This series is about AI’s disruption of entry-level work and what universities should be doing in response. Give us a quick picture of what you have going on right now at Foster and in your lab that ties to that question.

I’m Léonard Boussioux, Assistant Professor of Information Systems and Operations Management at the Foster School of Business. We are very dynamic in AI. We are iterating on curricula. We are asking all faculty to update their courses to include AI in one way or another. It can be in assignments, in the way they evaluate, in the content they teach, and we are trying to bring that culture throughout the entire school. We also host many lunch-and-learns, including for faculty, staff, and students. There are lots of workshops around many domains, including about things that are brand new, such as Claude Code. We organized days dedicated to AI to create a sense of community and foster engagement among students and the broader university.

On my side, I am the head of the Human-AI Operations Lab that I created to investigate how humans and AI can work together to democratize access to expert problem-solving. My goal is to find new methodologies to develop more creative ideas, knowing how to evaluate them at scale, also so that we can model, code, and prototype complex phenomena, and make sure that we have the proper human-AI collaboration throughout.

I care a lot about robustness, about multimodality, meaning how we can leverage multiple data sources together. And I love investigating how agentic AI and these autonomous systems can help us. So overall, there is a lot going on, and it’s moving very quickly. So the goal is really also to place the human at the center always, to make sure that we feel we are building something that we like, a world that we are comfortable in, and that this whole system participates in maintaining as much joy as possible.

In Poets and Quants last month, you said that at the individual level, Gen AI raises quality and novelty, but at the collective level, it compresses the idea space. Outputs converge, and the long tail of unconventional breakthrough ideas thins. Can you expand on that and what that actually means for the next generation of graduate students?

Absolutely. This is a very interesting finding, that, individually, AI can help you a lot by offering you different perspectives that you would not necessarily have on your own. It can also help you craft your content better. But as a result, people tend to over-rely on AI to come up with ideas, and they forget to use their own brainpower or brainstorming to figure out how to solve a given problem.

The challenge is that we typically use AI in similar manners. Of course, you can prompt a bit differently. AI has a bit of a different context for each one of us. However, because we all use the same models trained on similar data and tuned in similar ways, we end up with AI recommendations that are all similar to one another, even if they were generated by very different people.

The challenge, as a result, is that AI brings us the same 10 really good ideas, but we’ll never go into the 90 other good ideas that are rare, that are more unconventional, because AI loves to go with the most common ideas that are a priori good enough and safe.

That is a challenge, and I actually teach that content to my students as well. I give them exercises so they can use the power of their brains, their own creativity, and, above all, feel that they are necessary as humans. So I do a lot of experiments, and I ask them to come up with ideas, for instance, around the circular economy. Circular economy is an economic model designed to minimize waste and maximize resource efficiency. Think recycling, refurbishing, so it’s something that can do good for the planet.

So if we ask humans and AI, “What are your ideas?,” AI can match the creativity of a whole human crowd. But if we look at the set of ideas it proposed, it’s very small.

So we need a lot of innovations in the way we work with AI, by providing different contexts, by bringing some of our own ideas first, by pushing back the AI, by iterating, by creating multi-agent systems that can actually improve the performance of AI, and making sure we maintain a mindful friction through the whole process, such that the humans won’t accept immediately an answer from AI, and will be able to not cognitively surrender to whatever it gives us.

Your Crowdless Future paper in Organization Science shows that Gen AI matched crowd solvers on creative problem-solving tasks. The crowd was the entry-level layer of how innovation work used to be sourced. If AI does the crowd-sourced creative work, what should the business graduate be doing instead? And what does that mean for how you grade?

This is a very lovely question that always occupies my mind. I have multiple aspects to this answer. The first one is that AI can really help in the ideation phase. However, we still need to curate and maintain a sense of taste. If AI can propose 100 ideas to you, you don’t have the resources to make 100 ideas come true. That means AI can start doing them, but it’s also expensive for AI to compute a given idea. It costs tokens, so it’s already money. So we need to select which ideas we want to push.

And this selection factor is a moment where the human can provide their input and experience, their intuition of what will probably work. This instinct comes from all our experiences, from the past, from what we learned at school, and from the experiments we’ve been following during our work time. And as a result, this is something that we should absolutely maintain in the process.

So AI can augment our own set of ideas; it can augment our ability to execute, but we should always make sure we maintain a direction and a vision to obtain satisfactory results. So that’s aspect number one.

When it comes to teaching business school students, it’s hard to teach taste. So I have a lot of experiments I’m doing. For instance, I ask every one of my students to create a promotional website for a real local business. So I teach them what is called vibe coding. The idea of vibe coding is that you can use natural language to describe what you want, and the AI tool will execute your words and transform them into code. So anyone can now create a website directly from text interactions.

I asked my students to do that for real companies, so they actually help local businesses while learning how to vibe code. But I have something interesting, a little twist. At the end, when my 100 students have generated their websites, and they spend a lot of time, many of them spend many hours creating their website, I use a machine learning method to compare all their websites and show them that many of them converged to the same designs, the same user interface, the same user experience, even though they are doing vastly different things.

And why did this happen? Because they all use the same tool. That’s on purpose. I have them use a tool called Lovable. That is a really good one for web coding. So it’s a big moment of realization for them that these tools push us into a given direction. So if you want to distinguish yourself from others, to maintain competition, you need to provide that extra edge from your human side, to bring back that originality, that creativity.

And this is a factor that we have to maintain constantly, because AI will always tend to push us in a given direction, and sometimes we do not realize until it’s too late, or we don’t even have a realization that AI did push me in a direction that would have been otherwise different. So this is an example of what I teach the students, how I can teach them taste, and maintain that vision throughout the whole process.

I’m trying to vibe code myself. I am building a to-do list that uses stoicism and a lot of research from psychology, especially for people with ADHD, because I have ADHD. There was an issue with the UI. I told Claude to research the 100 most award-winning website design things, and fix the UI issue. I let it research for about an hour. In the end, I wasn’t happy with it, so I just thought of a solution myself. Is that the type of thing you mean to give that kind of edge? You need to provide a solution yourself? Or is that not enough? Is there anything more you need to put into it?

The example you described is one of the many things we need to do. In your situation, you asked AI to figure out the design. The problem is that AI does not really have eyes, feelings, or flow when it visits a website. It’s very different from us. When we see an interface, many things happen in our brains, very multimodal, actually. The AI is not feeling or processing in the same way. It just does pattern recognition based on whatever people consider good, and it will try to match that to whatever you ask.

However, AI does not have a good sense of your personal taste. There are so many types of ADHD; there is a whole spectrum. Many of my students also have ADHD, and many of my students also create to-do lists like you. What I realized was that the most interesting thing is that it’s designed for you. You’re creating a super personal helper, and no one else has the type of ADHD you have. The idea then is, what do I really want? What would actually help me the most? AI won’t be able to figure out that fine detail for you necessarily.

It can create a scaffolding, and it can help you see 10 solutions. Because what you can do, and this is what I suggest to my students, is generate 10 different possibilities for the same website, such that you get a few ideas from here and there, and you don’t settle down on one specific design. Because AI tends to anchor you in a given answer. So if it anchors you, you tend not to look through the whole range of websites, at the whole landscape of what’s possible.

So this is really it. When I tell the students, you have to maintain these different cues throughout the whole building process to make sure you never settle too quickly into something that might be suboptimal.

So your example is a really good one, where you realize, well, AI isn’t at a level yet where you can figure it out. And that’s also what I tell my students, it’s called the jagged frontier of AI. The task you asked for was outside of the comfort zone of the current tools. Within one year, if you ask the same question, you’ll get better results, as AI’s ability to go online, check websites, take screenshots, and understand what are good ideas has improved. So this is just evolving constantly.

Today, your taste was necessary to solve it finally. In one year, AI may actually be able to answer your question, which means: what do I do as a human if AI can do even more? But there is always something more we can contribute or more projects we can start. You can build a to-do list for you, but you can also build a list for others. Or maybe you make even a business. The personalized to-do list, where you arrive, and then the system will more quickly adapt to whatever is best to help you with whatever ADHD you have. So there is a business opportunity there. That could be the next step, or next year if AI is much better.

When Narek reached out to you in his outreach, he named you as someone rebuilding the curriculum so that graduates do not get squeezed out of entry-level roles. I want your comment on that, on what your response is to that, and also, from where you sit, what are other schools missing in their own AI response, if you can name the slowest parts.

Absolutely. Right now, we are witnessing a transition in the workplace where everyone wants to adopt AI, sometimes at any cost. They want to change the culture to make sure that we have, for instance, AI pods or an AI-native mindset. Many companies pivot. There is this whole landscape of change, and you have multiple strategies for navigating that change and the AI era.

Some companies decide to lay off employees so they can create that culture change, reorganize the whole company, potentially rehire later, or allocate this money to AI tokens. If you fire one employee, you might get more money to buy more tokens for other employees. To be clear, I am describing a strategy we observe in the market, not one I endorse.

Right now, many companies realize that by encouraging token-maxxing, which is the practice of spending as many tokens as possible, their employees start doing lots of new things, some not so new but cool things, but also it’s very expensive. So now we’re starting to realize, well, we cannot sustain that level of computing spending. It then creates a sudden compression. People realize it’s actually expensive to use AI, and it might sometimes be more expensive than having humans with taste, who won’t rely on AI to read 100 websites.

Here, by asking the AI to read 100 websites, your AI spent a lot of tokens and actually took maybe an hour, you told me. If you now have a UI/UX designer, they know immediately what’s required. In 10 minutes, they might tell the exact prompt needed to get the right change. So compare: you do not necessarily have the skills in UI/UX, but you know AI can do it, but it costs an hour of tokens, and then it was not good. Now you have a UI/UX designer, it takes 10 minutes, and you know it’s good.

This little example is actually emblematic of the new situation. When should you dedicate to an AI? When should you need a human? And those are very interesting open questions. We don’t yet have a playbook for how to organize this whole new workforce. So that’s aspect number one.

As aspect number two, new jobs will be created around interacting with AI. And it’s scary because it can take time for these new jobs to appear. There is a fancy job now called a forward-deployed engineer. The idea is that because AI models are not perfect yet, you need employees, for instance from OpenAI, to go to the companies that buy OpenAI’s models, to actually figure out how to use them. You have really smart models, but they’re hard and confusing to use.

For instance, you asked Claude Code to go through 100 websites. I would tell you, before you do that, that it won’t work, because I know that it’s outside of what AI will comfortably be able to do. It’s not good at finding 100 websites, then building it for you. Not yet. I expect it will be something it can do well next year. So it’s constantly evolving, meaning you may need to reorganize your workforce and develop new skills for your employees.

So I tell my students, I want you to be AI champions. I also want you to understand how to teach those tools to others, so you can lead your future teams toward the proper mechanisms, adoption, and collaboration with AI tools. So that’s the second thing, that they can lead those new teams, becoming AI champions. There will be a need for many AI champions.

Third, we are going to see many solopreneurs or very small teams that build very successful startups, because indeed you don’t need as many roles to create products very quickly. It used to be very expensive to launch a startup. Now, with a few hundred dollars or maybe even without the need for any money, you can use just free models and start building. You can create a website, a prototype, in a matter of hours. And this is a completely different game. It brings the ability to build, read, and solve problems into everyone’s hands. So I’m hopeful in that sense.

Fourth, many new issues due to AI. Cybersecurity is a big one. The consumption of electricity and water. Every time we use millions of tokens, this has an environmental footprint. Many people don’t think about that, but there is an opportunity to see how we can find more effective, smaller models. Those are potentially new jobs, new business opportunities.

So that’s my answer. There are a lot of things we can do. It will be a bumpy ride for sure. No one can really predict exactly what will happen, but I want to be on the hopeful side, on the side that encourages the responsible use that can empower us. But it also takes, first of all, leadership; it takes bravery sometimes to try new things. So I want to instill that culture in my students so they can go out into the world and spread that positive energy, so that globally we bring a mindset that will shape the society we want, not the society that feels scary.

Your Gen AI and AI for Business Applications course ends in a demo day, judged by people from Microsoft, Amazon, Google, and Meta, among others, exactly the firms hiring your graduates. Walk us through what a student is doing in week one versus the Saturday morning of demo day. And what are those judges actually hiring for now, that they were not three years ago?

This year, I taught two courses. The first one is Gen AI and AI for Business Applications. The second one is Advanced Machine Learning. I reshaped the whole curriculum. It’s only 10 weeks for the students, so it goes very quickly.

I used to teach fundamentals at the same time as I taught how to use AI models, but I realized that students first need to feel the magic and excitement of using AI. So from now on, and it was very successful this year, I start with just building. I teach my students how to build, how to ship products, and how to create agents. And they feel so excited, because now they feel they have the power in their hands.

That’s the first course. But as they build, they also start having deep questions about how does this work? Why is it not working? Earlier, you were describing a situation where you asked the AI to go through 100 websites and build it. It did not work. So I tell my students why. I teach them what happened. I teach them the process of looking for information online. I teach them that the prompt you wrote may not have led to the desired outcome. So those are things I teach them in the second course.

I also teach them how to teach others: why does a large language model sometimes fail? Why is a system prompt important? How do you consume fewer tokens? How are those models trained? This gives them greater maturity in using it.

Now, coming back to the first course. The first course is all about building. My students have five weeks to each create a real product that must be live, deployed online, and accessible to everyone. And on that Saturday, they have an agentic fair, where they all receive a table and some space to showcase their products. And they’re free to actually deal with physical objects, to bring some props, to bring their own large screens. So they have to design an attractive booth. It’s like a marketplace of projects.

And then I invite alumni from our program to help create continuity across years. Like this, you are excited to come back the next year to keep learning and seeing what new generations can do. So there is a legacy that keeps building. That’s important because there is friendship and support in the room. And the students then have an hour to present their work to whoever comes to their booths.

They have to demo live in a way that feels very motivated, and it has to be quick. So those skills are very helpful, because more and more in the workplace, one must demonstrate what they’ve been building. One must iterate on feedback very quickly. So there is this whole culture where judges can see 20 to 50 projects in a couple of hours. This is amazing because it brings them some creativity in the workplace, and we also hear them say, wow, we actually lack these kinds of skills, we wish to have someone as dynamic. You are actually solving a problem we are facing. And that creates a bond. I’ve seen a lot of connections form, and they can go for coffee chats or keep in touch after demo day ends.

So that’s aspect number one. Aspect number two: the judges. They typically work in the local ecosystem, in areas that may hire. They work at Amazon, Microsoft, Google, startups, Costco, many others. They face challenges in the company with AI adoption and upskilling. Once you leave school, it can be a little challenging sometimes to keep up to date.

When they see my students being AI champions, they feel, we want someone like these on our team to bring that energy and diffuse the knowledge. However, there is a challenge to fix: there is not always headcount. Those companies wish to hire and change the organization, but you need to create a job posting. And those job postings oftentimes are framed from the past, the way we would look for someone, although now the work has evolved. So it is a challenge that’s happening right now, and it’s why it’s hard to be hired.

So I’m trying to show the students here how they need to advocate for the new positions they would have in companies, and I really believe in that. You need to be very multidisciplinary. You are a product manager, you’re a technical product manager, you’re able to lead a team, you’re able to do marketing, you’re an engineer, you’re everything all at once.

Now, that’s why I bring the judges from the industry into the classroom, so they feel it, so they realize that we need these all-at-once capabilities, and that they don’t have them in-house so that you can hire those students. I can also tell you that what we’re looking for are these people who know how to create systems very quickly. We want more and more of these multidisciplinary engineers who can quickly understand what the latest tools can do and ship them into production, while also knowing how to talk to customers and other team members.

This is a need, but the challenge is that companies are still organized from a past perspective. And that’s why right now it’s the bumpy road. So I suspect that it’s going to stabilize in a few years probably. Still, right now it’s really about this one-on-one ability to advocate that, I can be the right person for your team, or you start building on your own until someone finds you and sees, actually, we want you, we will make the efforts to bring you in, which is in a bit of this networking-plus-building-in-public way of acting.

I also encourage my students to build in public so they can be seen and witnessed, and ultimately catch the attention of someone who realizes they need you and will create a job posting. And that’s challenging because it feels uncertain, scary, but that’s why I tell my students, trust your skills, trust what you can do, and be resilient. Keep pushing.

My final question, which is two questions in one. You and a few other people that I’ve spoken to live in this bubble that is on the front lines of technology and AI. I’ve spoken to friends that say, oh, our university bans AI. And if you actually look at the numbers, most have no idea what’s going on. They either ban it. They have no policy for it. What would your message be to the world of higher education when it comes to the attitude against AI? And then, if you had a crystal ball to predict the future, what would you say a business degree would need to certify in, let’s say, 2035?

My take is that schools that ban AI or discourage its use come from fear. A fear that our students won’t learn anymore. And I understand that fear, because there is this cognitive surrender. You want to learn accounting, you want to learn finance, you want to learn math, and AI already knows way more than you do, and maybe ever we know. As a result, you’re thinking, why should I even learn that? You’re being discouraged. So schools don’t want their students to surrender their faculties. And they adopt the approach of, let’s just get rid of it, let’s deny it, or let’s just maintain the old-fashioned way.

The merit of this approach is that you really want your students to make the efforts, and we need those efforts to feel the science, to ground it. Learning is sometimes painful. We need to go through that process. I do not like this approach, though, because AI will always be here from now on, whether we want it or not.

And my recommendation would rather be to find an enlightened use of the technology, a responsible use, where you maintain that learning for your students in a rigorous manner. Still, you also encourage them to be creative with the technology, so they can solve problems and complete tasks that were much harder before. That’s why I revisit my curriculum every year. What I was asking the year before is now too easy. So I need to ask my students to do harder things, which also means there are elements I don’t teach anymore, although they’re important. They can learn that on the go, if they realize it matters. But what I want to teach them is how they can solve really big problems.

I work to solve the global goals: poverty, hunger, climate change, and education for all. There is no shortage of big problems to solve in the world. So I tell them, let’s see how AI can support us in that mission. Here’s the approach I recommend: be more ambitious with what your students can do. Have them use the technology to do things you might not even know how to do yourself. But as AI will help you, many mistakes will be hidden. Some will be visible, some will be invisible. You need to teach the students that this matters. How can you catch an AI mistake in the process? How do you make sure you supervise these AIs? How do you manage a team of AIs? How do you properly communicate your intent?

Those are cool skills that one must master in the era of AI. How do you use ChatGPT versus Claude? How do you vibe code a website? There are so many things that can enter the room. So my recommendation for schools, then, is to do an introspection into why you fundamentally want to teach. How can AI support that? How can students try, for instance, to do an accounting sheet with AI to figure out little mistakes? What are the things that AI will catch, and that AI won’t catch? And if you’re able to show the students, look, AI got 90% of the mistakes, it’s amazing. It makes the 10% of them. These 10% matter a lot. How do you get AI to potentially catch those 10%, or why do you still need a human in the loop, and to even know there is 10% mistakes? You actually need to know the fundamentals. So I always teach my students why the fundamentals matter, and I illustrate it. I make it very visible and poignant because then they will remember. So that’s an element of my answer.

Another element of the second question you asked is that, I believe, our schools should teach our students to lead and manage. This does not change. The good old way of knowing how to shape a team and build a vision is as relevant as ever. I believe we should also teach how to build, how anyone can be an engineer. For instance, in business schools, maybe we do not emphasize enough that anyone can be an engineer as well, thanks to AI technology.

I’m hopeful for a golden era where everyone learns more skills and becomes more interdisciplinary, which will give us the ability to solve more challenging problems. And we are seeing a lot in AI for Science, for instance.

Some schools are doing well: the ones with resources, teaching in dynamic ecosystems where there’s a lot of technology. They naturally adopt and push it. It’s not easy, though, because many faculty members need to learn to use the technology themselves, and adapting can be daunting. When you’ve been teaching the same way for many, many years and suddenly you’re asked to update everything, that’s hard. So the pushback is often psychological, from the school’s perspective, or even from the students’.

My solution is to bring back the joy, to bring back the feeling that, as kids, we love building and exploring. I bring that into the classroom, and I also show it to my colleagues and other schools. I mentioned earlier that I visited many schools over the past month. That’s what I wanted to do. I went to all these universities: Stanford, Berkeley, Columbia, Chicago, UBC, MIT. I went there to show how I bring joy to the students, how they actually build and get excited, and how you can do it too. Anyone can do it. The tools are for everybody. That mindset, I believe, will bring beautiful things.

And I also want everybody to feel responsible. The carbon footprint, the water footprint, these matter too. So we need to educate. Not in the way that it should be, let’s not use AI, but how do you bring the responsibility throughout the whole process? How do you influence policy to build the world that we believe in?

So my crystal ball is, above the bumps in the roads, it’s going to be tough to keep up with smarter models. They’re just getting better every day. We are never ready for the next generation of models. A human side will matter so much.

I also predict that more and more people will start seeing so many cool applications. More young people who build entire companies and businesses so quickly with resources. The world may get more tools to solve the most challenging problems. I am hopeful we will properly address the carbon footprint, but it may take many years. In the meantime, there will be potential destruction of the environment, pressure on the electric grid. These are things we need to address. And cybersecurity is a big issue. There will be a lot of outages and challenges that we’ll have to go through.

But I’m a hopeful person around all of that. I believe we are going to discover new drugs, new materials. We are going to better understand the complex world of weather and nature.

I believe students will become AI-native. Right now, we’re in the world where the people around us did not know AI when they were born, when they grew up. But the new generations, the students who were born in the past maybe 10, 15 years, and especially those born now, will be AI-native. Remember the era of the internet, where people had to adapt. It’s the same thing. We’d have students who are so natural and have such an incredible intuition of how things work. Their brain will not operate the same as ours. And we cannot predict what will happen.

My suspicion is that they won’t see the barriers that we see, which will be good in the sense that they will make things happen. There is this famous little saying that is very cute. They did not know it was impossible, so they made it happen. And they will make mistakes that we won’t make, because they have no idea about, oh, that’s actually not correct to do that. But then we learn from those mistakes. The learning process will change.

So ultimately, I want to spread that message of joy, of hope, but also of responsibility, that we shall be the ones making it happen in a good way.

*This interview has been lightly edited for clarity and readability. The interviewee reviewed and approved the transcript before publication.