Decline Is a Choice: Marc Andreessen on Power, Censorship, and a Second Roaring Twenties

Marc Andreessen opens where the numbers already look insane. The United States is adding a trillion dollars to the national debt every 100 days, a figure now passing the Defense Department's annual budget and compounding as it goes. Soon, he says, it will be a trillion every 90 days, then 80, then 70. If that is not fixed, the endpoint is a hyperinflationary spiral and the country becomes Argentina or Brazil.

That is the first thing he says on Lex Fridman's podcast, episode 458, his second visit. The conversation runs nearly four hours, from a "roaring twenties" forecast through Trump, the censorship machine, universities, H-1B visas, AI coding, and whether God is the work of a third thirty-year stretch of life. Andreessen is the co-creator of Mosaic, co-founder of Netscape and Andreessen Horowitz, a Meta board member, an early Substack backer, and a volunteer — he calls himself an unpaid intern — around the Trump transition.

Asked for the best possible 2025 and 2026, he does not hesitate: the roaring twenties. Through COVID, politics, and everything else, the United States just kept growing while Canada, the United Kingdom, and Germany stopped, and some of those countries may now be going backwards. Geography is luck — a continent's worth of physical security, natural resources so abundant that the running joke is a North Dakota farmer kicking over a hay bale and finding a two-trillion-dollar deposit. Energy independence is not a capability problem. The last administration chose to turn American energy off; the new one has declared it will turn it on.

The other ingredients are people and technology. For fifty, a hundred, two hundred years the most driven people in the world moved here and raised children, producing what he calls the most aggressive population in any Western country. In software, AI, and advanced biotech the United States is overwhelmingly the leader, in part because the best scientists and engineers still come. Manufacturing is the notable gap. Put together, he says, the country has all the preconditions for a monster boom. All of its problems are self-inflicted. "Decline is a choice."

Fridman wants the human version of the same story: why the entrepreneurial intensity feels unique. Andreessen reaches for *Succession*. In the season-one finale Logan Roy, miserable back in ancestral Scotland, says he cannot wait to return to America, "where we can fuck without condoms." The line is a metaphor and the room laughs because everyone knows what it means. Four hundred years of Yankees, Scots-Irish, Southerners, Texans, California hippies, rocket scientists, and artists produced a national spirit that keeps bouncing back. The recurring American error, in his view, is trying to sand all of that off into something boring and mediocre.

He lived through one reset as a child. The 1970s felt like a nightmare: energy crisis, the Iran hostage crisis for some 440 days, inflation, demoralization, Jimmy Carter's Malaise Speech as the low point. Reagan arrived controversial and said it was morning in America and a shining city on a hill, and the national spirit came roaring back for a decade. The last ten years have felt to him like another seventies. Individualism in America probably peaked between the end of the Civil War and around 1931, the Second Industrial Revolution, when the United States took technological leadership from England and later won World War II and the Cold War. Line-go-up is not enough. A country is more than an economic zone.

The tangent is a nineteenth-century book: *The Ancient City* by Numa Denis Fustel de Coulanges, a Sorbonne classicist writing in the 1860s who had apparently absorbed every scrap of Greek and Roman literature. He reconstructed the Indo-Europeans who came down from the steppes — a warrior people, ancestors of Greeks and Romans — living under maximum survival pressure with no modern technology. Their social pattern was family, tribe, and city, with zero concept of individual rights. Morality was Nietzsche's master morality: strong is good because strong is survival. The head of the family, tribe, or city had life-and-death power. The form of the society, Andreessen says, was maximum fascism combined with maximum communism, and it only scaled to the size of a family.

Their religion was ancestor worship and nature worship. Fast-forward through Christianity to Nietzsche declaring God dead, and the most evolved modern secular experts have, in his telling, fallen back on the same structure: ancestor worship as identity politics, nature worship as environmentalism. Identity politics tags newborns with benefits or condemnation based on who their ancestors were. The Indo-Europeans would have recognized it on sight. He will not stop at the critique of tradition. Completely reinventing society every generation is Pol Pot's Year Zero and the Soviets, a bad history. Tradition ought to be evolutionary learning. The trap is gerontocracy on one side and, in the United States, kids setting the agenda on the other. Extremes on either side are a bad idea.

The Trump administration's part in the roaring twenties, Andreessen says, is getting the government boot off the neck of the American economy, the technology industry, and the American people. He calls the last decade a form of soft authoritarianism. It is not Lubyanka and not a 4 a.m. knock, but it is a suppressive blanket: a vetocracy in which forty people must sign off and any one can veto, plus the idea that progress, technology, capitalism, and success are bad. Tall poppies get chopped. Specific government actions sat on top of the mood — censorship regimes, debanking, deliberately kneecapping critical industries, social policies that amounted to letting criminals out of jail to see what happens. He calls the whole stretch a demoralization campaign. The Biden years reached the pinnacle of it.

He will not predict Washington. He has learned he cannot. The plans, staffing, and conversations around the incoming administration are consistent with rapid action by executive order, then legislation and regulation. He already feels a vibe shift in Silicon Valley, whether people were pro-Trump or anti-Trump. Hollywood people who were still vocally anti-Trump told him two weeks after the election that the ice had thawed, woke was over, and comedies might get made again. Business operators around the economy tell him the last ten years of misery are just over.

BlackRock's Larry Fink is his comic exhibit. Fink, he says, was first in as a major investment CEO on every dumb social trend, aggregating customers' voting control to force companies into ideological positions — "the Typhoid Mary of all this stuff in corporate America." In the last year Fink has been backpedaling as fast as he can, including pulling out of a corporate net-zero alliance. Andreessen compares it to Richard Pryor's backwards walk: leaving the room while looking like he is walking forward.

The NASDAQ board-diversity mandate produced the scene he cannot forget. Sitting on Meta's board with Peter Thiel when the rules landed, they had to figure out which of them counted as diverse. The attorneys explained, with a completely straight face, that Thiel counted by virtue of being LGBT. Thiel wrote a book called *The Diversity Myth*. He looked, Andreessen says, as if he had swallowed a live goldfish. The Court of Appeals later nuked the rule. The joke is the point: when a rule set detaches from reality far enough, you still have to sit there and perform it.

Preference falsification is Fridman's next question: how many people in tech actually hate Trump, and how many are just riding the vibe? Silicon Valley voting and donation records have been overwhelmingly Democratic for Andreessen's entire time there, donations running 90 to 99 percent one way. He is more interested in the top thousand or ten thousand — founders, CEOs, executives, engineers, VCs. The idea that people have beliefs, he has come to think, is mostly wrong. Most people just go along. The most high-status people may be the most prone to it, because they are the most focused on status.

His forbidden philosopher is Ted Kaczynski. The Unabomber was a lunatic murderer and a Harvard super-genius, and amid the madness he named over-socialization: modern Western elites overly oriented toward what people like themselves think. Before Andreessen even held his current views, he noticed that every dinner party had the same conversation, that everyone agreed on every issue, and that the agreement was precisely that day's *New York Times*. The positions were not the same as five years earlier, but everyone snapped into agreement every step of the way. Harvard and Yale believe the same thing; the *Times* and the *Post* believe the same thing; the Ford and Rockefeller foundations believe the same thing; Google and Microsoft believe the same thing. The content changes by decade. The lockstep does not.

Heretical ideas do not live at those dinners. They live one-on-one, face-to-face, in a whisper network. The secret handshake is a joke. If the other person laughs, the conversation is on; if not, you back slowly away. Laughter is involuntary, which is why fake laughs are obvious and why comedians can say what others cannot. Live comedy creates safety in numbers. Then come the alt dinner parties of six or eight thought criminals, the happiest rooms of the last decade because nobody has to fear getting yelled at for every third sentence. Online, until Elon Musk bought X and Substack took off, the same pattern played out as samizdat: conformity on the public networks, encrypted group chats underneath. WhatsApp and Signal plus disappearing messages unleashed it. Some people set the timer to seven days; friends insist on one day; every so often someone drops it to five minutes before sending something inflammatory.

Timur Kuran's *Private Truths, Public Lies* is the book Andreessen wants on the table. Preference falsification is two lies: you believe something and cannot say it, or you do not believe something and must say it. Václav Havel's greengrocer in *The Power of the Powerless* is the other half. In Prague in 1985 the shopkeeper still had to put "Workers of the world, unite" in the window every morning. Everyone knew it was fake — the grocer, the pedestrians, the police. It was no longer ideology. It was compliance. The social consequence is that nobody has an accurate gauge of how many people agree with them. If 10 percent agree, you get killed; if 90 percent agree, you win the revolution. Authoritarian systems cannot run an honest survey. You only find out by putting it to the test.

Someone sticks a hand up. In this cycle, Musk. Everyone watches to see if he gets killed. If he does not, a second hand goes up, then two, four, eight, sixteen, and then, as with the Berlin Wall in 1989, it rips. Suddenly people realize they are the majority. Elites do idea formation; the masses decide whom to follow. Eric Hoffer's *The True Believer* is the companion text. Musk and Trump are, in Andreessen's experience, the only two people he knows who say in private exactly what they say in public. That consistency is why they can catalyze shifts. Trump took over the Republican Party by saying out loud what the base already believed, standing among conventional Republicans who would not. The siren song of censorship is that if you keep people from saying things, you can keep them from thinking things. When it unwinds, it unwinds with speed, because the energy was bottled.

His percentage guess for his world is 20-60-20. Twenty percent are true believers in the current thing. Twenty percent are not on board but start as a small minority. The middle 60 percent have normal lives — work, kids, the game — and by default go along with the first 20. In the Valley that 60 percent spent a decade being woke and is pivoting in real time. They are just done. Kuran's warning is the swing the other way: people who will now claim they supported Trump all along, or never backed ESG, or never wanted that person canceled. The Democratic Party, he says, is heading into a civil war between de-radicalizing toward the center and doubling down. Meanwhile Harvard is still Harvard, and the *Times* is still the *Times*.

Fridman asks the hard version about Musk: he is becoming the most powerful human on Earth; how does the Ring of Power not corrupt him? Andreessen is not sure about the premise — Musk does not have the nukes. A lot of the controversy is Musk saying true things, as with the organized child-rape scandal in the United Kingdom that had been suppressed until he brought it up and the country exploded into open conversation. How worried should we be about someone whose power is that they get to speak the truth? After a decade of everyone lying about everything, he wants to run that experiment hard. On money in politics: Hillary outspent Trump three to one in 2016 and lost; in 2024 Kamala Harris raised and spent $1.5 billion on top of what Biden had spent, and Trump spent something like a third or a fourth of that. The topic almost vanished for eight years. Voters, the data suggests, actually have a voice.

Fridman tells a story that bothers him. A historian he admires, a heretical thinker with tenure, declined to come on the podcast because of the guest list — Trump, and Fridman's stated interest in talking to Vladimir Putin — and did not want the headache in faculty meetings. More than 80 percent of the guest list is left-wing. Andreessen wants to know if the man has tenure. He does. Then he should use it. The tenure system sang the same song for seventy years: work yourself to a professorship so you can say what you actually think without being fired. Academic freedom is gone. The entire moral foundation, paid for virtually 100 percent by taxpayers, is fake and rotten.

Universities run on four streams of federal money: the student-loan program in the many trillions and spiraling faster than inflation; research funding, of which the university rakes as much as 70 percent for central uses; tax exemption at the operating level; and tax exemption at the endowment. Pull those, plus state money for state schools, and they all instantly go bankrupt. Starting a replacement is blocked at the accreditation layer. The Department of Education does not accredit universities directly; it accredits nonprofit bureaus whose members are the existing universities. Incumbents control who gets federal student-loan money. Guess how enthusiastic they are about accrediting a new university. He calls it a government-funded cartel.

Fridman believes leadership can fix Stanford and MIT from within — cut funding as leverage, fire administrators, put students and faculty back at the center. Andreessen calls that underpants-gnome logic, from *South Park*: step one, collect underpants; step three, profit; step two, question mark. In politics the form is: we must do something, this is something, therefore we must do this, with no causal chain. Who runs the university if the president cannot fire the professors? The professors and what he calls the feral students, in a radicalization feedback loop. Tenured faculty know they can outlast any president. It is the same thing government bureaucrats know. It is the resistance: they sit there and resist.

Charlie Munger, asked how he would fix the culture at General Electric, said he could not even fix the culture at a restaurant. In business, nobody thinks you repair a terrible restaurant; you open a new one. Andreessen hopes he is wrong. Post-October 7, donors and board members such as Marc Rowan are genuinely trying. A friend on the executive committee of a top technical university is working overtime. He would love to see tenured professors get brave. The counter-question is whether it is actually happening at a single one of these places. Taxpayers do not yet understand the crisis. If they do, the funding evaporates, and the fuse is already lit.

Andreessen got the first ask to implement internet censorship in 1992, not 2013. He was making $6.25 an hour on a federally funded research project at a public university, and his boss wanted the browser to keep naked photos from showing up. This was pre-ImageNet. He briefly imagined a "breast detection algorithm," then declined. He did not know how to do it, and he did not want to build a censorship engine. The practical truth, he says now, is that no internet service can run at zero censorship, and no country has unlimited free speech. The First Amendment has twelve or thirteen Supreme Court carve-outs: incitement, terrorist recruitment, child abuse, child pornography. You are always going to have a machine. Once it exists, it is the Ring of Power.

The pivot came around 2012 and 2013, the start of Obama's second term, coinciding with the arrival of highly radicalized students — in school between the Iraq War and the financial crisis — who immediately mounted ban campaigns. The Democratic Party noticed. Executive teams and boards were almost all Democrats; Obama-era staff and FBI and intelligence people circulated in. The ring was lying on the table. They picked it up and ran. The original topics were hate speech and misinformation. Hate speech started reasonably: no N-word. Then, as Jordan Peterson has described, it became anything that makes people uncomfortable. Andreessen raising his hand to say that *that* idea made him uncomfortable did not count. Misinformation came later, as a consequence of the Russiagate hoax, and was more dangerous. Offensiveness can at least be argued. Truth, Western civilization spent eight hundred years learning, is not something a committee of experts gets to declare. God and the king used to do that. We do not have those anymore.

The straw that broke the camel's back was declaring lab-leak theories of COVID to be misinformation. Inside the companies, people realized the thing had spun completely out of control. The complex ran from about 2013 to 2023 and broke, in his account, three ways: Substack launched as a free-speech platform and stood up to press and platform pressure; Musk's purchase of X was the hammer blow; Zuckerberg is now the third. Jon Stewart's appearance on Stephen Colbert's show was the catalyzing public moment. In a climate where you had to believe COVID was entirely natural or you were a fascist, Stewart said it was not plausible to rule out a virus appearing three hundred yards from the Wuhan Institute of Virology. Nicholson Baker's *New York Magazine* piece was the first mainstream undressing. Andreessen remembers a board meeting after which everyone looked around and agreed they did not need to censor that anymore. The next question was why they had. The question after that was what the rest of the regime was worth. That was the thread in the sweater.

Zuckerberg on Joe Rogan, Fridman says, was a courageous act. Andreessen now puts him in a third category with Musk and Trump: public and private synchronized. Zuckerberg runs services used by more than three billion people and has never had Elon's flexibility. CEOs are lonely. You cannot talk to executives, who have agendas; you cannot talk to a board that can fire you, unless, like Zuckerberg, you have control. Companies can withstand most external pressure if they keep internal coherence. The moment the internal team cracks, the solid turns to liquid. Government pressure is the exception. Any CEO who thinks he is bigger than his government has that notion beaten out of him in short order. Twitter CEOs, he says, all had the same experience: daily beatings.

Fridman asks whether the government crossed into criminal pressure. Andreessen's answer is flat: flagrant criminality, felonies, obvious felonies, and he can cite the laws. Three public records, he says, are enough. The Twitter Files are 100 percent representative of what he saw at every other company. Representative Jim Jordan's Weaponization Committee produced a Facebook version of the same story, some 2,000 pages, from documents the company turned over. Zuckerberg described it on Rogan. In 2020 and 2021 the government decided the First Amendment did not apply to them, called companies, threatened, bullied, and forced censorship. There was no due process for the speech regime or for debanking.

The statutes are 18 U.S.C. §241 and §242, one for government employees and one for private actors, covering deprivation of rights and conspiracy to deprive rights. It is not legal for government employees, or private entities in conspiracy with them, to take away constitutional rights. Some of those rights are enumerated; some are not. The Constitution does not say you only have the rights it lists. Debanking, the right of access to the financial system, sits under the same laws as speech. Companies will argue they were victims, not perpetrators, and he thinks they have a case. The people in the middle — federally funded third-party censorship shops, Stanford Internet Observatory the obvious example — are, in his view, acting as agents of the government and badly exposed.

How a bill becomes a law is a cartoon. Most of what companies live under is not legislation. It is regulation, tens of thousands of rules that can wreck a company, then raw administrative power: the government calling to tell you what to do. Facebook built Libra, a global payments cryptocurrency meant for three billion people, with anti-money-laundering controls ready. Two senators wrote to the big banks: we are not telling you that you cannot work with Facebook, but if you do, every aspect of your business will come under greatly increased regulatory scrutiny. That, he says, is the equivalent of admiring a corner restaurant and noting it would be a shame if someone tossed a Molotov cocktail through the window. Then it culminates in White House calls telling you what to do — a king's power, not a president's. In the actual ranking, legislation was the least important instrument. Threats were the most.

The fix, as he lists it: elect people who do not do this; transparency, including the broad truth-and-reconciliation process Peter Thiel argued for in the *Financial Times*; investigations; ultimately prosecutions, because object lessons require people going to jail. The last two are up to the new administration, and he will not speak for them. The Ring of Power in Tolkien is infinitely tempting and inherently corrupting. Gandalf will not put it on even to end the war. People who started out thinking they were patriots defending democracy ended up authoritarian, anti-democratic, and anti-Western. He does not know whether any of them feel that today. They should.

Robert Michels's iron law of oligarchy reaches Andreessen through James Burnham's *The Machiavellians*, which he calls probably the best book on politics written in the twentieth century. Classic theory offers three forms: democracy, oligarchy, monarchy — rule of the many, the few, the one. Michels asked whether democracy had ever existed, mounted a historical search, and found almost never. He then walked into what he took to be the most democratic private organization on earth, a German autoworkers' union devoted to the workers of the world uniting, and found six men at the top leading the membership by the nose. Every union has a Jimmy Hoffa; the number two is his son. The masses cannot organize, whether 25,000 in a union or 250 million in a country. Only a minority can. Every political structure in human history has been a small organized elite ruling a large dispersed majority.

Florence tried direct democracy for about two years in the Renaissance; it was a disaster, never repeated. California's proposition system, an overlay on the legislature, is in his view a catastrophe. The good news is the American founders understood this. They did not give the country a direct democracy. They built the oligarchy into a representative system — Congress, executive, judiciary — and then a second layer: a House apportioned by population and elected every two years, a Senate equal by state and elected every six, both required for legislation. Many of them were in their twenties, Einsteins of politics, shape rotators rather than word cells, students of Greek, Roman, and Renaissance history whose seminar notes survive as *The Federalist Papers*. The system has lasted coming on 250 years. He would not trade it.

The reductionist test inside any institution is who can get whom fired. Who has more power, a newspaper columnist on $200,000 a year or a CEO on $200 million? He has seen the columnist get the CEO fired. He has not seen the reverse. A CEO runs a company until shareholders, the management team, employees, the *New York Times*, or the Senate come for the seat. The live case is the *Los Angeles Times*. A biotech entrepreneur bought it about eight years ago for $500 million, has probably burned another $500 million keeping it running, and spends something like $50 to $100 million a year. The paper endorsed a litany of left-wing candidates. Just before the November election the owner said they would not endorse for the presidency. The newsroom flipped out. The journalists hate him with the fury of a thousand suns. For eight years the reporters ran the paper. Now the owner is showing up, setting policy, removing people, and claiming the prerogatives of ownership. In a fight between employees and employer, it is not crystal clear the employer wins.

Fridman puts Bill Ackman in the courage category for going after journalists in public. Andreessen notes that a lot of people had not heard of Ackman before October 7 and the Harvard fights, but he was a famous activist investor for some thirty years. The skill set, in the *Taken* sense, is taking control of situations without formal majority: a 5 percent stake, other investors rallied, entrenched management forced to change. Andreessen has been on the receiving end inside companies. Ackman has now brought that playbook to universities and newspapers, with deep pockets and courtroom experience, and is teaching the rest of the room in real time. His own investors, Andreessen says he knows for a fact, have been remarkably supportive, because a lot of them agree with him. He is the tip of the spear for something more people already believed.

Andreessen is careful: he is not in the administration, not in the room when people are selected, not speaking for anyone. He is an unpaid intern, a volunteer. The unusual structure is an incumbent who left for four years and is coming back. The first term did not arrive with a team. The institutional Republican Party was full of Bush people, many of them Never Trumpers; it took two to three years to staff the government, they had it for maybe eighteen months, and then COVID hit. Officials from that term came under legal assault, some went to prison, and many were debanked, including the president's wife and son. The live question was whether a second term could even match the first. What he has seen in the appointments looks much better: a deeper talent pool, people who know the game and still want in, a generational shift with cabinet and second- and third-level people in their thirties and forties instead of a gerontocracy.

Fridman worries about sycophancy around Trump. Andreessen says reality warps around any powerful leader, including CEOs, and that Trump at this point in his life seems aware of it. The caliber of the people taking the jobs is his evidence. He was on an HHS interview team, did not know the candidate list in advance, saw Jay Bhattacharya's name, and almost fell out of his chair. Bhattacharya proved himself under extraordinary pressure and would, in Andreessen's view, have been a gigantic upgrade as head of NIH. Marty Makary at FDA is the same story; the CDC pick is the same story. He has been spending time with Dr. Oz, a walking encyclopedia on the healthcare system. Steve Feinberg as number two at the Pentagon. Two of Andreessen's own partners are going in. Pound for pound, he says, it may be the best White House team since the 1990s, the 1950s, maybe Eisenhower.

DOGE, again with a disclaimer that he is not on it, is a time-limited eighteen-month commission advising the executive branch, not a formal agency. Three streams: money, people, regulations. The headline is $2 trillion. How many federal agencies are there? Somewhere between 450 and 520; nobody is quite sure. How many people work for the federal government? On the order of four million full-time employees and maybe up to twenty million contractors; nobody is quite sure. Most federal buildings are still empty. Two years ago the Supreme Court went after the idea that the executive can create agencies and issue regulations without corresponding legislation. Most of the federal government that exists today is not enforcing laws Congress passed. The last White House did nothing with that. The new one could say unwinding illegal regulation, spend, and headcount is compliance with the Court, not a challenge to the laws.

Washington is cynical because it knows every failed attempt, from Reagan to Clinton-Gore. What is new is social media. These people can bypass the press, expose spending in daylight, and run victory laps. Andreessen frames it as a Machiavellian sandwich: a president as rule of one, combined with the people as rule of many, against Congress and the permanent bureaucracy as the oligarchy in the middle. Experts call a $30 billion broadband plan that delivered broadband to nobody, or $40 billion of charging stations that built eight, a rounding error. Divided by 300 million taxpayers, $30 billion is $100 a year — a meal out, a babysitter, a trip to the park. The contempt for the taxpayer, he says, is off the charts. Impoundment is the other tell: by late-century interpretation the executive is not allowed to underspend, so every September 30 there is a budget flush of hundreds of billions spent unproductively on purpose. That is not what the founders meant.

On H-1B, Andreessen sat out the Christmas pile-on and read everything. He is not working the issue in government. He has made the standard Silicon Valley argument many times in Washington: brain-drain the world, the newcomers create companies and industries and more jobs for native-born Americans than previously existed, a positive-sum flywheel with no trade-offs, and anyone who objects is a racist. He believes a lot of it. His firm and its companies feel like the United Nations; he has worked with people from some eighty countries. The best and brightest already come here to get educated. Staple a green card to the technical degrees. He also grew up in a small farming town in rural Wisconsin. Almost nobody makes that trip into high tech. There is no talent flow from the Midwest, the South, the Sunbelt, or the Deep South. Those are the people who put a man on the moon, built the World War II machine, the railroads, the telephone network, Detroit and Cleveland. Philo Farnsworth, who invented television, grew up in the middle of nowhere. The blank spot on the map is also Trump country.

Either you believe the strong Vivek Ramaswamy thesis that the middle of the country is too culturally rotten to do what their grandparents did, or something else is happening. What educational institution in the last sixty years wanted farm boys from Wisconsin? All of them, he says, rejected them. The Harvard and UNC Supreme Court records showed admissions as cartoonishly racial as a ring of power would predict, including text messages of crude stereotypes and SAT cutoffs that can differ by 400 points across ethnic groups. Asians need to perform that much higher to get in; the cliché in the Valley and in medicine is that an Asian from Harvard is guaranteed to be Einstein, because otherwise they were never admitted. He does not believe a single institution is complying with the ruling. After October 7 it also became possible to talk about Jewish applicants being managed down toward 2 percent of the population after a century in which they had gone from exclusion to as much as 30 percent of elite campuses.

The piece that blew his mind was a 2004 *New York Times* story, quoting Henry Louis Gates and Lani Guinier, asking which Black students were being admitted. At the time, half of the Black admits at a place like Harvard were not American-born — Nigerian, other West African, or West Indian. Kamala Harris and Colin Powell are his West Indian examples. If affirmative action is meant to remedy American slavery, importing someone from Nigeria who never experienced it is, he says, a tremendous insult to Black Americans. For sixty years the country has run a social-engineering experiment excluding native-born people — white, Asian, Jewish, Black — from the same scarce educational slots and jobs that high-skill immigration fills. Google only hires so many L7 engineers. Over the entire Christmas debate he did not see a single person pounding the table for more high-skill immigration who was also pounding the table to get smart kids already here into those seats. Not one.

The rest of the nuance makes the slogan even less clean. The Valley mostly uses O-1s, the super-genius founder visa with a real evidentiary bar. H-1Bs flow to a handful of big tech companies that hire in volume and to consulting mills such as Cognizant, whose model is large numbers of Indian workers living near company housing, doing mid-tier IT consulting at $60,000 to $100,000 a year rather than $300,000 in the Valley. Startups mostly cannot use the program. High-skill immigration also has only four permanent sinks: the United States, Canada, the United Kingdom, and Australia. Those four countries drain talent from everyone else. After the Greek crisis the smart young people were gone. Ukraine and Russia have been brain-drained too. The West treats physical-resource extraction as the height of colonial evil and human-capital extraction as a flowering of justice. National Merit Scholars already identify the top 0.5 percent by PSAT and SAT, race-blind, for a $2,500 scholarship. How many universities use it as a scouting system? Zero. They scout basketball. The SAT itself has been dumbed down, g-loading pulled out, made more coachable — a subsidy to families who can pay, which his hometown did not have. An 800 no longer separates the top hundred kids you would send to MIT. The optimistic forecast is to do both: high-skill immigration and a fair shake for families in the middle of the country. That would defuse the issue.

Asked for advice to people shipping software in 2025, Andreessen says we are in a legit revolution. AI coding is the biggest earthquake to hit software in his life, maybe since software was invented, because code can be validated, which means synthetic data, self-training, and reinforcement learning all apply. It is already great; YouTube is full of kids making apps in ten minutes. For a long time, people who can code will still be the best at making AIs code, because they can evaluate the work. His nine-year-old is in coding classes through 18. Non-coders can still say "teach me how to code." He names Khan Academy's free work and Replit, originally built for kids, now with AI built in. AIs are spectacular at explaining code, writing glue into Stripe, generating game art and voice, even drafting a business plan. William Gibson's line applies harder than ever: the future is here, it is just not evenly distributed. In the late 1980s a famous book, *The Decline and Fall of the American Programmer*, predicted the end of the job. Programming jobs exploded by a factor of a hundred. His guess is an order of magnitude more coding jobs in ten years, orchestrating AIs, because demand for code is effectively unlimited.

Little tech is not shrinking. He is seeing the AI hiring boom of all time at big companies and startups alike. Weak-form companies add AI as the sixth bullet on a slide that already existed — Gmail offering to summarize email when he wants it answered. Strong-form companies rebuild from scratch with AI as the first citizen; Cursor is the example, and he just met a team building an AI-native email system. When the transformation is powerful enough you reconceptualize the product, which usually means a new company, because incumbents will not. Clay Christensen's *The Innovator's Dilemma* is the reason: well-run large companies often fail to adapt *because* they are well-run, optimizing for existing customers. Startups exist to break glass. He is waiting for the first company whose org chart is redesigned around AI — a VP of engineering supervising a hundred coding agents, or an AI as the VP, or an autonomous agent that builds itself a business. The systems are not quite ready, and it will take a founder willing to break all the rules.

He will not pick a 2025 winner among OpenAI, Meta, Google, and xAI. There are five or six trillion-dollar questions hanging in the air: big versus small models, open versus closed, whether synthetic data works, how far chain-of-thought and reinforcement learning go, chips, and policy, which the US and EU have both been flunking. In December 2022 it looked as if OpenAI would run away with everything. Today at least six world-class model companies produce remarkably similar results. Once you know a Turing-test-passing model is possible, you can have five more within a year. Scaling still takes billions, and the only checkbooks are hyperscalers and foreign sovereigns. OpenAI's nonprofit-to-for-profit conversion would require a new entity to buy the assets at fair market value; the last round was some $150 billion, which means raising $150 billion out of the chute. Hallucinations are the creative/correctness fork; lawyers have already submitted briefs with made-up citations. In checkable domains — math, code — synthetic data and RL might kill hallucinations. That itself is a trillion-dollar question. Training data is within a factor of two of all human-generated text; some labs are hiring doctors and lawyers to write more by hand. How many countries want their education, healthcare, and politics run on a model trained to extreme left-wing California politics? He thinks not very many. The EU is locking in a regime so harsh that American AI companies currently will not launch new products there. The election, in his view, tipped the United States from a timeline of social-media-style AI censorship to one that could be quite good, and crypto as programmable money for billions of agents becomes possible again.

On X he is not supposed to be tweeting at all. Instantaneous global communication is the glory; a single tweet out of context is the danger. He once made the entire nation of India hate him overnight, front pages and television news in every time zone. He should write more long-form, and Substack has become a center for it. Yann LeCun is, he thinks, the smartest and most credible critic of LLMs as the path, with a worked-out theory that they are an evolutionary dead end. Andreessen is investing behind LLMs as hard as he can and uses LeCun's arguments to grill the bulls. The other two of the three godfathers of AI he calls far-left, borderline Marxist, their doomerism feeding a push for ruinous legislation; LeCun as anti-doomer and open-source advocate is useful against that. Andrew Huberman convinced him to stop drinking, which ruined the evening scotch and, he jokes, his emotional health while improving his physical health. Sleep advice involving natural light he ignores; a blue screen for three hours then sleep is a test of manhood. Success, for him, is contribution plus a demystifying of happiness: Jefferson meant the Greek sense of fulfillment, not the first ice cream cone. He spent thirty years on machines, is fifteen years into people, and maybe God is the last thirty. Ben Horowitz quotes his father: if a man does not have a real religion, he makes up a fake one, and the fake ones go very badly. Eric Voegelin argued that communism and Nazism were religions in the deep sense, the ones Nietzsche forecast after God died. Woke, Andreessen says, had saints, texts, and sin, and no redemption — perhaps the most dangerous religion you can design. Fridman closes with Thomas Sowell: it takes considerable knowledge just to realize the extent of your own ignorance. Andreessen looks ready for twenty more hours.