Let’s explore how technological shifts—specifically AI automation and remote work—are restructuring task allocation, changing the demand for early-career workers, and redefining the premium on human expertise.

Research Roundup

Workin’ the Chain Gang

How we organize AI at work is facing a quiet, structural conflict: AI productization vs optimal hybrid intelligence.

While my research argues that the ideal human-AI relationship is a tight, step-by-step collaboration where human and machine work side-by-side, each correcting and elevating the other, recent economic research suggests that market forces are actively working against this dynamic.

A new working paper models how firms deploy AI in practice, and it’s core finding centers on a concept of "task chaining". Because passing work back and forth between humans and AI creates high coordination costs (hand-off friction, context switching, translation errors), firms try to avoid interleaving humans and AI step-by-step. Instead, they bundle AI-executed steps into contiguous, uninterrupted blocks—or "chains".

While this makes sense for short-term cost optimization, it changes the nature of work to “human-in-the-loop” by pushing humans to the absolute boundaries of the workflow. Instead of collaborating with the AI to refine ideas, the human is relegated to supervising inputs and signing off on outputs.

This reveals a deeper challenge: hybrid intelligence isn't just rare because people aren't trained for it or because AI models aren't ready [1]. It is rare because firms’ economic optimization actively selects against it. The economic gravity of coordination-cost minimization works directly against the high-value, fine-grained human-AI interaction [2].

Interestingly, the authors’ model hints at where this logic breaks down: "comparative advantage" of humans on ill-posed problems fails if workflows are just chained factoryline metaphors. When tasks are deeply interconnected, trying to segment them into isolated AI blocks fails because the messy, interdependent handoffs are precisely where the highest-value, emergent insights are generated.

Right now, the industry primarily benchmarks AI models in isolation. This gives firms the confidence to build longer automated chains to bypass human coordination costs. By doing so, we risk systematically dismantling human-AI synergy to save on hand-off friction.

Are we building a future of collaborative partners or just optimizing ourselves out of the loop?

[1] Both of these are true, though. We absolutely do need greater human capital in the form of foundation skills and meta-earning, and we need AI designed for hybrid intelligence rather than autonomous artificial intelligence. But above human capital and AI design is management and policy change.

[2] Despite the fact that the biggest empirical gains from AI have been in high talent, experienced labor.

Double Team

Is AI truly replacing junior-level employees? A new paper challenges this common assumption, pointing the finger instead at work-from-home. Does lockdown explain it all away?

Analyzing a dataset spanning 243 million hires and 407 million job postings reveals that the white-collar jobs most exposed to AI are also the most remote-work-friendly. When both factors are controlled for simultaneously, the apparent “AI replacement effect” on junior hiring virtually disappears—WFH exposure better predicts the drop in early-career hiring.

The argument has real substance: remote work makes mentoring, supervision, and "learning by osmosis" incredibly difficult and expensive for firms. Companies are not necessarily choosing models over young people; rather, remote environments make investing in early-career talent look like a bad financial bet.

This is a vital data point to keep us grounded AND we should be careful not to dismiss the very real, structural shift AI is driving. Other recent data points to an unmistakable trend:

The "Three-Cent" Arbitrage: Recent research on task-level outsourcing indicates that for every $1.00 companies save on human contract labor, they spend only $0.03 on AI tools. This is a massive, asymmetric reallocation of labor spend.

The Rise of the "Senior-Heavy" Organization: A recent Harvard Business School and INSEAD study of over 2,900 Y-Combinator startups shows that AI-native firms operate with 25% fewer people. They are flatter, capital-efficient, and—critically—heavily senior-weighted.

When you synthesize these pieces of evidence, a worrying picture emerges: agentic AI tools act as a powerful multiplier for highly experienced, elite workers. A senior engineer, marketer, or designer with an advanced AI stack can now execute tasks that once required an entire team of junior assistants. [1]

This creates a systemic apprenticeship crisis. If remote work makes onboarding young professionals too expensive and AI allows senior staff to easily automate away the "busywork" traditionally given to rookies, the bottom rungs of the professional ladder are effectively being dismantled.

Even if remote work policies are the immediate culprit behind today's cooling junior job market, the long-term trend points to a permanent drop in entry-level demand. What happens when all industries become YouTube-style sink-or-swim talent pools? [2]

This is a human capital policy crisis. If corporations are no longer willing or able to fund the "apprenticeship phase" of a career, we must actively design new pathways for young professionals to gain experience, build skills, and transition into high-value expertise.

[1] This has been my experience since 2008. AI has been a massive force multiplier in my work long before LLMs came along. It’s that now everyone can join in the…fun? Well, for me…yes.

[2] See my SciFriday post on The Amazing Digital Circus (below)or the talk around Backrooms for the upside.

The Automation Antidote

𝐓𝐡𝐞 𝐀𝐮𝐭𝐨𝐦𝐚𝐭𝐢𝐨𝐧 𝐀𝐧𝐭𝐢𝐝𝐨𝐭𝐞. When we worry about automation displacing workers, we often comfort ourselves with the idea that technology will simply "create more jobs" (the Jevons Paradox). But a new paper from David Autor and colleagues says this misses the point. [1]

The new analysis of historical census files and modern ACS data reveals a difference between "more work" (expanding existing roles) and "new work" (the emergence of entirely novel occupational roles).

The research finds that:

• New work is characterized by scarce, highly specialized expertise that commands significant, persistent wage premiums.
• These wage premiums are not permanent; they slowly decline as the new expertise diffuses through the broader workforce.
• Crucially, new work serves as a powerful counterweight to automation-driven displacement—not by merely inflating headcount, but by generating entirely new domains of human expertise.

Historically, this premium-carrying "new work" has been disproportionately captured by younger, highly educated workers.

However, this demographic skew is likely less about age and more about meta-learning and life friction.

This elite human capital formation is not a sign that skills are easily fungible. Rather, it suggests that the core engine of "new work" is the foundational ability to learn how to learn. Younger workers tend to fill these roles first because they are less constrained by existing institutional commitments, allowing them to adapt to demand shocks with lower pivot costs.

The implication is highly encouraging for seasoned professionals: older workers with high meta-learning capacity who find themselves unburdened by rigid career commitments—serial entrepreneurs, victims of high-level tech layoffs, or individuals experiencing life transitions—likely look and perform very similarly to their younger, elite peers.

The premium wages in an AI-driven economy won’t just belong to the young; they will belong to those with the foundational meta-skills and the logistical freedom to rapidly build expertise in whatever "new work" emerges next.

[1] As have I for some time. Jevons Paradox says that as technology lowers the cost of a form of labor demand goes up, “paradoxically” improving the welfare of workers. The problem is no one seems to have thought about what happens if Jevons holds for demand but not supply.

Media Mentions

I’m honored to be cited in David Brooks's new essay in The Atlantic, "The People Who Will Thrive in the AI Age".

He draws on my research—neural markers of cognitive effort drop sharply in most people using AI—to make a larger point I care deeply about: when intelligence becomes a commodity, what distinguishes us is “volition”, the will to do hard mental work rather than outsource it.

In my research this surfaces as curiosity, resilience, intellectual humility, fluid intelligence, and more. Brooks’s conviction is the heart of Robot-Proof: if machines have all the answers, our job is to build better people. Read his whole article!

SciFi, Fantasy, & Me

The Amazing Digital Circus came to me sideways, through an episode of the How Did This Get Made? podcast, with no idea what I was in for. What I found was hard to pin down: comforting and challenging, experimental and amateur and slickly professional, sometimes all in the same scene. It's all over the place…and I mean that as praise.

A woman named Pomni wakes up trapped in a candy-colored virtual-reality circus with 5 other humans, all at the mercy of a manic AI ringmaster and their own buried traumas. Picture I Have No Mouth, and I Must Scream [1] by way of a children's toy commercial. All nine (short) episodes are out; I watched the whole run in one sitting.

A tiny independent team built a global phenomenon on YouTube—a billion-plus views, a theatrical send-off, no corporate studio in sight—and made something genuinely strange and personal in the process. Imperfections and all, this may be what the future of entertainment looks like.

[1] Definitely references in the show.

Stage & Screen

• July 7, MIT: I'm giving the keynote for the MIT App Inventor Global Education Summit taking place this year at MIT CSAIL.
• July 8, NYC: It a book talk for Robot-Proof at the Harvard Club...how swanky!
• July 14, SF: AI+ education means...what? We discuss at WESTED's annual board retreat.
• July 14, Napa: Deep thought about AI and Society.
• September 15, Amsterdam: How might AI change the world of investing?
• September 15, SF: Innovation Day with INSEAD!
• September 16, DC: AI and education–beyond dreams and dread.
• September 19, Phoenix: I'm giving the keynote for the Association of Science & Technology Centers annual conference.
• September 21, Stanford: We're still working on the details, but hopefully I'll be talking about my research on machine learning and neurodiversity for Stanford's Neurodiversity Project.
• September 24, UC Berkeley: It's my annual Berkeley Change-makers Lecture!
• September 24, NYC: Culture Shifting Deal Making Summit
• September 29, Cincinnati: Still baking...
• September 30, Irvine: Hybrid Intelligence for innovation!
• October 6, SF: UCSD Alumni Association
• October 6, SF: Giving a talk at the Draper Richards Kaplan Foundation
• October 21-23, Warsaw: So much good stuff is in the works for my first visit to Poland (and maybe time in Germany as well!)
• October, Toronto: The Future of Work...in the Future
• November 19, NYC: Secrets in the dark!

Vivienne L'Ecuyer Ming