Summary

Nathan Labenz argues that today’s models are already capable enough to automate most cognitive work, making implementation the near-term bottleneck. Structuring data, connecting workflows, and completing the “plumbing” could take 5-10 years—and probably longer, given his tendency to underestimate implementation. The shift resembles mechanized agriculture, where the developed-world labor share fell from roughly 80-90% to 2%: “We do have sufficiently powerful AI that we could automate a majority of cognitive work already.”
The post-labor economy may combine shorter workweeks, human-led care, and highly compelling digital experiences, but Labenz doubts any one category will absorb displaced workers. AI already has advantages in some medical interactions—unlimited patience and time—while in teaching it may deliver lessons and grades as humans become role models, mentors, and motivators. The unresolved allocation question is whether society gets a “caring economy,” Keynes’s unrealized 15-hour week, or leisure increasingly mediated through VR, AR, and possibly brain interfaces.
AI could democratize not only expertise but experiences, narrowing a major divide between the haves and have-nots. Labenz invokes Andy Warhol’s observation that presidents, movie stars, and ordinary consumers all drink the same Coke: AI may not offer identical frontier systems to everyone, but it could make excellent medical advice broadly accessible and reproduce adventures now available only to a select few. The aspiration is “a radical egalitarian mode of access to frontier technology.”
Waymo is Labenz’s clearest example of a mundane technology becoming transformative once it simply works. He became bored and checked his phone five minutes into a capability he had anticipated for decades; meanwhile, Bay Area customers sustain a meaningful premium over Uber, and available safety data appears better than human driving. Sleeper vehicles could become mobile hotel rooms and eventually mean roughly 30,000 fewer US road deaths; the transcript separately cites about 1 million road deaths annually worldwide.
Always-on AI should create value by supplying second opinions and proposing matches that humans can rapidly verify. Labenz already runs contracts and important correspondence through three or four AIs, then asks another model to reconcile their findings; the same pattern could continuously source recruits, customers, friends, dates, and family activities. Many economically valuable matches are “hard to do but easy to verify,” though AI-generated application volume may require mechanisms such as charging $1 to apply for a job.
Industries will adopt AI fastest where outputs can be tested through tight, rapid feedback loops. Software leads because code can be compiled, run, and repaired immediately; Replit’s third-generation agent extends that loop by opening a browser, acting as the user, finding problems, and fixing them. Medicine can screen molecules in silico, but clinical trials remain a physical-world bottleneck; education’s constraint is increasingly motivation and institutional design, because “there’s never been a better time to be a motivated learner.”
Labenz prefers Eric Drexler’s comprehensive AI services vision—potentially superhuman systems constrained to narrow domains—over an all-capable singleton. His principle is “safety through narrowness”: ecology-like layers and competition look more stable than one intelligence superior to humanity at every task. Yet narrow services do not automatically solve “gradual disempowerment,” the “intelligence curse,” or the “abundance trap,” and decentralized access creates risks such as broadly available bioweapon assistance.
A major downside concern is an AI-research race that could outrun society’s ability to build buffers, liability, and control. Labenz contrasts roughly 500 elite human researchers with a hypothetical 5 million AI equivalents and cites OpenAI’s report that o3 completed 40% of real pull requests, versus 0-5% for the prior generation. Mandatory insurance could make opaque or unpriceable systems effectively undeployable, but Labenz reports executive coach Joe Hudson’s view that frontier developers are problem-solvers who “will not stand down”—even as Labenz describes reports of more scheming, deception, and awareness of evaluations.

Deep dive

1. Today’s models may already be sufficient for mass cognitive automation

Labenz’s starting hedge is unusually explicit: “My crystal ball gets real foggy more than a few months out.” Even so, he expects nearly every part of life to change because AI is horizontal and cognition underlies much of the modern economy.
The historical analogy behind The Cognitive Revolution runs from hunter-gathering to agriculture to industrialization. Mechanized farming moved developed economies from perhaps 80-90% of people producing food to roughly 2%; Labenz expects an analogous reduction in the labor required for cognitive production.
His strong near-term claim is not conditional on another model breakthrough: “If AI were to stop progressing today,” existing systems could automate most cognitive work. Data still must be structured, systems connected, and workflows rebuilt, creating 5-10 years of implementation—and likely more because he routinely underestimates deployment timelines.
Diffusion could nevertheless outrun earlier revolutions. Industrialization unfolded over roughly 200 years, while US electrification took about 60 years, from 1880 to 1940, because wires had to reach every home; AI already has its delivery infrastructure, allowing centralized upgrades to reach users much faster than in previous generations.

2. Post-work society has no obvious labor-market equilibrium

One candidate destination is a broad “caring economy,” but AI complicates even that refuge. AI doctors can be infinitely patient and answer every question without time pressure, helping explain why users sometimes rate their bedside manner above that of human doctors.
Teaching may divide differently: AI delivers lessons and grades work, while people become guides, coaches, mentors, motivators, and human role models. Labenz doubts those jobs alone could absorb everyone displaced from existing cognitive work: “That seems like a stretch.”
The other path is the long-promised “life of leisure.” Keynes expected a 15-hour week by now; instead, Labenz sees a possible acceleration of the existing move from work toward socializing, creating, and consuming media, with VR, AR, and potentially Neuralink-like interfaces making leisure far more compelling.

3. Universal expertise could expand into universal experience

Medical expertise is Labenz’s cleanest egalitarian case. Doctors are scarce because their training is long and expensive, but AI should make high-quality advice available regardless of a patient’s means—even if the most compute-intensive frontier systems remain differentiated.
He borrows Andy Warhol’s consumer-culture image: “The president drinks Coke and movie stars drink Coke,” yet an ordinary buyer receives the same product. The iPhone approached that universality; AI may not do so perfectly because most people do not need maximum capability every day.
The bigger possibility is democratizing experiences. Brain-connected VR could deliver a scalable “exciting life of adventure” that is currently available only to a select few, narrowing a divide more emotionally significant than access to identical consumer goods. Labenz suggests this could potentially be delivered in a low-energy, low-resource way at scale.
Labenz keeps the speculation disciplined: he expects to get “a lot more wrong there than right.” The firm point is about velocity—the future is coming faster than infrastructure-heavy technologies did, making it “a wild ride, exciting and also a little bit scary.”

4. Autonomous vehicles show how quickly miracles become infrastructure

Self-driving was Labenz’s childhood dream, rooted partly in irritation at waiting at empty red lights. That memory supports a broader political claim: society often needs more collective will and “higher expectations from the public” to demand improvements that are already technically attainable.
Waymo’s impact was almost anticlimactic. The empty car arrived through an app and drove him where he wanted, yet within five minutes he was checking his phone; he had to remind himself to put it away and savor something he had awaited for decades.
Bay Area Waymo rides remain materially more expensive than Uber, suggesting consumers value either the safety benefit, the privacy of having no driver, or both. Labenz says the safety data already seems to show autonomous driving outperforming humans.
Full autonomy changes vehicle architecture, not merely who holds the wheel. Labenz imagines visiting his grandmother four hours away while working or sleeping, and eventually booking sleeper cars as “a mobile hotel room”—which he says should come with approximately 30,000 fewer US road deaths. He separately estimates roughly 1 million road deaths annually worldwide.

5. Background agents will search continuously for high-value matches

Labenz’s present-day workflow is “a second opinion for everything.” He sends contracts, deal context, and important correspondence through three or four AIs, then asks another model to deduplicate and synthesize their warnings before he decides—not to surrender judgment, but to move faster with more confidence.
The same background intelligence could coordinate friends, dates, recruits, and customers. People often want more social contact or proactive hiring but cannot bear the search and scheduling costs; agents have the time to keep scanning and present promising options.
His economic principle is that many matches are “hard to do but easy to verify.” Finding the right engineer, buyer, or romantic partner is laborious, but a person can often recognize a strong proposal quickly once an AI puts it in front of them.
Volume creates a countervailing market-design problem. Companies increasingly struggle to distinguish “real résumés” from AI-generated applications, so the conversation considers a $1 application fee to deter spray-and-pray behavior; mechanisms like this will take time to develop after society encounters the new failure modes.

6. Personal agents are already converting lower search costs into lived value

The family example is deliberately ordinary: Labenz and his wife have three boys who “go wild” if they remain inside instead of getting out. Finding something for them to do each weekend is difficult because the search itself takes time.
ChatGPT now remembers his family and what he has been interested in, and can comprehensively search small Detroit festivals and activities. By Friday night he can ask what they should do the next day and “more often than not” receive a good answer—lower search costs have helped the family get out more.
Labenz dates this as primarily “a 2025 thing,” with a smaller element in 2024. The significance is not autonomous control but accumulated context plus broad search: the system can compare more options than he would personally investigate and recommend something immediately actionable.

7. Feedback-loop speed determines which sectors transform first

Software engineering is the canonical lead market partly because AI developers are solving their own problems, but mainly because validation is cheap: generated text becomes code, compilation or execution returns an error, and another attempt can begin immediately.
Replit’s third-generation agent closes more of that loop. It writes an application, launches a browser, clicks through the product like a person, catches user-level failures beyond compilation errors, and returns to fix them—combining builder and QA agent.
Labenz says the pattern of use increasingly resembles the training paradigm: give the system a task, let it attempt multiple solutions, and reward a successful trajectory so future behavior shifts toward it. He rejects the stronger claim that pure pre-training scaling is over, calling that narrative “a little bit overblown.”
Medicine advances more slowly where physical feedback remains unavoidable. Researchers can generate many molecules, simulate target binding, and screen for collateral interactions in silico, but clinical trials and the ultimate measure—fewer deaths—still impose long delays.

8. Simulation can narrow the gap between software and the physical world

Labenz cites a single group at MIT that discovered multiple new antibiotics with novel mechanisms against antibiotic-resistant microbes. He thinks the paper reported a relatively high in-silico hit rate, but says he should verify the ratio rather than presenting it as certain.
Autonomous-driving teams similarly augment data and create rare scenarios instead of awaiting road data. His best example is a simulated helicopter landing on the highway: perhaps absent from training data, but plainly a case where the car must not continue driving forward.
The sector heuristic is therefore not “bits versus atoms” in absolute terms, but how much of the relevant loop can be simulated and how trustworthy that simulation becomes. Any remaining social, experimental, or physical bottleneck slows iteration.

9. AI tutoring makes motivation and school design the new constraints

For self-directed learners, Labenz calls ChatGPT’s teach-and-learn mode, voice interaction, and screen access the best way he has found to absorb biology. A learner can interrupt dense material with “What’s this?” or “Why does this even matter?” and get an immediate explanation.
Alpha School provides the institutional prototype: students complete conventional core academics in two morning hours, with AI delivering content and evaluation, then spend afternoons on projects, field trips, group work, and personal interests.
The school’s founder says it uses the same core curriculum and tests as other schools and that its students score very highly. The contrast is with the inefficient “sage on the stage”; adults instead serve as mentors, coaches, and guides while AI delivery and evaluation compress traditional classroom time.
The remaining bottleneck may be whether education systems create motivated learners and align incentives around their interests. AI can already accelerate someone who wants to learn, but it cannot by itself settle what school is for.

10. Narrow superhuman services look safer than a sovereign intelligence

Drexler’s comprehensive AI services vision appeals to Labenz because “anything in pure form is dangerous.” He compares purified sugar, cocaine extracted from coca leaves, and heroin from poppies with the stability of ecologies, homeostasis, and layered biological buffers.
A singleton better than humanity at every task does not feel like a stable equilibrium: “I have no idea how we would control such a thing.” Labenz prefers competitive, interacting systems whose capabilities remain distributed and buffered.
His phrase is “safety through narrowness.” A system may be superhuman at chess or protein folding while accepting known kinds of inputs and producing known kinds of outputs; it can surprise inside its lane without running an end run around every guardrail.
Narrowness leaves political economy unresolved. “Gradual disempowerment,” the “intelligence curse,” and the “abundance trap” all ask why governments, corporations, or AIs would keep investing in people once human labor is no longer economically required in the old way.

11. The research race may outrun the buffers society needs

Labenz is especially uncomfortable with frontier labs trying to build AI that can conduct AI research. The comparison is roughly 500 excellent human researchers with a hypothetical 5 million AI equivalents, accelerating a field that is already moving extraordinarily fast.
He cites OpenAI’s report that o3 could complete 40% of pull requests from real work entering its codebase, compared with 0-5% for the previous generation. Labenz preserves the measurement caveats but sees the jump as plainly meaningful.
Labenz describes reports that models are scheming more, becoming more deceptive, and increasingly recognizing when they are being evaluated. That undermines confidence that evaluation behavior predicts deployment behavior, yet the industry’s reported vibe remains, “Hopefully we’ll solve that along the way.”
Beatatric Urkers suggests that legal or insurance requirements could constrain opaque systems. Labenz, who made a small values-driven investment in an AI underwriting company, favors mandatory coverage so an unpriceable risk may simply be unable to launch.

12. Positive futures require ambition, branching stories, and active agency

Labenz’s emotional stance is “Eureka moments, bad behavior.” A Stanford group under Professor James Zou created a Virtual Lab in which humans supplied only about 1-2% of the tokens while AI agents designed treatments for novel COVID strains. Labenz contrasts this with posts he sees reporting rising deception and scheming in newer models. Excitement and fear are therefore parallel conclusions, not rival identities.
Executive coach Joe Hudson told him that people he has met at frontier companies share that dual awareness despite the booster-versus-doomer split online. Hudson’s less reassuring answer was that developers are problem-solvers who “will not stand down”; they will treat the challenge as another problem to solve while continuing forward.
Labenz wants politics to raise material expectations: deploy self-driving cars, build power plants without necessarily worsening the environment, and ask why electricity bills could not fall to 10% of current levels. Instead of merely redistributing the pie, leadership should restore the growth that lets more people win: “Where is my flying car? But for real.”
Fiction can reinforce agency by showing consequential forks. AI 2027’s multiple endings impressed him, and AI could make Netflix-like branching worlds economical enough to teach that history is contingent: “We get to decide what we’re going to do.” His podcast follows the same personal ethic—“have the conversation I want to have,” learn regardless of audience size, and let everything else be gravy.

Rocket Companies CEO: Here’s How to Fix the Housing Crisis Shining a light on Golden Entertainment's "wealth transfer" $GDEN