build. These things happen. And so you don't want to be in the receiving end of it. You want to be the leading part of it. From my perspective, to make that happen and then put on your designer hat and think about what does the design of your organization look like. I would encourage abundance thinking instead of efficiency thinking.

Karim, thank you so much for taking the time to have this conversation.

You're welcome. You're welcome. Always, always, always a pleasure to chat with em.

That's Kareem Makhani, a professor at Harvard Business School who specializes in artificial intelligence. He co wrote the book Competing in the Age of Strategy and Leadership. When Algorithms and Networks Run the World. Next time, how Generative AI Changes Creativity. HBR Editor in Chief Adi Ignatius talks to an artist as well as innovation researchers about the impact of this tech on creative work. That's next Thursday right here in the HBR IdeaCast feed after the regular Tuesday episode. This episode was produced by Kurt Nickish. We get technical help from Rob Eckhart. Our audio Product manager is Ian Fox, and Hannah Bates is our Audio Production Assistant. Special thanks to Maureen Hoch. Thanks for listening to how generative AI changes everything. A special series of the HBR IdeaCast. I'm Amy Bernstein.

Most AI delivers promises. Publicis Sapiens just delivers enterprise ready AI that works. Not pilots, not prototypes. Modernization up to 75% faster, cost cut in half billions unlocked in new revenue trusted by Fortune 100 companies because quite simply, what they build works. Publicis sapient AI that is built to deliver. Learn more@publicisapien.com it.