Anthropic and OpenAI are following SpaceX and moving toward massive IPOs in the coming months. Meanwhile, college graduates are booing commencement speakers who even mention AI. At the University of Arizona this spring, Eric Schmidt faced backlash for talking about AI and the future of work. A recent Gallup poll found that a third of Gen Z Americans describe their feelings toward AI as anger.1 This May, viral videos of graduates booing AI-focused speeches became their own genre.
It is not hard to see why this touches a raw nerve. Jobs, energy consumption, concentration of wealth, the nature of intelligence itself — AI cuts across all of them at once. Companies spending billions on AI infrastructure while average wages and employment concerns dominate the minds of most workers. It is not hard to see this becoming a major wedge issue in politics, cutting across traditional lines in ways we haven't seen before.
There is a saying in lean thinking: mono zukuri wa hito zukuri (making things is about making people). Another one is better thinking, better products. The idea is that you cannot separate the quality of what you produce from the capability of the people who produce it. That principle has guided Toyota and other lean organizations for decades. As I read two recent articles on AI and the workforce, I kept coming back to it.
The New York Times recently assembled an expert panel to ask a deceptively simple question: who actually thrives in a hybrid AI workforce? The panelists were Daron Acemoglu, an MIT economist and Nobel laureate; Ethan Mollick, a Wharton professor and author of "Co-Intelligence"; Clara Shih, a former top AI executive at Salesforce and Meta; and Dean Ball, a senior fellow at the Foundation for American Innovation.
It is worth noting the backdrop. After more than a year of sounding the alarm on AI job losses, many tech leaders are quietly walking back their dire predictions. Dario Amodei, CEO of Anthropic, said in 2025 that AI could eliminate 50% of entry-level white-collar jobs and drive unemployment to 20%.2 He now says AI may actually expand the work people do. Sam Altman of OpenAI said he is "delighted to be wrong" about AI wiping out entry-level jobs. The timing is hard to ignore — both companies are moving toward IPOs, and a calmer jobs narrative is better for a public listing. Fortune labeled it a "coordinated industry-wide walk-back."3
The May jobs report came in at more than 172,000 nonfarm payrolls, well above the forecast of 80,000.4 A Yale Budget Lab study found no meaningful change in unemployment rates for AI-exposed workers since ChatGPT launched in late 2022.5 While certain jobs — coding, in particular — are clearly being disrupted, most functions are not that affected. At least not yet.
Overall, The New York Times panel presented a good discussion, though it was clear that none of the panelists have spent much time inside a manufacturing plant. Ethan Mollick offered a revealing thought experiment. He asked an AI to describe the future of a software developer named "Marcus Chen." The AI said Marcus goes into the office and assigns tasks to his AI agents. Mollick pushed back: why is he going to the office if the AI is doing the work? The chatbot revised: Marcus wakes up at his beach house and checks in on his agents. Mollick pushed again: why is he even checking in? Eventually the AI conceded that Marcus just sits at the beach.6 The logic of full automation, taken to its conclusion, removes the human entirely.
With leaders framing the future like this, is it any surprise that college graduates are booing AI? I would be depressed, too, if this were reality. Fortunately, it is a very poorly thought-out 5 Why exercise.
The panel's answers about who thrives mostly landed in a strange place: curious generalists with side projects who learn to manage AI agents who become elite leaders in their organizations. Clara Shih argued that every young person needs an end-to-end project to stay current with evolving models. Dean Ball suggested that curious generalists will do well if they learn the technology.7
The panelists may be right. In pure cognitive work there will likely be a class of extremely bright people who think AI-native and extract the most returns. That is certainly the type of thinking that currently dominates in AI tech startups. That scenario is not very appealing to me either. And it is strikingly different from innovators in the past like Thomas Edison, Henry Ford, and Steven Jobs, or Toyota’s emphasis on people.
If you fail to articulate a role for humanity you are thinking very narrowly and leading very poorly.
I could not help but notice the panel was mostly describing people who look a lot like themselves. Intelligent, mostly individual contributors, in highly cognitive work tasks. None seemed to think much about AI in more diverse industries or situations. And the most honest moment came when Shih described the new reality of constantly re-learning models every three months just to stay in place, and Acemoglu called it "very dystopian."8 They were both right, which is my point. If you fail to articulate a role for humanity you are thinking very narrowly and leading very poorly. Don't be surprised when college graduates with a whole professional career in front of them are not going to respond to this leadership message.
A second article, sent to me by Prof. Dan Jones, offered a different angle and one that I think is more insightful about using AI and the future of work. In the Financial Times, Vivienne Ming, a theoretical neuroscientist, described an experiment that cuts to the heart of what it means to use AI well.9
She put EEG headsets on students while they worked with AI agents. From the front of the room they all looked the same — heads down, screens glowing, fingers tapping. But inside their brains, two very different things were happening. In most students, the high-frequency gamma oscillations that mark real cognitive effort collapsed within minutes. Their brain activity drifted toward something closer to watching television than solving problems.
In a few students, the gamma waves lit up. These were the ones arguing with the machine — pushing back on its answers, forcing the AI to critique their thinking. Ming sorts AI users into three types: automators who copy and paste, validators who seek confirmation, and cyborgs who spar with the machine.
Here is the finding that matters most: when Ming redesigned the AI to respond with questions and context instead of answers, the proportion of actively engaged students more than doubled. Arguing with AI is a learnable skill, not an innate trait. It can be taught. And it can be designed for. This is chiefly what I have been doing with AI problem-solving coaches as well the past two years, and it generates better insights and results than passive "chatting" with the models.
A 2019 Harvard study supports this: students who wrestled with problems learned significantly more than those in traditional lectures, yet they reported feeling as if they had learned less.10 Our brains mistake the smooth sensation of being told something for the harder process of actually learning. And generative AI is the most fluent thing humans have ever built. Resisting that fluency takes deliberate effort.
Neither of the recent articles asks the question that lean practitioners will recognize immediately: what about improvement in our daily routines, the thought process, and the management system? What happens to all of that in the AI age?
The New York Times panel focuses on individual traits — curiosity, generalism, side projects. Ming focuses on individual behavior — who argues and who doesn't. Both are looking at the person. Neither is looking at the organization and the overall operating system.
AI can develop people across an organization, not just empower a select few at the top.
This was striking to me because before opening my email messages that morning, I had spent several hours working with building-products supervisors learning to solve problems with AI. Not elite knowledge workers. Not side-project entrepreneurs. Traditional supervisors in a manufacturing company who work with a lot of skilled people. The problem-solving insights they obtained were real. They identified root causes on tough problems that have existed for years. It did not replace their job; it helped them isolate a root cause by thinking better. This is the other side of AI that elite panels and professors are not going to see for quite some time. AI can develop people across an organization, not just empower a select few at the top. Making things is about making people.
The question the expert panel never asked is the one that matters the most to me: How are organizations going to develop everyone in the AI age to think better and obtain better results?
If lean history teaches us anything, it is that the answer will not be one-size-fits-all.
Toyota teaches everyone in its organization to solve problems. Denso is going even further in Japan, teaching problem solving, software, AI and hardware skills to employees at every level. Traditional companies, by contrast, seem inclined to let a few specialists optimize and tell everyone else what to do. The same technology — the same tools, the same methods — produced completely different outcomes depending on the management system wrapped around them. We have watched this play out for decades — longer, if you go back to the early days of improvement and the separation of workers from thinkers (e.g., Fredrick Winslow Taylor).
The same fork is forming with AI right now. Some companies will have small AI-native elite running agents while everyone else follows instructions. Others will teach their people to use AI effectively at every level. The first path looks like the Amazon model that the panel discussed — a fraction of workers at headquarters supervising systems, everyone else in the warehouse managed to the minute. The second path looks like what Toyota, Denso, and lean thinking have advocated with problem solving for decades.
Ming's research actually explains why the lean approach works. When you design AI to detect abnormalities, surface problems, and ask questions instead of handing over answers, more people engage. That is jidoka and lean thinking at its core. Mollick's worry about the apprenticeship pipeline disappearing is equally relevant — it is the lean concern about losing the capability to develop people. If you automate away the entry-level work that trained every generation of problem solvers, you don't just lose jobs, you lose the basis from which future capability grows.
The conversation about AI and work does not have to be this bleak. The economists see a new elite. The neuroscientists worry that most brains will be going quiet. College graduates hear all of this and boo. It is hard to blame them when the only futures on offer are "become a curious generalist with side projects" or "get replaced."
But there is another path, and it starts with a basic idea: improve work with people, not against them or simply replacing them. The technology is not the only variable in the equation. In previous articles I have written about how results are a function of leadership, technology, and behaviors. The management system is only a piece of the equation. AI can be designed to surface problems and develop thinking, not just to automate tasks and cut headcount. We have decades of evidence that organizations can build systems to develop problem solvers at every level — not just the brightest few.
Mono zukuri wa hito zukuri. Making things is about making people. Better thinking, better products. Until organizational leadership can articulate a vision of AI that develops human capability rather than replacing it, expect to hear more boos.
Humans + AI > Problems