AI-Wiki

What I tell my Stern Business School students on how to thrive in the age of AI

18 min read

What I tell my Stern Business School students on how to thrive in the age of AI

▶ Watch on YouTube · World Economic Forum · 32:59

For today’s undergraduates, the impact of artificial intelligence on the jobs market means massive job uncertainty. So how should young people prepare themselves for the unknown unknowns of a career in the age of AI?

“The uncertainty will be overwhelming for a fraction of them, but I think for a fraction of them they will look back in 10 years and say this was a period of such great opportunity,” says Arun Sundararajan, a professor of entrepreneurship, technology operations and statistics at New York University’s Stern School of Business.

This episode was co-produced with China’s CGTN and co-hosted by Xin Guan, CGTN anchor and chief business news editor.

TL;DR

A ~33-minute episode of the World Economic Forum’s Radio Davos podcast, published 10 September 2026, hosted by Robin Pomeroy and co-produced with CGTN — co-host Xin Guan, CGTN’s anchor and chief business news editor, puts roughly a third of the questions and supplies most of the China material. The guest is Arun Sundararajan, Harold Price Professor of Entrepreneurship and Professor of Technology, Operations and Statistics at NYU Stern, author of The Sharing Economy: The End of Employment and the Rise of Crowd-Based Capitalism (2016).

A scoping note the title does not give you. The title advertises student advice; that material occupies roughly 12:26–14:50, about two and a half minutes of thirty-three. Slightly under half the episode (from 18:43) is about AI governance, frontier-model national-security policy, and sovereignty — a body of material the title does not signal at all. Read this as a two-part interview.

  1. The ten-year scoreboard on his own prediction — work is moving out of employment, not out of existence. Sundararajan’s 2016 thesis was widely misread: “a lot of people, when they saw that subtitle, the end of employment, they assumed that I was talking about humans not having anything to do. That wasn’t the implication… it was really about a change in the arrangement of work.” The prediction was that work would be arranged “less and less in the form of jobs or employment and increasingly as non-employment work arrangements.” His figures: ~40 million Americans in some non-employment work arrangement when he wrote the book, over 70 million last year; and, attributed to the chief economist of JD.com that same morning, over 40% of China’s workers — where, Xin Guan notes, the term used is “flexible employment.”

  2. The technological-determinism fallacy, stated cleanly. On whether AI’s new capabilities mean immediate displacement: “just because AI can do what a human does or a machine can do what a human does doesn’t mean that it will instantly start to do it in the place of the humans. That’s the fallacy of technological determinism.” His conclusion: “I think that we’ve got some time. There are lots of other factors that make this transition more viscous.”

  3. A three-cause account of the entry-level slowdown, and an explicit refusal to adjudicate. He treats the slowdown as established and its cause as open: “I don’t think there’s a lot of debate, at least in the United States, that there has been a slowdown in entry-level hiring. I think there is a very vibrant debate about what’s causing it.” The three candidates he names:

    • Generative AI breaking the apprenticeship bargain. “A company would hire someone who was promising in an entry-level coding role, banking role, consulting role. They do fairly simple things — PowerPoint, spreadsheets, routine coding — in exchange for being sort of like these investments that would be made into them to create the next generation of partners… And now because the generative AI can do a lot of that sort of somewhat routine cognitive work, that contract is fragmenting.”
    • Uncertainty-driven hiring pauses. “There’s so much uncertainty that organizations are feeling about what will the role of human beings be in the future that a lot of them are just saying, well, let’s hold back. Let’s not hire people right now. Let’s wait for this to shake out over the next 2, 3 years.”
    • A COVID-era confound. “We reorganized and restructured a lot of work during the COVID shutdowns… I think that in some ways has lowered the return on these investments in entry-level workers, because it’s one thing if you’re in the organization 5 days a week absorbing from these senior people. It’s another if you’re in strategically two or three days a week.”

    His own epistemic warning: “anyone who tells you they know exactly what kinds of work AI is displacing and what’s going to happen in the long run, I would not take too seriously — but there’s no doubt that there’s going to be a lot of displacement over the next decade.”

  4. Role compression, and the muscle-then-mind analogy. Where work is delegated to AI, “the set of things that the human does shifts to verification, problem formulation, certain kinds of judgment, certain kinds of accountability.” He expects this “perhaps in western economies as well, but more so in Asian economies.” The historical rhyme: “100 years ago, muscle got automated. There was a certain kind of deskilling as well… the skilled machinist, the craft manufacturer — that kind of skill was put out of business by the machines and work became more routine. But there were other cognitive capabilities that started to command a premium.” What he will not do is name the new bundle precisely: “we haven’t put our finger on exactly what that bundle is, but verifying, being able to know when to trust the machine, what to delegate to the machine, how to exert the judgment that allows you to complement rather than be substituted.”

  5. The student advice — the title’s actual payload, in three parts. Asked what he tells his students: “I tell them to run for the hills. No, I don’t.” His framing premise is that AI is “a technology that is going to create an incredible amount of value… simply because the barriers to creating things of value have never been lowered so rapidly by any other technology.” The three prescriptions:

    • Be an entrepreneur while still a student. “If the barriers to creating things of value have been lowered, go out and start to create them while you’re a student.” Two payoffs: “it creates a portfolio of what you’re capable of” and “it builds that muscle that allows you to be flexible and resilient and adaptable.”
    • Be an AI force multiplier. “It’s not like companies will not want any humans in the short run. They may need fewer entry-level humans, but the ones that they need are going to be the ones who know how to dramatically increase their output using AI. And there will actually be a premium on that kind of human.”
    • Network, and exploit student status as a wasting asset. “There’s never a better time to do that when you’re a student, because you can always go up to someone and say I’m a student, I’d like to learn from you. Once you’re no longer a student that doesn’t work as well.”

  6. One-person companies — the China-side observation, and the composition change inside it. Xin Guan describes the OpenClaw phenomenon: “we have this open-sourced AI agent called OpenClaw… and it just went viral in China and people start their own company… we call it a one-person company, and he deploys several AI agents working for him.” The part she flags as the interesting change is who can now do it: “in the past, if you are IT student it’s easier to start a business because they know how the internet works, they know how to code. And now if you are study like literature, philosophy — but you can just tell AI agent what you want to do, tell it about your vision, and they’re just starting a company.” Sundararajan’s agreement adds a selection claim: “there’s going to be a lot more people who earn their livings as not working for someone else but creating things of value working by themselves. And the early indications are the people who can harness the agents certainly have a leg up.”

  7. Aging Asia inverts the threat. For “China, Japan, South Korea, where there has been both a potential and realized threat of an economic slowdown because of a labor shortage,” AI and “in particular the embodied AI, the robotic AI, is probably going to be a boon rather than a threat.” The mechanism he states: “the fact that we are seeing the kind of rapid progress that we’re seeing in AI and robotics today is going to mean that the GDP slowdowns that would have occurred are not going to occur in these economies.” Supporting datum, cited loosely as something he read: “more than half of the world’s industrial robots that were installed last year were installed in China.”

  8. Redistribution — and a demand on his own discipline. “There may be a call to come up with new ways to redistribute. And I’m hoping we’re going to get beyond the ideas of an AI tax or a universal basic income. We can get to some other ideas where I feel like the economists have to do better than tax or UBI.” The purpose he assigns to redistribution is transitional rather than permanent: to ensure “that as we get to that eventual destination of massive value creation, the process is not too painful for a significant fraction of the workforce.”

  9. On government’s role — a named institutional gap. Educational infrastructure, he argues, was built for the start of a working life: “K-12 education, college. What we need are infrastructures that allow people to shift, because entry-level work is what’s being threatened now, but there’s going to be a lot of mid-career transition in the pipe.” The obstacle he names is a prestige asymmetry: “most of the glory comes from creating Tsinghua University — not creating a network of community colleges that transitions people in their 30s and 40s.” He also wants governments to champion AI’s “non-income, non-wealth equalizing effects” in education, healthcare and access to opportunity.

  10. A fifteen-year transfer of governance power from states to platforms — and China as the counter-case. “In many countries that are not China, over the last 15 years we’ve seen a redistribution of governance power between governments and private entities” — over what content is accessible, who provides infrastructure, who is responsible for security. “A lot of those roles shifted away from government and towards platforms… and this happened de facto in many ways. As a consequence, if I look at the US today, a lot of governance has to be done by the companies, because we’re in a world where that is the system.” China he describes as “much more of a top-down approach to platform governance… more active algorithmic governance, where there’s better visibility into how the algorithms are worked and governed,” set inside “much stronger planned industrial policy that is much more broad-based, so that it’s not just a lot of innovation in a few areas, but across the board — robotics, solar, electric vehicles, not just generative AI,” with the stated effect of avoiding “over-capitalization in any particular slice” and “no disorderly allocation of capital.” Enforcement runs “through administrative directives rather than through a courtroom battle.” His prediction: “you’ll probably see an AI governance structure settle in China much sooner than in the EU, and certainly much sooner than the United States.”

  11. Mythos as the first important case study of US AI governance — read as two distinct phases. He calls it “a fascinating case study and one of the most important early case studies on how AI governance will play out in the United States.” Phase one is platform self-governance: the model’s “code-generating capabilities being so advanced that it posed a threat to the computer security… this capability could be harnessed to hack into any system,” and “the platform proactively said, as a responsible actor, we are not going to release it.” Phase two is state control: “the US government has decided that it may be a threat to national security for non-US citizens — not just people in other countries but non-US citizens — to have access to it, and Anthropic has made the choice to simply not make it available to anyone… in part because they have a number of non-US citizens who work for them.” His reading of what it signals: “we’re in unchartered territory when it comes to who’s going to be making the decisions… I expect that over the next year or two we’re going to have more incidents like this where we’re going to be making it up as we go along, and it’s going to be both the government doing that and the platforms doing it.”

  12. AI sovereignty, deflated into a layer-choice problem — plus a motive that is not economic. He sets it inside the longer digital-sovereignty story: “part of a nation’s strategic autonomy is going to be shaped by having some indigenous digital capabilities, or not being too dependent on a different nation state for critical digital infrastructure — and AI happens to be the latest such digital infrastructure.” The deflation: “it’s very unlikely that any other country is going to have anything resembling blanket AI sovereignty where everything is produced by the country. I personally don’t think that that’s necessary or pragmatic. I don’t think any country has full-stack AI sovereignty anyway — we have semiconductor dependencies, manufacturing dependencies, cloud dependencies.” What follows is a strategy question rather than an autarky question: “it’s really a question of what is the layer of AI that is most important for me to have strategic autonomy. Should I do it completely by myself or should I form an alliance?” And he flags a driver he did not expect: “an aspect that I’ve seen surprisingly be of considerable importance to a lot of countries is the issue of cultural autonomy, which has less to do with I want the technology to be domestically produced so I can preserve my bargaining power, but more about I don’t want a foreign technology teaching my second graders in AI-enabled classrooms, because I want the culture that they learn when they’re in school to be my country’s culture.” Xin Guan frames the same question as US wall-building versus Chinese open-source advocacy; he declines the binary — “there’s an argument to be made for openness. There’s also an argument to be made against openness from a security point of view. So it all remains to be seen.”

  13. On “losing a generation” — the closing answer, and the line WEF pulled for the description. “Compared to people who were in their position 10 years ago, there isn’t a predictable path to starting their careers. It’s something they’re going to have to make up as they go along. So it’s going to take a lot more resilience and adaptability. I do think that during this period of adjustment, the uncertainty will be overwhelming for a fraction of them. But I think for a fraction of them, they will look back in 10 years and say this was a period of such great opportunity — compared to 2016, where everything was stable and predictable.”

How this source touches the Warner & Wäger cells

  • contextual/external-triggers — the source treats two distinct external triggers. The first is technological: generative AI arriving at “non-routine cognitive” work, which he argues does not mechanically produce displacement (see the determinism passage at ~5:42). The second is geopolitical and occupies the episode’s back half: frontier-model export-style restrictions (the Mythos sequence, ~23:52), G7 trusted-ally access tiers, and the fragmentation of cross-border technology cooperation that Xin Guan raises at ~27:07. Both are conditions firms and states respond to rather than choose.
  • digital-transforming/redesigning-internal-structures — the apprenticeship bargain (~6:53–7:27) is described as an internal structure that is fragmenting rather than being redesigned, with the COVID-era shift to two-or-three-day attendance named as an independent structural cause of the same effect (~7:40–8:14). His prescription for government — infrastructures for mid-career transition rather than career-entry (~19:11–19:48) — is the same structural argument at national scale.
  • digital-transforming/improving-digital-maturity — the workforce-capability half. Role compression toward “verification, problem formulation, judgment, accountability” (~10:19–10:53) and the “AI force multiplier” prescription (~13:49–14:21) are both claims about which human capabilities an organisation should now be building, which is what this cell names.
  • strategic-renewal/business-model — the one-person-company passage (~14:50–16:09) is a value-creation-and-capture claim, not a tooling claim: agents make a firm-of-one viable as an organisational form, and Sundararajan extends it into a distributional prediction (“a lot more people who earn their livings as not working for someone else”). His 2016 crowd-based-capitalism thesis is the same cell at the previous technology generation — platform-mediated commerce replacing the vertically integrated employer.

Linked entities and concepts

Source quality

  • Format and provenance. YouTube upload of a Radio Davos podcast episode, auto-generated (ASR) captions, cleaned at acquire time. Not a human-curated transcript. Chapter headings are creator-supplied (17 of them), not inferred. The raw file’s notes: block lists every ASR normalisation applied.
  • Two ASR-derived proper nouns were verified rather than assumed. “Mythos” and “OpenClaw” both reached the transcript through automatic captioning; both were checked against external reporting on 2026-09-15 before being written here, and both accounts match the public record. The raw file records the checks.
  • Evidential tier: expert testimony, not measurement. This is an interview. It contains no original data. Every quantitative claim in it is recalled or second-hand, and should be treated as such:
    • ~40M → >70M Americans in non-employment work arrangements — his own figures, no source or definition given for “non-employment work arrangement,” and the definition is load-bearing (it plausibly spans gig platform work, independent contracting, and secondary self-employment).
    • “>40% of China’s workers” — attributed to the chief economist of JD.com, in conversation, that morning. Uncheckable as stated.
    • “>half of the world’s industrial robots installed last year were installed in China” — prefaced “I read recently that,” no source.
    • There are no numbers at all in the governance and sovereignty half.
  • Standing. Sundararajan is an academic working directly in this area and is describing his own decade-old published thesis against the subsequent record, which is the strongest part of the episode. He is also an advisor to the Internet Society of China, which he discloses in passing and which is relevant to how his comparatively favourable read of Chinese algorithmic governance and industrial policy should be weighted.
  • Framing to keep visible. The episode is a WEF–CGTN co-production. CGTN is a Chinese state broadcaster, and the co-host supplies most of the China material, including the one-person-company account. Nothing in the episode is obviously promotional, and the co-host’s questions are substantive — but the China-favourable comparisons here arrive through a channel with an interest in them, and the wiki should not treat the China governance passages as disinterested comparative analysis.
  • Scope. Full episode ingested, 0:00–32:48 of a 32:59 runtime.

Relationships

Supports

  • The Biggest AI Opportunity Isn't Replacing People | Stanford Economist — both state that entry-level employment has fallen. Brynjolfsson measures it in ADP payroll records for workers aged 22–26 in the most AI-exposed occupations; Sundararajan states it without measurement and names three candidate causes — generative AI absorbing routine cognitive work, employer uncertainty producing hiring pauses, and COVID-era remote restructuring lowering the return on junior hires
  • Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence — both address the age-concentration of AI-related employment effects. Canaries measures it by occupation-exposure quintile in monthly payroll records; Sundararajan describes the mechanism he believes sits underneath it — the fragmenting of an apprenticeship bargain in which juniors did 'PowerPoint, spreadsheets, routine coding' in exchange for investment toward senior roles
  • How Companies Are Building Their Own Intelligence | Sonya Huang, Sequoia Capital — both address what sovereignty over AI means in practice. Huang argues the enterprise case for owning rather than renting intelligence; Sundararajan argues the nation-state case and states that no country holds full-stack autonomy — 'we have semiconductor dependencies, manufacturing dependencies, cloud dependencies' — so the decision is which layer to be autonomous in
  • Hugging Face's CEO on why companies are done renting their AI | Equity Podcast — both address national positioning in AI and the role of open models. Delangue argues from download share that open-source leadership produces AI leadership; Sundararajan describes China's approach as top-down algorithmic governance situated inside a broad industrial policy, and predicts AI governance settles in China sooner than in the EU or the United States

Contradicts

  • AI and Jobs series: How entry level jobs are changing in an AI era — both state a position on whether the entry-level hiring slowdown is established. Brown, from a WEF–PwC employer survey: 'dare I say sometimes those headlines aren't backed up by evidence and data', citing organisations increasing entry-level hiring by 25%. Sundararajan: 'I don't think there's a lot of debate, at least in the United States, that there has been a slowdown in entry-level hiring. I think there is a very vibrant debate about what's causing it.' Same podcast channel, seven weeks apart

Graph View

Backlinks