April 11, 2026
‘The Generalist’ via PubsforSubs · email · 22 mins
Hanns Scharff, the Luftwaffe’s most effective WWII interrogator, never asked a direct question. While Nazi interrogators relied on threats and violence, Scharff took downed Allied pilots on casual walks in the Taunus hills northwest of Frankfurt. He extracted military secrets by saying something wrong and letting pilots correct him — telling one American that a chemical shortage caused white tracer smoke, the pilot rushed to explain it was intentional design signaling low ammo. The pocket was picked without the wallet’s owner registering a rustle.
John Nolan’s 1999 book Confidential is deliberately hard to obtain — and that scarcity is part of its mystique. Nolan spent 22 years as a spy before founding a corporate espionage consultancy; his firm allegedly orchestrated the P&G vs. Unilever “secrets of shampoo” intelligence campaign in 2000. The book is out of print, sells for several hundred dollars on eBay, and is prized by intelligence officials — 350 pages interleaving psychological technique with espionage history and corporate case studies.
Elicitation is defined precisely to exclude interrogation and interviewing. Interrogation targets someone who hasn’t admitted to having information and knows why you want it; interviewing targets someone who has admitted to having it and knows why you want it. Elicitation specifically “avoids direct questions and employs a conversational style to reduce concerns and suspicions” — the source neither confirms having the information nor realizes they’re being mined for it.
Direct questions trigger a defensive cascade that filters and degrades answers. When someone asks a sensitive question, the recipient runs a rapid internal loop: Why are they asking? What do they want? What do I gain or lose? By the time an answer emerges, it has been heavily edited. Cognitive psychology research cited by Nolan shows humans also remember questions more clearly and longer than ordinary statements — so asking directly flags your interests and makes the source more guarded in future interactions.
The “Purposefully Erroneous Statement” exploits humans’ near-compulsive urge to correct. Cunningham’s Law — “the best way to get the right answer on the internet is not to ask a question, it’s to post the wrong answer” — is the same mechanism. In Nolan’s factory-bid example, saying “you couldn’t possibly hit 500 units a week” produces the correction “we’ll be at 1,135 in two months,” plus cost structure and competitive bid details volunteered in further defense. A direct question — “how much will you bid?” — would have received a guarded non-answer.
“Quotation of Reported Facts” works at every knowledge level, including total ignorance. Rather than stating something wrong yourself, you cite a (real or invented) external report and let the source weigh in. Nolan tested this by telling three different employees that analysts expected their quarterly earnings down 20% due to a Kuala Lumpur facility problem. The insider corrected the figure to 8% and named Thailand as the real culprit; the mid-level employee revealed months of operational dysfunction in KL from firsthand experience; the outsider disclosed a two-week production halt and a pending Italian government contract. Each level of knowledge produced different but actionable intelligence.
“Provocative Statement” and “Quid Pro Quo” manufacture a cover story for the entire conversation. A mildly confessional opener — “I really wonder why I’ve stayed with my company so long” — invites the source to ask why, at which point the elicitor responds with a personal story (Quid Pro Quo) that prompts the source to reciprocate with their own. If the source later wonders how the conversation turned to sensitive topics, the answer is: they asked. The elicitor can honestly say the source initiated the direction.
“Criticism” and “Feigned Disbelief” exploit the impulse to overprove. Cable TV executives were unfazed by complaints about price or service, but criticizing programming content made them “launch into a defense” — industry-specific hot buttons can be identified and used. Persistent disbelief in one Nolan investigation — suggesting a fired executive was dismissed only for “wearing black socks with dress shoes” — ratcheted up the source’s corrections until the headhunter disclosed compromising information about underage interns on company property. The elicitor never asked; the source volunteered it to overcome perceived skepticism.
Introverts make better elicitors than extroverts, and naive beats dumb. Ego is the enemy: an elicitor who wants to demonstrate their own knowledge ruins the dynamic. Appearing naive positions you as a student, unlocking the other person’s desire to teach — the explosive-welding example shows a source explaining classified fabrication processes in detail because the elicitor played credulous. But there’s a narrow failure mode: one Nolan elicitor blew cover by accidentally using an industry acronym mid-session, revealing knowledge they’d claimed not to have; the source shut down immediately.
Word repetition and timed silence are the simplest and most overlooked elicitation mechanics. Restating a source’s words — using synonyms rather than verbatim parroting — signals engagement and invites elaboration with no apparent agenda. Paired with 10–20 seconds of silence after a whimsical observation, it reliably prompts extroverts to expand on whatever was just said. “Emphatic loading” — repeating the same phrase with stress on different words — generates entirely different conversational branches: “Jim Johnson?” signals disbelief, “The Congressional Medal of Honor?” signals suspicion, all without asking anything.
The most susceptible targets are also the most trusted insiders: frontline workers, professors, and salespeople. Factory workers and customer service reps have current, granular operational knowledge that management rarely tells them is sensitive — so they share freely. Professors have a professional identity built on informing others. Salespeople are extroverted and confident, which makes them poor defenses against a carefully planned conversational path they don’t notice being led down. Lawyers are blind to elicitation precisely because they’re trained to defend against direct questions; elicitation operates in their dead angle.
Everyday corporate espionage is far more pervasive than most targets realize. At trade shows, Nolan describes a company planting a fake bathroom attendant — older, vest-wearing, European accent — near the sinks, collecting conversations between men who checked for feet under stalls but ignored the “attendant.” The “elevator trick” had employees riding up and down for 30-minute shifts listening to hallway conversations that continued onto the elevator: people stop talking when someone boards after them, but not when someone is already there. Frederick the Great’s line applies: “It is pardonable to be defeated, but never to be surprised.”
Quotable:
“Under interrogation, nobody behaves normally. People who are stupid act intelligently. Intelligent people act stupid. The guilty look innocent as day, and the innocent look dreadfully guilty. And, just occasionally, people act as they are and tell the truth as they know it. Of course, they’re the wretched souls who get caught out every time. There’s nobody less convincing to our wretched trade than the blameless man with nothing to hide.” — John le Carré, quoted by Nolan on the failure modes of direct interrogation
The Information AM · email · 9 mins
Three senior OpenAI executives who built the original Stargate data center initiative have left or are imminently departing: Peter Hoeschele (already gone), Shamez Hemani (compute strategy and business development), and Anuj Saharan (compute organization leader). All three are joining the same unnamed new company. The proximate cause: late last year OpenAI hired former Intel CTO Sachin Katti as head of compute and infrastructure, reporting to President Greg Brockman — executives who previously reported directly to Brockman, including Hoeschele, were moved under Katti, and the departures followed.
Meta is forcibly reassigning its top engineers into a new Applied AI Engineering (AAI) division under VP Maher Saba, who reports to CTO Andrew Bosworth. In an internal memo seen by The Information, Saba designated AAI as “P0” (Meta’s highest priority designation) and confirmed that transfers are “not optional” — the team’s job is to generate training data, run hackathons over the next two weeks, and support Meta Superintelligence Labs (headed by chief AI officer Alexandr Wang) in reaching state-of-the-art model performance.
Fed Chair Jerome Powell and Treasury Secretary Scott Bessent on Tuesday summoned leaders of Citigroup, Bank of America, Wells Fargo, and other major U.S. banks to discuss cybersecurity risks from Anthropic’s new flagship model, Claude Mythos. Anthropic simultaneously restricted Mythos access to only 40 organizations — including Apple, JPMorgan Chase, and the Linux Foundation — for security vulnerability testing. A previously leaked Anthropic draft had warned Mythos could help hackers “exploit vulnerabilities in ways that far outpace the efforts of defenders.”
Mercor, the three-year-old AI training-data startup, crossed $1 billion in annualized gross revenue in early 2026, doubling from its $500M pace in September 2025 — after CEO Brendan Foody had called it “the fastest growing company of all time” (from $1M to $500M ARR in 17 months). Since 60–70% of top-line revenue goes to contractors, net revenue is $300M–$400M, and the company is profitable on a free cash flow basis. This milestone came just before hacking group Lapsus$ claimed to have breached Mercor via a supply chain attack on open-source project LiteLLM; Mercor raised $350M at a $10B valuation from Felicis last November.
Meta has committed to spend $35 billion total with GPU cloud provider CoreWeave through 2032 — a new $21B deal (2027–2032) on top of an existing $14B contract through 2031. This is on top of a separate $27B deal with Nebius for Nvidia chips, and Meta’s total 2026 capex guidance of up to $135 billion. Meanwhile, Amazon’s AWS chip business (Graviton CPUs + Trainium AI chips) is running at $20B+ annually; CEO Andy Jassy said it would be ~$50B if Amazon sold chips directly to outside firms and floated that selling racks to third parties is “quite possible.”
Quotable:
“The transfers aren’t optional.” — Meta VP Maher Saba, responding in internal comments when an employee asked whether engineers selected for the new Applied AI Engineering division could decline
Clouded Judgement by Jamin Ball · email · 8 mins
Stanford’s Meta-Harness study proves the orchestration layer around a model matters as much as the model itself. Keeping weights fixed and changing only the harness — the code governing what context a model sees, what it stores, and what it retrieves — produced a 6x performance gap on the same benchmark. The AI-optimized harness beat best hand-engineered solutions by 7.7 points on text classification while using 4x fewer tokens, reached #1 on an actively contested coding benchmark, and its discovered harnesses transferred across five models never seen during the search process.
Compressing context to save tokens costs 15 benchmark points at median. When researchers gave the harness optimizer raw, full execution traces (complete prompts, tool calls, model outputs, and state updates from every prior run) instead of summarized versions, performance jumped 15 points. Most teams default to summarizing everything to cut costs — this puts a hard number on how much performance that tradeoff sacrifices.
Anthropic launched Claude Managed Agents to own the harness layer on behalf of developers: sandboxed execution, context management, error recovery, permissions, and long-running sessions, priced at $0.08/session-hour plus standard token rates. Notion, Rakuten, Asana, and Sentry are already building on it. The strategic logic is clear — productize the orchestration layer to compound switching costs on top of model lock-in.
For vertical AI founders, Anthropic’s general-purpose managed harness is fine, but fine isn’t the same as great — and the 6x gap from domain-specific tuning is the whole competitive ballgame. The right build/buy split: purchase foundational infrastructure (sandboxing, auth, session management) but build the orchestration intelligence — which context to surface, when to retrieve it, how to handle domain-specific edge cases. The long-term path for the strongest application companies is: killer harness → proprietary data collection → post-trained model → eventually a pre-trained model of their own.
Meta-level: Stanford’s system used Claude Code as the agent that writes better harnesses — AI writing orchestration code for other AI systems, producing measurable results on real benchmarks. This is a practical, working instance of recursive self-improvement: agents improving agents, happening now in production engineering contexts rather than as a theoretical future risk.
Quotable:
“If a 6x performance gap comes from harness quality on a generic benchmark, imagine what a deeply tuned, domain-specific harness could do in your vertical.” — on why domain-specific orchestration is where competitive moats are built
Byrne @ The Diff · email · 8 mins
TV bookers choose pundits for entertainment value, not predictive accuracy — so the incentives that would create a pundit accountability system have never existed. Books like Philip Tetlock’s Expert Political Judgment and organizations like the Good Judgment Project measure average error rates across pundits, which tells you almost nothing about any specific one. Bryan Caplan’s model of demanding explicit wagers and publicly tracking them is more useful but requires counterparties willing to put money on the line and honest enough to honor the bet.
Vistadex, a new multi-market prediction platform, solves this by using LLMs to scrape every statement made by opinion-makers across all known platforms, extract predictions, match them to existing prediction market contracts (e.g. Polymarket), and make play-money bets on their behalf — automatically generating a continuous, objective scorecard for each pundit. The long-term implication is treating pundits as implicit fund managers: you could allocate a portfolio to “things Robin Hanson says will happen” as a tail-risk hedge alongside equities and bonds.
This business was technically conceivable twenty years ago but economically impossible: you’d have had to manually read every pundit statement, encode each bet, and generate prior odds for every proposition — two enormous, independent lifts neither of which was feasible alone. The LLM revolution and the maturation of prediction markets like Polymarket (neither of which was built with Vistadex in mind) crossed a threshold that flipped the venture from ridiculously low-ROI to potentially profitable and socially useful.
This is what long-run economic growth looks like at the micro level: as the count of discrete industries grows linearly, the count of potential connections between them grows quadratically, and each connection is a new niche. As IT and financial services get more efficient — cheaper to aggregate information, cheaper to execute trades — those niches get identified and exploited faster. Vistadex sits exactly at the intersection of LLM infrastructure and prediction market liquidity, and neither Anthropic nor Polymarket needed to anticipate it for it to become possible.
Elsewhere in the issue: OpenAI told investors that Anthropic’s capacity-limited Mythos launch proves Anthropic lacks compute — but this actually illustrates the structural logic of the AI business: the lab with the best current model has the least spare capacity to train the next one, so the smart bet is always on “the weakest lab that isn’t actually dying.” Family firms are more common in politically unstable economies (South Korea, India, Taiwan) because multiple balance sheets act as a financial buffer — LG’s chairman literally adopted his nephew to formalize succession. OpenAI’s new $100/month tier follows a common retail pattern: pick the price point first, then engineer a product bundle that justifies the margin. Iran now collects taxes in crypto and buys weapons with them, meaning war economies can skip hyperinflation by dollarizing via stablecoins as they lose ground.
Quotable:
“The more discrete industries there are, the more—by a big margin—potential connections there are between industries. And the cheaper it is to aggregate and respond to information, i.e. the more efficient the IT and financial services sectors get—the faster those niches will be identified and exploited.” — on why Vistadex is a microcosm of how growth happens
Bloomberg · email · 7 mins
PE firms are sitting on ~13,000 companies and refusing to exit. JPMorgan CEO Jamie Dimon flagged in his annual shareholder letter that private equity holding periods have doubled to 7 years — and assets held beyond 6 years are classified as “potentially stranded” by With Intelligence (an S&P Global data provider), meaning they face write-down risk from declining performance or inability to attract new capital. Despite stock markets near all-time highs in recent months, PE sales fell more than a third to ~$103 billion in Q1, as AI disruption fears and Middle East uncertainty spooked buyers. Sponsors who failed to hedge borrowing costs during the cheap-money era remain trapped: they can’t find buyers or float companies at valuations they’ll accept, and limited distributions from older vintages are making new fundraising harder.
Private equity’s entire track record was built on a 2,000-basis-point interest rate tailwind that no longer exists. Howard Marks of Oaktree Capital Management wrote this week that PE “was born and existed through 2021 in an interest rate climate that was supportive of it in the extreme” — a 40-year decline in rates that he calls “the most impactful event in the financial world in the last half-century.” Now that rates have normalized, sponsor returns will depend on actual skill in selecting, growing, and financing companies rather than leverage arbitrage. Direct lenders who financed these buyouts face cascading stress: struggling portfolio companies have turned to payment-in-kind debt (deferring cash interest) to survive, and when those loans mature, lenders face the choice of further extend-and-pretend or recognizing losses. Marks, paraphrasing Buffett, warned that “the tide has begun to go out for direct lending,” with “bare bottoms” about to be exposed.
Software’s dominance in PE portfolios is now a liability, not an asset. Software and technology services accounted for roughly half of all private equity deals in recent years — a concentration bet that produced market-beating returns for almost two decades. That premium has now collapsed: the tech loan premium in the leveraged loan market has completely broken down in 2026, private credit funds are turning away software borrowers outright, and falling leveraged loan prices historically precede stress spreading into private credit. Bruce Richards, co-founder of Marathon Asset Management, predicted on Bloomberg TV that direct lenders face three straight years of double-digit default rates in software, with recovery rates of just 0–30 cents on the dollar, partly because software companies that took on 8–10x leverage have no refinancing options.
Zombie funds are multiplying, retail investors are fleeing, and regulators are closing in. Hundreds of sponsors raised capital at peak multiples and now face the prospect of never raising again — With Intelligence expects the count of zombie funds to grow significantly. Carlyle Group’s $7 billion private credit fund has already been forced to limit withdrawals after investors requested to redeem 15.7% of the fund. On the regulatory front, the Bank of England is moving to restrict “funded reinsurance” — a tool PE-backed offshore reinsurers use to absorb corporate pension liabilities from insurers — warning in 2024 that PE’s growing role in life insurance could pose systemic risks. The US Treasury is simultaneously convening meetings with insurance regulators over private credit market exposure.
Quotable:
“Recovery rates will be zero to 30 cents on the dollar on your average software default.” — Bruce Richards, Marathon Asset Management co-founder, on the coming wave of direct lending losses in software
Bloomberg Odd Lots · email · 7 mins
March Core CPI came in lower than expected — helped by a 4.1% drop in legal services — but energy components surged dramatically: gasoline rose a record 21.2% and fuel oil jumped 30.7%. Airlines were the first downstream casualty, with airfares up nearly 3%, because jet fuel costs feed through almost immediately to ticket pricing.
WD-40’s Q1 earnings reveal the transmission lag that will hit the broader economy: the company says it takes 90–120 days for crude price changes to flow through to its cost of products sold due to production and inventory lifecycles. WD-40 doesn’t expect gross margin to be materially impacted until Q4 fiscal year 2026, meaning the current inventory buffer is masking the real hit — which hasn’t arrived yet.
WD-40 is hedging against sustained oil exposure by rolling out an 85% bio-based version of its flagship product in Europe first, with global expansion planned for the following fiscal year. If oil prices stay elevated, this reformulation could become a meaningful structural offset — though management cautioned it won’t generate significant revenue for multiple years.
The University of Michigan consumer sentiment survey dropped to a record low even as one-year inflation expectations rose — a dangerous combination, since inflation psychology can become self-fulfilling when workers demand wage increases to compensate for expected price rises. The divergence between soft headline CPI data and deteriorating consumer expectations signals the inflation story is far from over.
Quotable:
“There is typically a delay of between 90 and 120 days before changes in cost of raw materials impact our cost of products sold, due to production and inventory lifecycles… We do not expect that our gross margin will be significantly impacted until the fourth quarter of fiscal year 2026, based on current inventory levels.” — WD-40 management, Q1 2026 earnings call
Robert Armstrong · email · 6 mins
US small caps are beating large caps by 8.5% in 2026 — a striking reversal after roughly six years of chronic underperformance. The most straightforward driver is sector mix: energy, the year’s best-performing sector, has a 6.5% weight in the small-cap S&P 600 versus only 3.5% in the large-cap S&P 500, and smaller energy stocks (up 41% YTD) are outrunning their large-cap peers (up 29%), consistent with the principle that small companies benefit most in cyclical expansions.
The flip side of energy is tech: information technology is down 5% in the S&P 500 (software carnage partially offset by a semiconductor rally), and IT makes up a third of the S&P 500 but only 12% of the S&P 600 — so small caps simply carry less deadweight from the AI selloff. This two-factor story (more energy exposure, less tech exposure) would ordinarily be sufficient, but the data throw up a paradox: small-cap tech is itself having a very good year, outperforming large-cap tech, even though small software companies should logically be at least as exposed to AI disruption fears.
A second anomaly compounds the first: within small caps, growth stocks are leading, not value — even though the broader macro environment (energy up, Mag 7 selling off, software weak) strongly favors value. This contradicts the usual expectation that small-cap indices outperform because they are de facto value indices. Armstrong’s tentative explanation is that years of overvalued big-cap US growth stocks are mean-reverting, but he explicitly admits he doesn’t understand all the catalysts driving this rotation and invites readers to send theories.
Quotable:
“Expensive stocks don’t fall because they are expensive. There is always a catalyst, and I don’t think I understand all the catalysts here.” — on the puzzling growth-within-small-caps outperformance
The Core · email · 8 mins
Trump posted “A civilization will die tonight” on Truth Social, threatening Iran with destruction of bridges, power plants, and desalination plants — effectively telegraphed war crimes. Pope Leo XIV, the UN Secretary General, and the French foreign minister all publicly condemned the threats. India, as 2026 BRICS chair and self-declared champion of the Global South, issued no statement — a conspicuous silence when a fellow developing nation was threatened with annihilation by the world’s most powerful leader.
The US-Iran ceasefire, mediated by Pakistan in Islamabad with JD Vance set to lead the US delegation, is already wobbling. Israel’s continued bombing of Beirut and occupation of southern Lebanon (south of the Litani river, 250+ killed) is the fault line: Iran insists Lebanon is covered by the truce; the US and Israel say it isn’t. Meanwhile, Iran is charging a $1-per-barrel toll on the ~20 million barrels that transit the Strait of Hormuz daily — effectively ~$20 million/day in extraction — and experts warn that even a held ceasefire locks in Iran as permanent gatekeeper of global energy flows. Gold hit $4,743/oz on the uncertainty.
India’s domestic energy situation is in crisis, directly tied to the Hormuz blockade. As the world’s second-largest LPG importer, India is rationing industrial gas to protect household cooking supply. Oil minister Hardeep Singh Puri flew to Qatar for emergency talks; GAIL bought three spot LNG cargoes and plans to borrow up to Rs 60 billion for expansion. Reliance capped fuel purchases at Rs 1,000 per visit across its 2,000+ Jio-BP outlets. India is also granting case-by-case port waivers for an ageing LPG tanker and a US-sanctioned crude vessel — bending sanctions compliance to keep the lights on.
India’s GLP-1 weight-loss drug market hit Rs 1,600 crore (moving annual total) and is accelerating sharply after Novo Nordisk lost its semaglutide patent in India. In just the last 10 days of March 2026, injectable semaglutide unit sales jumped nearly 6x (25,000 → 63,000 units); semaglutide’s share of the GLP-1 market climbed from 25% to 33% in a single month. Torrent Pharmaceuticals seized 8% market share (Rs 4.7 crore) within days of launch; 13 domestic firms are now selling 26 brands across oral and injectable formats. Novo Nordisk, which held 98% as recently as February, retained 76% by March — rapidly eroding but still dominant.
Quotable:
“It is strange leadership of the Global South for India to stay mute when the leader of the most powerful nation on earth threatens a fellow developing country with annihilation.” — The Core, on India’s silence over Trump’s Iran threats
The New York Times · email · 6 mins
The Artemis II crew — Christina Koch, Jeremy Hansen, Reid Wiseman, and Victor Glover — faced the most dangerous leg of their 10-day lunar mission on April 10: re-entry. The capsule hit Earth’s atmosphere at roughly 24,000 mph, encountering temperatures hot enough to melt portions of the metallic structure. The heat shield is the only protection, and the Artemis I mission already revealed it is flawed — with no backup system. A former NASA astronaut and a former NASA engineer told the Times the risk of shield failure was significant enough that the mission should have been postponed; NASA officials disagreed and proceeded.
VP JD Vance flew to Pakistan for U.S.-Iran nuclear peace talks scheduled for April 11, but Iran raised the stakes en route: a senior Iranian official demanded that Iran’s frozen assets be released before negotiations even begin — a new precondition on top of its existing demand that Israel halt strikes on Lebanon. Iran’s military also signaled it intends to maintain control of the Strait of Hormuz, which remained at a trickle of traffic even after a cease-fire was announced. Back in the U.S., the Israel-Iran war’s energy shock pushed inflation to 3.3% in March.
NYC Mayor Zohran Mamdani, 34, hit his 100-day mark leading the largest U.S. city. The democratic socialist has logged some wins — accelerated bus-speed projects, a rest stop for delivery workers — but has already backpedaled on his campaign pledge to cede control of public schools. New Yorkers are split on his performance.
Quotable:
“When doctors told us he may never speak again, I said, ‘Have you met my dad?’” — Matt Pinfield’s daughter, on the radio and MTV personality who suffered a stroke in January 2025, fell into a coma for over a month, and has since returned to broadcasting and attending concerts
Benn Stancil · email · 6 mins
Anthropic’s unreleased model “Mythos Preview” spontaneously found thousands of high-severity vulnerabilities in every major operating system and web browser — capabilities that were never explicitly trained but “emerged as a downstream consequence of general improvements in code, reasoning, and autonomy.” Anthropic withheld the model from public release due to safety concerns, launching “Project Glasswing” instead to deploy its capabilities defensively. Their own system card notes that “Anthropic’s capability trajectory bent upward in the period leading to Claude Mythos Preview,” with prior AI models not yet meaningfully accelerating their successors’ development — meaning the curve is bending without AI-on-AI bootstrapping.
The K-shaped AI economy is not new in kind, only in speed — jobs have always appeared and vanished, and K-shaped displacement has been predicted for years. What AI is doing is compressing the K’s angles: “get good at something, be the best, and make your money, before the walls close in.” The vertigo of reading Mythos’s system card evokes the precise childhood feeling of a 7th-grader who couldn’t get his middle school coach’s attention learning that Alex Rodriguez was already being scouted by MLB scouts — the gap not just in one field (baseball, engineering, law, medicine) but across all of them simultaneously.
Money is gravity that cannot be escaped even in a post-AGI world. OpenAI warned early investors that “it may be difficult to know what role money will play in a post-AGI world,” but scarcity is always relative — when intelligence and labor become abundant, a new bottleneck emerges and money follows whoever clears it. Society will always have a scoreboard. The more interesting question AI forces open is not “what would you do if you didn’t need money?” but what you’d pursue if you were suddenly free from the tyranny of being able to make money at all — a rare moment of existential weightlessness the current disruption actually offers.
Quotable:
“We are made anxious by those who have the new skills we’re supposed to have, like taste, judgement, and agency. We are jealous of those who are winning the games we’ve long played. But we are moved by those who have the courage to leave all of those old gravities behind.” — closing reflection on the AI transition