Featuring
Spice rack
Will AI create better jobs before it destroys the old ones?
Original point: The AI job-loss story is mostly theoretical: like earlier technological revolutions, AI will automate middle steps, raise productivity, and move people from rote work into more sophisticated jobs rather than create lasting mass unemployment.
What everyone argued
Jason Calacanis
Jason accepts that AI creates wealth but says GDP is not wages and many exposed workers own little equity. He points to weaker outcomes for young graduates and leaner headcounts at highly profitable tech companies, arguing that replacement jobs may arrive more slowly than automation removes entry-level, driving, warehouse, and knowledge-work roles.
David Sacks
Sacks argues that AI helped turn second-quarter growth from ordinary to exceptional and that today's systems remain 'middle to middle': humans still set objectives, prompt, validate, and iterate. He uses agriculture and the Model T to argue that visible displacement will be outweighed by new industries and higher-value work that cannot yet be named.
David Friedberg
Friedberg says innovation creates demand for workers before it removes old roles: firms must hire, train, and build the new factories and systems first, then workers voluntarily leave obsolete jobs for better-paid ones. He rejects victim framing and predicts recruitment will empty the old occupations before demand for them disappears.
Winner circle
Jason wins the risk assessment. Sacks and Friedberg make a credible long-run growth case, but they do not meet the burden for a smooth transition; both assume that new work arrives in the right place, at the right wage, for the right people. The honest position is neither doom nor denial: aggregate gains can coexist with painful, concentrated displacement, and current graduate data make that risk more than a media story.
Commentary
Jason Calacanis
Assumptions and fact checks
AI is a major cause of rising unemployment among young graduates and developers.
Why it mattersExposure is plausible, but the cited labor data are not causal. Remote-work hiring patterns, post-pandemic overstaffing, high interest rates, and sector mix also affect entry-level outcomes.
Current big-tech headcount patterns predict economy-wide job displacement.
Why it mattersFrontier technology firms are important leading indicators but are unusually scalable, capital-rich, and fresh from pandemic hiring swings. Their staffing ratios cannot be projected across health care, construction, hospitality, government, and small business without more evidence.
Labor-market conditions for recent college graduates were worsening.
CheckThe New York Fed reports recent-graduate unemployment at about 5.7% in 2026 Q1, with underemployment at 41.5%; its March comparison put unemployment at 5.6%, up from 3.6% in March 2019. That validates the stress Jason identified, though not his causal attribution to AI.
David Sacks
Sacks argues the century better than the next five years. The agricultural analogy supports long-run adaptation; it does not pay a 24-year-old developer's rent while the new occupational ladder is still under construction.
Assumptions and fact checks
Roughly 40% of second-quarter 2025 GDP growth was attributable to AI, making AI the difference between 2.3% and 3.8% growth.
Why it mattersAI-related equipment and software investment contributed to growth, but 'AI' is not a discrete BEA component and the counterfactual depends on classification and multiplier assumptions. The official release emphasizes imports and consumption, not a single AI attribution.
Because prior general-purpose technologies eventually created more work, AI replacement jobs will appear before affected workers suffer prolonged displacement.
Why it mattersLong-run job creation does not guarantee a smooth path for particular regions, ages, or occupations. Transition speed, retraining, bargaining power, and geographic match are precisely the unresolved variables.
Real U.S. GDP grew at a 3.8% annual rate in the second quarter of 2025.
CheckBEA's third estimate reported 3.8% annualized real GDP growth. BEA said the increase primarily reflected lower imports and higher consumer spending, so the separate claim that AI alone supplied 40% of growth is better treated as a model-dependent assumption, not an official GDP fact.
About half of the U.S. workforce worked in agriculture in 1900, falling to about 2% by 2000.
CheckUSDA reports 41% of the workforce in agriculture in 1900 and 1.9% in 2000. The century-long collapse is real and the endpoint is right, but 'one out of every two' materially overstates the starting share.
David Friedberg
Friedberg has an appealing conveyor belt: build the new industry, recruit everyone upward, retire the old one. Real labor markets have stairs, locked doors, and moving trucks; he needed evidence that the same people can board the belt in time.
Assumptions and fact checks
Workers in declining occupations will generally be recruited into higher-paying new jobs before their old jobs vanish.
Why it mattersSome booms work this way, but skill, location, credential, age, and wage mismatches can produce unemployment or lower pay even when total vacancies are high. The claim needs occupation-level evidence rather than a universal innovation-cycle story.
Describing automation as job loss wrongly casts workers as passive victims.
Why it mattersAgency matters, but an employer's automation decision can eliminate a role regardless of the worker's preference. Reframing involuntary displacement as voluntary upgrading avoids rather than resolves the distributional question.
The United States had the lowest unemployment rate in history in late 2025.
CheckBLS reported a 4.4% unemployment rate for September 2025. The postwar series includes substantially lower rates, including roughly 2.5% in 1953, so 'lowest in history' is plainly incorrect.
Would a National Guard deployment improve San Francisco safety, or disrupt a city already recovering?
Original point: Market Street still functioned as an open-air drug market, so a targeted federal operation could clear the remaining blight faster than local officials constrained by San Francisco's politics and legal system.
What everyone argued
Jason Calacanis
Jason proposes a negotiated, three-month deployment limited to BART stations and a few Tenderloin corners. He treats visible Guard presence as a deterrent and argues that Mayor Daniel Lurie could preserve local consent by defining the mission, location, and duration in advance.
David Sacks
Sacks accepts that Lurie and District Attorney Brooke Jenkins are improving the city, but says a stubborn Market Street drug market remains. He points to Washington, D.C., as a cooperation model and argues that federal authorities could target trafficking networks, deport noncitizen dealers, and accelerate a cleanup that local institutions have not completed.
David Friedberg
Friedberg argues that the city's own data show a sharp recovery: crime, homicides, tents, and car break-ins were falling while police ranks and convention demand improved. Because local officials were already arresting offenders and winning convictions, he says the administration should take a beat rather than trigger backlash with federal military policing.
Winner circle
Friedberg wins the deployment question. He brought evidence that the city was improving and forced proponents to explain the Guard's incremental role; they never did. Sacks deserves credit for the narrower point that federal partnership can help, but the later successful model used civilian law-enforcement agencies rather than soldiers on street corners.
Commentary
Jason Calacanis
Jason supplies the most operational version of the pro-deployment case, but the mission remains a costume without a job description. Naming the legal authority, arrest role, command structure, and measure of success would have turned a political gesture into a policy.
Assumptions and fact checks
A tightly limited Guard deployment would win broad public acceptance despite predictable protests.
Why it mattersDuration and local consent would reduce opposition, but military presence carries symbolic and escalation costs that cannot be inferred from support for ordinary law enforcement. Jason offers no polling or comparable San Francisco test.
Guard presence at transit stations and Tenderloin corners would add meaningful deterrence beyond reinforced civilian policing.
Why it mattersVisible presence may deter some conduct, but displacement to nearby blocks is possible and the relevant comparison is against trained civilian officers with arrest, investigation, and service-referral authority.
David Sacks
Sacks is strongest when he asks for targeted federal help and weakest when he treats the Guard, DEA, ICE, FBI, prosecutors, and local police as interchangeable uniforms. The later joint-enforcement results vindicate coordination, not the claim that soldiers were the missing tool.
Assumptions and fact checks
National Guard personnel could directly police drug markets and make the proposed cleanup easy.
Why it mattersFederalized Guard members face Posse Comitatus restrictions on direct civilian law enforcement unless a constitutional or statutory exception applies. State-controlled Guard support is different, but search, seizure, and arrest remain legally and operationally more natural for civilian agencies.
The remaining drug market primarily reflects ideological constraints rather than capacity, court rules, housing, addiction, or trafficking economics.
Why it mattersPolicy choices matter, but Sacks reduces a multi-agency problem to mindset. The later decline involved police staffing, technology, prosecution, federal partners, and treatment-related work rather than one ideological switch.
San Francisco provides about $2 billion a year to nonprofits serving its homeless and addicted population.
CheckThe city's adopted appropriations list roughly $846 million for the Department of Homelessness and Supportive Housing in FY2024-25 and about $677 million in FY2025-26, including far more than nonprofit grants. A broader multi-department total can be larger, but the specific $2 billion annual flow to nonprofits is not supported.
David Friedberg
Friedberg keeps the burden of proof where it belongs: extraordinary deployment needs evidence of an extraordinary gap. His case would be even tighter with neighborhood-level Tenderloin data, but the year-end city results aged very well.
Assumptions and fact checks
Continued local improvement made a Guard deployment unnecessary even in the worst remaining blocks.
Why it mattersThe burden belongs on proponents of military involvement when civilian institutions are improving and federal civilian agencies can support targeted operations. Concentrated disorder justified more work, but not necessarily a new military actor.
San Francisco was on track for a roughly 30% crime decline, a 70-year homicide low, and a multidecade low in car break-ins in 2025.
CheckSFPD's final 2025 report recorded a 25% overall crime decline, 28 homicides—the lowest count since 1954—and a 43% drop in car break-ins. The exact interim percentages differed, but the direction and historic scale of Friedberg's claim held up.
Do rare-earth price floors fix China's market power, or should America rely on deregulation?
Original point: Price floors inflate input costs and invite long-term distortion; the United States should instead lower regulatory barriers and use broad incentives so private investment can discover, process, and compete.
What everyone argued
Chamath Palihapitiya
Chamath says ordinary take-or-pay contracts cannot protect U.S. projects when Chinese state support can flood the spot market and crater prices. He favors a federal buyer of last resort and a strategic reserve that purchases against long-run demand, then supplies manufacturers during shocks.
David Sacks
Sacks argues that rare earths are already a non-market national-security contest: China can subsidize production, dominate processing and magnets, then cut prices to strand Western investment. A floor gives investors enough certainty to finance U.S. capacity and prevents dependence from becoming geopolitical leverage.
David Friedberg
Friedberg says permitting costs and dirty, outdated separation methods pushed mining and processing abroad. He argues that modern exploration, safer processing, deregulation, and tax incentives could unlock domestic resources and let U.S. firms out-innovate China without embedding an artificial floor in input prices.
Winner circle
Sacks wins the narrow mechanism debate. Friedberg is right that America still needs faster permits and better processing, but those reforms do not protect a newly financed plant from strategic dumping. A bounded, benchmarked floor with shared upside is a defensible bridge to capacity; Chamath's reserve is useful insurance alongside it, not a substitute for production.
Commentary
Chamath Palihapitiya
Chamath sees the bankability problem that a simple 'let markets work' answer misses. He would be stronger if he treated a reserve as insurance inside a portfolio of policies instead of the lone antidote.
Assumptions and fact checks
A government buyer of last resort is the only practical defense against Chinese price manipulation.
Why it mattersA reserve can absorb shocks, but allied supply, recycling, substitution, permitting reform, tax credits, long-term offtake, and targeted price guarantees are complementary tools. Calling one instrument the only answer overstates the case.
Government can forecast long-run mineral demand well enough to build reserves without costly overbuying or backing obsolete inputs.
Why it mattersStrategic inventories can insure against concentrated supply, but technology mix, recycling, and substitution make exact demand uncertain. Transparent targets and release rules matter as much as the decision to stockpile.
The United States later created a strategic stockpile mechanism for commercial critical-mineral users.
CheckUSGS described Project Vault in February 2026 as a critical-minerals stockpile focused on rare earths, lithium, nickel, and related materials for American industry. This is hindsight confirmation of the mechanism Chamath advocated, not proof that every purchase will be efficient.
David Sacks
Sacks wins because he answers the actual financing problem rather than the label 'price control.' The unanswered question is governance: the floor needs competitive milestones and an exit ramp so resilience does not become a permanent subsidy hammock.
Assumptions and fact checks
A price floor can preserve strategic capacity without causing unacceptable cost, favoritism, or complacency.
Why it mattersThe mechanism protects investment from predatory pricing, but its quality depends on the benchmark, duration, competition, cost-sharing, and performance conditions. A defensible tool can still be badly designed.
China dominates the processing and permanent-magnet stages of the rare-earth supply chain.
CheckIEA estimates China held 91% of global refined magnet-rare-earth output and 94% of sintered permanent-magnet production in 2024. That concentration is large enough to create the leverage Sacks describes.
The U.S. government used a long-term price floor to support MP Materials' domestic NdPr production.
CheckMP Materials disclosed a ten-year $110-per-kilogram NdPr price-floor commitment plus offtake, equity, and loan support from the Defense Department. The structure directly addresses the financing risk Sacks identifies.
David Friedberg
Friedberg correctly insists that resilience requires physical capability, not clever contracting alone. He loses because deregulation answers why projects are expensive, not how investors survive a strategic price crash after construction.
Assumptions and fact checks
Permitting reform and innovation can make U.S. rare-earth projects competitive without price protection.
Why it mattersLower cost and faster permits are essential, but they do not neutralize below-cost foreign supply or guarantee downstream demand. The thesis needs a credible answer to strategic dumping during the years before scale.
New exploration could reveal more than a thousand times today's proven U.S. rare-earth reserves.
Why it mattersEarth MRI is designed to identify undiscovered resources, but the thousand-fold figure is not supported by the cited public USGS program and confuses geological potential with economically recoverable reserves.
The United States is completely dependent on Chinese suppliers for rare earths.
CheckUSGS reports that imports supplied 80% of U.S. rare-earth consumption in 2024 and China directly supplied an estimated 56% of consumption. Dependence is severe, especially downstream, but neither figure is 100% and domestic production exists.
Jason wins by asking who receives the productivity dividend and when. His caution is stronger than his causation: the entry-level squeeze is real, but proving AI did it requires job-level adoption and separation data, not a tour of four corporate org charts.