Marc Andreessen on Why AI Is Arriving Right on Time and What It Means for Builders
Lenny's Podcast
In this conversation with Lenny Rachitsky, Marc Andreessen, co-founder of Netscape and Andreessen Horowitz (a16z), argues that we are living through one of the most significant periods in modern history. His view on AI is more hopeful than most of the public debate. Andreessen does not expect AI to trigger mass unemployment. He thinks it arrives just as slow productivity growth and falling populations would otherwise push Western economies toward decline. From that starting point he covers how he is raising his 10-year-old, what happens to product managers, engineers, and designers, why he still thinks people should learn to code, how the most AI-forward founders are rethinking the company, and why he distrusts confident predictions about moats and AGI.
A historic moment on three fronts
Rachitsky asked how big a deal the current moment is. Andreessen called 2025 "maybe the most interesting year" of his career and probably his life, and said he expects 2026 to top it. He sees three large shifts happening together.
The first is what he calls a full-scale collapse of trust in legacy institutions, which he says is backed by a lot of data. Many structures people relied on for a long time have, in his words, proven "not up for the challenge." The second is a liberation of national and global conversation. He describes a revolution in freedom of speech and thought, where people can openly discuss things they could not discuss a few years ago, and he believes this is now "a one-way train." The third is a set of major geopolitical shifts in the US, Europe, China, and Latin America, where assumptions are being "pulled out into the daylight and re-examined."
AI lands on top of all three. Andreessen thinks we are at the very beginning of each shift. He compared the combined moment to historical breaks such as the fall of the Berlin Wall in 1989 or the end of World War II, and said it "certainly feels like that."
AI is working, but it is entering a stagnant, shrinking world
On the technology itself, Andreessen said it is now "pretty clear" that AI is really working. When ChatGPT arrived only three years ago, the open question was whether a system that could write Shakespearean sonnets and rap lyrics could also reason in domains that matter, such as medicine, science, and law. In his view the last 12 months, and especially the last three, have answered yes. He pointed to AI producing new math theorems and to AI coding hitting "critical mass" over the holiday break, when some of the world's best programmers, Linus Torvalds among them, said AI was now coding better than they could. He expects AI to become very good at reasoning in any domain with verifiable answers.
He thinks the common industry frame is wrong: the idea that because the technology works, AI will simply sweep the world and change everything. That frame, he argues, rests on an incomplete picture of the world of the last several decades, and he singled out two facts.
The first is productivity growth, which economists use as the measure of technology's impact on the economy. It feels as though the West has lived through 30 to 50 years of rapid technological change, but Andreessen says the statistics show the opposite. By his account, US productivity growth in our lifetimes has run at about half the pace of 1940–1970 and about a third of the pace of roughly 1870–1940. AI is therefore entering an economy that has seen very little real technological progress for a long time.
The second is demographic collapse. Reproduction rates are falling fast, below replacement in many countries including the US, and he says many countries, China included, will depopulate over the next century. He concludes that we need AI to work in order to raise productivity and economic growth, and that we will need machines to do jobs for which there will not be enough people.
Why the job-loss panic is "reductive"
Rachitsky raised the widespread fear that young people's jobs will disappear. Andreessen called the job-substitution model "very reductive" and "overly simplistic." His reasoning goes back to the productivity numbers. After a long period of slow change, and therefore unusually low job churn by historical standards, even a tripling of productivity growth, which would be "a massively big deal," would only bring job churn back to what it was between 1870 and 1930. He says people who lived through that period described a world "awash with opportunity," in which young people built careers in new fields and much of the modern world was invented.
He expects task-level and job-level substitution, but thinks it will be "swamped" by the macro effects of growth and innovation, with hiring booms in many places. He adds the demographic point. He said he did not want to get into politics, but he thinks the world is broadly reversing course on the immigration levels of the last 50 years amid rising nationalism. If declining population is combined with less immigration, "the remaining human workers are going to be at a premium, not at a discount."
When Rachitsky asked whether this depends on the timing lining up, Andreessen reversed the question. Without AI, he said, we would be panicking about depopulation, because depopulation without new technology means a shrinking economy with fewer jobs, fewer new fields, and less consumer demand. He described that scenario as an economy "self-euthanizing" over time. In his view "the timing has worked out miraculously well," since AI and robots are arriving "precisely when we actually need them."
The utopian scenario, followed mechanically
Andreessen said mass job loss would require productivity growth of something like 10, 20, 30, or 50 percent a year, which is orders of magnitude beyond anything any economy has recorded. He said this is possible, admitted his own "utopian temptation," and then walked through what he calls straightforward economics rather than a bold prediction.
Massive productivity growth means more output from less input. That produces gluts of goods and services in the affected sectors, and gluts drive prices down: something that costs $100 falls to $10 and then to $1. He says this is equivalent to giving everyone a large raise, and the new spending power feeds growth and new fields. If unemployment remains, the social safety net becomes much cheaper to provide because the things welfare pays for, such as healthcare, housing, and education, have also fallen in price. Under this scenario, he argues, there is no version where "everybody's just poor."
He was clear that this is not what he actually expects. He thinks change will be more incremental, and that the incremental path will also be overwhelmingly good news.
Revising his debate with Peter Thiel
Rachitsky listed predictions Andreessen got right: the importance of the web, "software eating the world," and his 2011 forecast of 5 billion smartphone users within ten years, which Rachitsky said ended up closer to six billion. He also mentioned a debate in which Andreessen argued against Peter Thiel's claim that technological progress had stalled. Andreessen joked that he has buried his wrong predictions "behind the shed," and then said he has come around much more to Thiel's view.
Thiel's real argument, as Andreessen now puts it, was that there has been plenty of progress in bits and very little in atoms. Andreessen says he glossed over this at the time because he was focused on defending progress in bits. He now agrees the built world is not very different from 50 years ago: bridges from 1930, dams from 1910, cities founded in 1880. He asked where the new cities and dams are, and where California high-speed rail is.
This is also why he does not expect AI to change everything overnight. Much of the world is "wrapped up in red tape": bureaucratic process, regulation, politics, unions, cartels, and monopolies that resist change. His example was healthcare. He called doctors, nurses, and hospitals cartels and noted the push toward nationalized systems, meaning government monopolies, which do not like rapid change. He said ChatGPT is "almost certainly a better doctor than your doctor today" but cannot get a license, prescribe medication, or perform procedures. His optimistic hope is that AI might push society to revisit these assumptions for the first time in decades.
Raising a kid for AI: agency and the philosopher's stone
Andreessen and his wife homeschool their 10-year-old, and they think a lot about how to prepare him. He sees two effects of AI on individuals. AI raises the average by making people who are good at something very good. More striking to him, it makes the truly great "spectacularly great." He said his friends who are excellent coders tell him they are now 10 times as good as they used to be. His goal for his son, and his advice to parents, is to help children become deep in some field in a way that lets them fully use AI.
Rachitsky summed this up as "agency." Andreessen said the term puzzled him at first because it seemed obvious. He came to see it as less obvious in a society where people are trained by default to follow rules and where breaking them causes alarm. He suggested the K–12 system has become increasingly focused on rule-following. He also keeps some structure at home. The night before the interview he told his son that "in order to lead, you must first learn to obey." Still, he sees a large premium on people who take full responsibility, lead projects, and create new things, and he thinks AI is the ultimate lever for a kid with that kind of agency.
That led to the image he uses for AI. Isaac Newton, he noted, spent decades on alchemy, trying to find the philosopher's stone that would turn common lead into rare gold, and never succeeded. With AI, he said, we now have a technology that turns sand, "the most common thing in the world," into thought, "the most rare thing in the world." AI "is the philosopher's stone," and teaching his son to use it is central to how they educate him. He rejected the meme that Silicon Valley people keep their kids off computers. In his experience, the more plugged in people are, the more they want their kids to understand the technology.
One-on-one tutoring for everyone
On education more broadly, Andreessen separated the national question of educating all children from the question of what is best for a single child. For the individual, he said, the answer has been known for centuries: one-on-one tutoring. Royal and aristocratic families always used it. His example was Alexander the Great, tutored by Aristotle. He also cited Bloom's two-sigma effect: tutoring reliably raises outcomes by two standard deviations, taking a student from the 50th to the 99th percentile, because the child stays at the edge of what they can do and is corrected in real time.
Until now only the richest could afford this. AI makes it a realistic option, since a child can ask an LLM unlimited questions, ask it to "dumb it down," and ask it to quiz them. He expects a hybrid of traditional schools plus AI tutoring. He mentioned Khan Academy's nonprofit work, many startups, and Alpha School, which combines in-person teachers with heavy use of AI tutoring. He said there is "a magic formula in here" that he thinks will apply much more broadly.
The three-way Mexican standoff among PMs, designers, and engineers
Asked about the future of product managers, engineers, and designers, Andreessen described a John Woo-style three-way Mexican standoff. Every coder believes that with AI they can also be a product manager and designer. Every product manager thinks they can code and design. Every designer knows they can do the other two. He joked that eventually all three will realize AI can also be a better manager and aim "up the org chart." His view is that they are all roughly correct: AI is now good at many of the tasks in all three jobs.
He does not think any one role wins. He expects the silos to weaken and the most talented people in each role to become "superpowered" and competent in all three, which makes them very valuable because they can design and build new products from scratch. For coders specifically, the first step is to understand how the job is changing, from writing code by hand to orchestrating "a dozen instances" of coding bots.
Jobs are bundles of tasks
To make this concrete, Andreessen used the economists' framing: the job is not the atomic unit of work, the task is, and a job is a bundle of tasks. People talk about job loss when they should be looking at task loss. His example was the executive and the secretary. In 1970 a vice president dictated memos. When email arrived, the secretary printed emails, the executive handwrote replies, and the secretary typed them back in. Today executives handle their own email, and admins focus on travel, events, and other work. Both jobs survived while the tasks shifted, and the executive actually took on more clerical work.
"The job persists longer than the individual tasks," he said, and jobs change only once enough tasks have changed. For individuals, the task is to keep swapping in new tasks, adopting new tools, and expanding scope. He does not know whether the job will be called coder, designer, or "I tell the AI how to build products" in ten years, but he said whoever orchestrates the AI will be "incredibly important."
From human calculators to coding bots, and why you should still learn to code
Rachitsky said it is becoming clear that engineers will soon not write code at all. Andreessen answered with a history of abstraction. "Calculators" originally meant rooms of people, sometimes thousands, doing arithmetic by hand for insurers or the military. Then came machine code in ones and zeros, punch cards, assembly language, and higher-level languages like C. When scripting languages took off (he noted JavaScript came out of Netscape, alongside Python and Perl), there was a fight over whether scripting was "real programming." It was, and most coding today uses scripting languages. Rachitsky said he worked in Perl for years and remembers being looked down on; Andreessen said the skeptics were partly right early on and clearly wrong in the end. AI coding, in his framing, is the next layer: it abstracts away writing the scripting code.
The best programmers now tell him they spend their days orchestrating around ten coding bots in parallel, moving between terminals, arguing with the bots, debugging, and revising specs. That requires knowing how to write code yourself, because otherwise you cannot judge what the bots produce. His son uses Claude, ChatGPT, and Copilot, vibe-codes games on Replit, and "spends two hours at dinner arguing with an AI for fun." Andreessen still tells him he must fully understand code so he can tell when output is wrong or slow.
He puts it bluntly. If your goal is to be a mediocre coder, let the AI do it, since it will generate "infinite amounts of mediocre code." If you want to be among the best, you should understand every layer down to assembly, machine code, the chip, and the network, and also how the AI itself works, because you are always more productive when you know how the machine works. AI is also the best tutor for learning those layers. He expects programmers to become 10, 100, or 1,000 times more productive, with humans still "absolutely" involved.
Capital-D design and the value of taste
Rachitsky suggested that taste and design may become more valuable. Andreessen agreed in a specific sense. Task-level design, such as producing the perfect icon, will be something AI does "all day long," generating a thousand options. "Capital D design" will matter more: what a product is for, how it fits into people's lives, whether it makes them happy or challenges them in the right way. His example was a 25-year-old designer who wants to be Jony Ive in a decade. Ive did everything without AI, and a young designer who fully uses AI could become "the best designer the world's ever seen," with more time for the higher-level questions most designers never reach. Rachitsky said design feels hardest to learn through exposure alone. Andreessen admitted he has always wanted to be a cartoonist but has no art skills, and said "it might be time."
Don't be fungible: the T, the E, and the auteur
Rachitsky proposed a T-shaped strategy: be excellent at one role and good enough at the others. Andreessen connected this to career advice from Scott Adams, whose recent death he called a tragedy. Adams said he could have been a pretty good cartoonist or pretty good at business, but being a cartoonist who understood business made him "spectacularly great" at Dilbert, which neither the best cartoonist nor the best businessperson could have made. The lesson, as Andreessen puts it: being good at two things is worth more than double, being good at three is worth more than triple, because "you become a super relevant specialist in the combination of the domains."
He gave Hollywood as an example. The superstars are writer-directors, the auteurs. Hollywood has its own three-way standoff among director, writer, and actor, each of whom thinks AI can replace the other two, and he thinks each is again correct and able to expand sideways. Rachitsky suggested an "E" or sideways "F" shape with several deep legs. Andreessen agreed and added Larry Summers's advice: "don't be fungible." Someone who is only one thing can be swapped out, while someone with a rare combination becomes one of the few people who can do it.
Andreessen stressed one point he thinks people underuse: AI will teach you. People focus on what to have a model do for them and forget to ask it what it can teach them. He said people who want to develop their careers "should be spending every spare hour" asking AI to train them, for example "I'm a coder, train me how to be a product manager," including having it set assignments and evaluate the results.
Rachitsky shared two tactics he has heard from guests: watch the agent's reasoning as it works to learn architecture, and after getting unstuck, ask the model what you could have said to avoid the problem. Andreessen agreed with the first and tied it to theory of mind. If a colleague brings back the wrong thing, you need to understand their thinking to give useful feedback, and the same holds for AI, which will explain and critique itself all day. He also likes having one AI write code and another debug it, playing them against each other. Rachitsky noted people call these "LLM councils."
How the most AI-forward founders think, in three layers
On founders, Andreessen described three layers. The first is whether AI redefines products. Some technologies, such as flash storage, just get added to existing products. Others, such as the internet replacing on-prem software, remake categories and replace companies. His example was Adobe Photoshop, a roughly 40-year franchise: is AI a feature added to image editing, or do people stop editing images because regenerating one with Nano Banana is easier? He said a16z is betting heavily that many categories will be reinvented.
The second layer is jobs inside companies: whether a founder with budget for 100 coders needs only 10 superpowered ones or keeps 100 who each do ten times more.
The third, which "hasn't quite dropped yet," is whether the idea of a company changes. Can a founder run an entire company by overseeing AI bots, reaching the "one-person billion-dollar outcome"? He named Bitcoin as the most spectacular precedent, followed by Ethereum with a small team, plus Instagram and WhatsApp. The furthest version is AI bots on a blockchain operating autonomously as businesses and paying dividends, which some founders are pursuing. Rachitsky was skeptical. Running his newsletter mostly alone involves endless support tickets, edge cases, and forms. Andreessen answered that Satoshi pulled it off, conceded the open-source community complicates the count, and said he does not claim to have answers, only that many of the smartest people he knows are thinking hard about it.
Moats: humility about structural predictions
Asked about moats, Andreessen said major technological shifts take a long time to play out, with structural effects spreading over years. Early on there is a "rush to judgment," with confident claims about which companies, incumbents, and layers will win, and the media amplifies them because it prefers definitive answers to open questions. Reading internet coverage from 1993 to 1997, or even through 2005 or 2010, he said, shows that almost all such confident statements were wrong, often badly.
He argued both sides for models. Billions in cost, scarce compute and data, engineers paid "like NBA stars," and heavy regulatory and reputational burdens all point toward an oligopoly or monopoly, which software has produced before. But if someone had told him at the ChatGPT Christmas that within a year to a year and a half there would be five other American companies with comparable products, five in China, and open-source equivalents running on a fraction of the hardware, he would have called that fast commoditization, and he says that is what happened. He listed Google, Anthropic, xAI, Meta, DeepSeek, and Kimi. Apps are similarly unclear: models may absorb everything, or the value may shift to apps adapted to domains such as medicine, law, and coding. He knows smart people on both sides.
His example was Claude Code, which he called "a Claude wrapper" and an agent harness, and Cowork, which Rachitsky noted was built with Claude Code in about a week and a half. That is impressive, Andreessen said, but it also raises the question of how much barrier to entry something built in a week and a half can have. He expects every other model company to build similar tools and wondered whether Claude Code could be "lapped" the way GitHub Copilot was. He said some of the smartest people he knows admit, after a couple of drinks, that there may be few secrets among the big labs, with DeepSeek, which he called "basically a hedge fund in China," as evidence. On the other hand, well-paid researchers may hold a dozen hard-to-copy ideas. He called it a complex adaptive system and said he puts a "big discount" on his own forecasting. As a VC, he said, that means betting across every strategy, and he has not met anyone who has it all figured out.
Indeterminate optimism as a strategy
Asked whether a16z's approach, including what Rachitsky called the largest fund in history, is simply to place many bets, Andreessen used Peter Thiel's two-by-two of optimism versus pessimism and determinate versus indeterminate. Thiel has criticized Silicon Valley's "indeterminate optimism," the belief that things will get better without saying how, and praised determinate optimists such as Elon Musk with concrete plans like electric cars, solar, and Mars.
Andreessen places himself and a16z "firmly" in the indeterminate camp and argues it is not wishful thinking. The system is made up of hundreds, thousands, and tens of thousands of determinate optimists, the founders, each with a specific plan. The best response to an unknowable future is to run as many experiments as possible. Founders must be determinate optimists. He acknowledged their complaint that VCs can diversify while founders make one bet, and replied that founders get to steer. History remembers Henry Ford, not his seed investor, and founders deserve "99.99999%" of the credit, though he thinks indeterminate optimists in the background help keep the cycle going. Later he extended this to Silicon Valley itself, which he called a "company town" whose company is the Valley. It no longer makes silicon, yet with AI it has moved into what he counts as its ninth major platform without anyone planning the sequence.
AGI: the cosmic version overstates, the prosaic version understates
Andreessen said he has always struggled with the concept of AGI. The "cosmic" definition is the singularity: a self-improving takeoff in which machines decide faster than humans and human judgment stops mattering. He does not think we live in that world. The "prosaic" definition is AI doing economically valuable tasks as well as a human. Rachitsky noted an Anthropic co-founder frames it as a basket of the most valuable tasks. Andreessen said we are getting close to that, if not already there.
He thinks the prosaic definition understates what is coming, because nothing suggests human skill is a ceiling. Human limits come from biology. He said human IQ tops out around 160, the Einstein and Feynman level where new physics comes from. He then gave rough figures: about 140 for top authors, scientists, and doctors, 130 for a good lawyer, 110 for a good line manager, 105 for a good small-business accountant. He said current models test around 130 to 140 and are arguably approaching 160 on math, and he expects 180, 200, 250, even 300 fairly soon. He considers this as good as getting more Einsteins, and expects AI coders, doctors, and lawyers better than the best humans. In his view, reaching human equivalence will be "a footnote," something that happened "on a Tuesday in 2026," with the real story being exploration beyond human capability.
He explained the emotional side of this from his own experience. He is constantly frustrated that he lacks the hours or years to learn enough, cannot do the math in his head, and forgets most of the ten books he reads on a subject. He also regularly meets people smarter than he is: he takes notes through the first half of the conversation and feels outthought in the second, then goes home "to have a drink." Machines without those limits, working for us, excite him more than he thinks people appreciate.
Media diet: X and old books
Andreessen describes a "barbell strategy": he reads X for what is happening now and old books that have lasted 50 years, and he is skeptical of everything in between. He suggests rereading last week's newspaper and noticing that "none of this happened" the way it was predicted, and says magazines are worse because of their longer cycles. He thinks direct content from practitioners, through Substack, newsletters, and podcasts like Rachitsky's and Lex Fridman's, is still "dramatically underrated." He grants that people "talk their book" but finds they love explaining their work, and says there is "tremendous amounts of alpha" in listening to leading experts. Rachitsky noted Silicon Valley's culture of sharing. Andreessen said that from a single CEO's view, leaking secrets and departing employees are painful, but the ecosystem benefits as a whole. The two joked that everything is now a wrapper, making them all "sand wrappers."
His favorite film of the past year, and he said possibly of the past 15, is Eddington. Joaquin Phoenix plays a crusty right-wing sheriff and Pedro Pascal a young progressive mayor in a New Mexico town of about 600 people, starting in March 2020 and running through COVID and the George Floyd protests, with a thinly disguised Meta building an AI data center nearby. He praised it as the first film to engage directly with 2020 and to show how people experienced real-world events through phones and social media. He said he and the director would probably disagree on much, but credits the director for grabbing "all the third rails."
Product diet and closing
His son discovered Replit on his own about three months earlier. Andreessen noted that by age 10 anything a parent works on is "deeply uncool," and his son now spends hours vibe-coding, mostly Star Trek simulators in the Next Generation LCARS interface style. Andreessen is also taken with AI voice products. His dinner-party trick is Grok's foul-mouthed raccoon avatar Bad Rudy. He mentioned a16z portfolio company Sesame, which went viral for emotionally intimate voice experiences, and said he expects a wearables revolution including pendants and Meta glasses. He praised Whisper Flow, a voice transcription app that understands instructions like "I want bullet points" as commands rather than text to type.
To close, he recommended a recent piece by Packy McCormick that he called the best explanation of what a16z does and how it thinks, along with a16z's YouTube channel.
If we didn't have AI, we'd be in a panic right now about what's going to happen to the economy. We've actually been in a regime for 50 years of very slow technological change in the face of declining population growth. The timing has worked out miraculously well. We're going to have AI and robots precisely when we actually need them. The remaining human workers are going to be at a premium, not at a discount.
How big of a deal is the moment in time that we are living through right now?
This is a very, very historic time. AI is the philosopher's stone. Now we have a technology that transfers the most common thing in the world, which is sand, converted into the most rare thing in the world, which is thought.
We spent a lot of time with the most cutting-edge AI-forward founders. The most leading-edge founders are thinking of, can you have entire companies where the founder does everything?
There's all this concern that young people, jobs are not going to be there for them. AI is replacing them.
Everybody wants to talk about job loss, but really what you want to look at is task loss. The job persists longer than the individual tasks.
What's your sense of just the future of three very specific roles? Product manager, engineer, designer.
There's like a Mexican standoff happening between those three roles. Every coder now believes they can also be a product manager and a designer because they have AI. Every product manager thinks they can be a coder and a designer. And then every designer knows they can be a product manager and a coder. They're actually all kind of correct. What happens is the additive effect of being good at two things is more than double. The additive effect of being good at three things is more than triple. You become a super relevant specialist in the combination of the domains.
People aren't fully grasping how much this is changing. And people who really want to improve themselves and develop their careers should be spending every spare hour, in my view, at this point talking to AI, being like, "All right, train me up."
Today my guest is Marc Andreessen, one of the most seminal figures in tech and in business. He invented the web browser, built the world's largest venture firm. He's also a multi-time founder and an investor in essentially every generational tech company and is also one of the most clear-minded, lateral, and insightful thinkers about both the past and the future of technology. In this very special conversation, we chat about how unique and significant the moment that we are all living through right now is, what skills he's teaching his kids to thrive in the AI future, what happens to product managers, designers, and engineers in the coming years, where moats exist in AI, what the most AI-native founders are doing differently, and so much more that is just scratching the surface of this very deep and important conversation.
You are going to walk away from this chat being smarter about what is going on in the world right now and where things are heading.
A huge thank you to my newsletter community and followers on X for suggesting topics and questions for this conversation. If you enjoy this podcast, don't forget to subscribe and follow it in your favorite podcasting app or YouTube. It helps tremendously. And if you become an insider subscriber of my newsletter, you get a year free of over 20 incredible products, including a year free of Lovable, Replit, Bolt, Gamma, n8n, Linear, Superhuman, Devin, PostHog, Descript, Wispr Flow, Perplexity, Warp, Granola, Magic Patterns, Raycast, ChatPRD, Mobbin, and Stripe Atlas. Head on over to lennysnewsletter.com and click Product Pass. With that, I bring you Marc Andreessen after a short word from our sponsors.
Today's episode is brought to you by DX, the developer intelligence platform designed by leading researchers. To thrive in the AI era, organizations need to adapt quickly. But many organization leaders struggle to answer pressing questions like which tools are working? How are they being used? What's actually driving value? DX provides the data and insights that leaders need to navigate this shift. With DX, companies like Dropbox, Booking.com, Adyen, and Intercom get a deep understanding of how AI is providing value to their developers and what impact AI is having on engineering productivity. To learn more, visit DX's website at getdx.com/lenny. That's getdx.com/lenny.
If you're a founder, the hardest part of starting a company isn't having the idea. It's scaling the business without getting buried in back office work. That's where Brex comes in. Brex is the intelligent finance platform for founders. With Brex, you get high-limit corporate cards, easy banking, high-yield treasury, plus a team of AI agents that handle manual finance tasks for you. They'll do all the stuff that you don't want to do, like file your expenses, scour transactions for waste, and run reports, all according to your rules. With Brex's AI agents, you can move faster while staying in full control. One in three startups in the United States already runs on Brex. You can too at brex.com.
Marc Andreessen, thank you so much for being here and welcome to the podcast.
Awesome, Lenny. Thank you. It's great to be here.
I want to start with just a big picture question. I have a billion directions I want to go, but I think this is going to give us a little bit of a frame of reference. How big of a deal is the moment in time that we are living through right now?
This is a very, very historic time. I think 2025 was maybe the most interesting year in my entire career and probably life, and I think I would expect 2026 to exceed that.
Wow, that says a lot.
Yeah, I've seen some stuff. So it feels like two things are happening. One is the trust that a lot of people have had in kind of what you describe as kind of legacy institutions around the world is, I think, in kind of full-scale collapse right now. By the way, there's a lot of data to support that. And so I think there's just a lot of structures and orders and institutions that people have just relied on for a long time that have just proven to not be up for the challenge.
And then kind of corresponding with that is the national and global conversation have become, like, let's say, liberated. And so, you know, this sort of incredible revolution that we have in kind of, you know, what I've described as freedom of speech, freedom of thought, ability for people to openly discuss things that maybe they couldn't discuss even a few years ago, you know, is just dramatically expanded. And I think that's now on a one-way train for just a much broader range of discourse.
And then, you know, there's also just these incredibly massive geopolitical shifts that are happening. And obviously the US is changing a lot, Europe is changing a lot, China is changing a lot, Latin America, by the way, is changing a lot, very dramatic, you know, events playing out down there right now, you know, kind of all over the world. Like, I think a lot of assumptions are being pulled out into the daylight and re-examined.
And then it's kind of the fact that all these things are happening at the same time, right? And so you've got all of these countries and industries, you know, where things are kind of increasingly in upheaval, but you have AI as this kind of new technology that's going to really affect things. And then you've got, you know, people, you know, citizens being able to fully participate and being able to argue things out. So it's kind of like those three kind of big mega things are kind of all colliding at the same time. And I think we're probably just at the very beginning of all three of those. And those all feel like kind of, you know, historical, you know, moment shifts, you know, comparable in magnitude to maybe the fall of the Berlin Wall in 1989, you know, maybe the end of World War II, you know, kind of moments like that. It certainly feels like that.
Good God, what a time to be alive.
Yeah. In terms of the AI piece, which is where a lot of people are trying to figure out what to do, what do you think isn't being priced in yet in terms of the impact AI is going to have on, say, the world or just people listening?
I think at this point it's pretty clear, with, you know, our technology hats on, that this stuff is really working now, right? And so there was this, you know, kind of, when there was the ChatGPT moment, you know, three years ago — it was only, by the way, only three years ago, right, was the ChatGPT moment — and the big question was, all right, this is incredibly fun and creative, and like, we have machines now that can compose Shakespeare and sonnets and rap lyrics, and, you know, this is amazing. But then there was, you know, this big question: can you harness this technology for reasoning and for, you know, problem solving in domains that really matter, you know, medicine and science and law and so forth? And, you know, it turns out the answer to that is yes, right?
And, you know, the last 12 months, and especially even just the last three months, have really proven that AI can really do — like, you know, you're seeing it all now. You know, AI is now developing new math theorems. You know, over the holiday break, you know, it feels like the AI coding thing, you know, really hit critical mass, and the world's best programmers, right, including like Linus Torvalds, you know, for the first time over the holiday break basically said, yeah, AI is now coding better than we can. And so that's incredibly, incredibly powerful. And I think we all, you know, kind of, I think, assume that AI now is going to get really good at reasoning in any domain in which there are verifiable answers. And so that, you know, that's going to include many very important domains. So for the technology, it feels like it's moving fast and it's going to be working really well.
I think the thing that is not well understood — I think a lot of people in the industry have kind of what I would describe as this one-dimensional thing, which is, okay, as a result of the technology now working, AI just kind of sweeps the world and changes everything. And I think that's kind of the wrong frame. I think it's based on an incomplete understanding of the world that we live in, or the world that we've been living in for the last, you know, 80 years. And I would call out two things in particular.
So one is, I think it's felt to us, like in the US and the West, for the last, you know, whatever, 30 years or 50 years, it's felt like we've been in a time of great technological change. But actually, if you look for actual evidence of that, like statistical evidence of that, analytical evidence of that, you basically can't find it. And in particular, economists have a way of measuring the rate of technological change in the economy, that is productivity growth, which we could talk about what that means, but basically it's sort of the mathematical expression of the impact of technology on the economy. And productivity growth for the last 50 years has actually been very low, not very high. So we all feel like it's been very high, there's been lots of technological change. What's actually happening is it's been very low. And in fact, the pace of productivity growth in the US, in my lifetime, in our lifetimes, has been running at about half the pace that it ran between 1940 and 1970. And it's been running at about a third the pace that it ran between about 1870 to about 1940. And so statistically, in the US, in the West, technology progress in the economy, technology impact on the economy, has actually slowed way down. And so, you know, the AI thing is going to hit, but it's hitting an environment in which we have actually had almost no technological progress in the actual economy for a very long time. So we could talk about that.
And then there's this other just incredible thing that's happening, which is the, you know, demographic collapse, right? It's sort of a Western phenomenon, an increasingly global phenomenon, which is, you know, the rate of reproduction of the human species is in rapid decline. And, you know, there are many countries, you know, including the US, where, you know, the rate of reproduction is, you know, under two, you know, meaning that, you know, many countries around the world — by the way, including China, which is a really big deal — are actually going to depopulate over the next century.
And so you have this kind of precondition that says there's actually been very little technological progress happening in the world, and the world is going to depopulate. And so AI is going to enter a world in which those two things are true. And I think this is incredibly important, because we actually need AI to work in order to get productivity growth up, which is what we need to get economic growth up. And we actually need AI to work because we're going to need, you know, machines to do all the jobs that we're not going to have people to do, because we're literally going to depopulate the planet over the next hundred years. And so I think the interplay of these factors is going to be much more interesting and, frankly, more complex than a lot of people have been thinking.
I'm going to follow this thread about kids. I know you have a kid, and one of my favorite lenses into how people think and what they value is what they're teaching their kids, what they're steering their kids towards. Are there specific skills or even careers that you're steering your kid towards?
The way I think about this — and, you know, yeah, we have a 10-year-old, and, you know, we actually homeschool, and so we think a lot about this. So I think the way to think about the impact of AI on people, specifically on people as individuals — a lot of people just focus on kind of this, you know, very, I would say, straightforward and/or overly simplistic view of just literally job gains, you know, job losses, which we could talk about. But there's two specific things at the level of an individual person and individual kid.
So I think it's pretty clear that AI is going to take people who are good at doing things and it's going to make them very good at doing things, right? And so it's going to be a tool that's going to sort of raise the average kind of across the board. And, you know, look, you see that playing out already. You know, anybody who's in a position where they need to, you know, write something or design something or write code or whatever, if they're pretty good at it today, they use AI and all of a sudden they're very good at it. And so there's sort of that aspect to it. And I think the way the education system writ large is going to teach AI is going to be based, you know, hopefully a lot on that.
But then there's this other thing that's happening, which we're also starting to see, and we're really seeing it particularly in coding right now, where the really great people are becoming spectacularly great, right? And so you kind of use the term, you think about, like, the superpowered individual, right? So the individual who is really good at coding or really good at making movies or really good at making songs or really good at designing, you know, making art or whatever those things are, or, you know, podcasting, or, you know, hopefully venture capital — you know, if you're very good at it and you can really harness AI, you can become spectacularly great and super productive, right? And, you know, I'm sure
You have a lot of friends in this category as well, but the really, really good coders are experiencing this, right? My friends who are really good coders are like, "Oh my god, all of a sudden I'm not twice as good as I used to be. I'm like 10 times as good as I used to be." And so I think at the unit of n equals one of an individual kid, I think the question is kind of how do you get them in a position where they're kind of this superpowered individual such that they're going to be really deep in whatever it is they're going to do, but they're going to be deep in a way that's going to let them fully use the power of AI to be not just great, but to be spectacularly great. And I think that's going to be the real, you know, that's the real opportunity, and at least that's what we're shooting for and that's what I would encourage parents to shoot for.
So, what I heard there is essentially agency, this word that we see on Twitter all the time, is building agency, them not waiting for someone to tell them what to do, figuring out what to do.
Yeah. So this term agency that's become very, very popular [laughter], certainly in California for the last couple years. It's really interesting because I had a lot of trouble with this early on, because I'm like, agency, what are they talking about? And what they're kind of talking about is initiative, willingness to, you know, you could just do things. What is it? Samo Burja has the great term live player. You can be a primary participant in events. And at first I was like, well, yeah, that's kind of obvious, right? Of course. And then I'm like, oh, actually it's not so obvious anymore, because, kind of to your point, I think so much of our society is based on like there are all these rules and everybody gets taught kind of by default you're supposed to follow all these rules, right? And then if you break the rules, everybody gets freaked out. It's like, oh my god, he broke the rules. And so we have somehow worked our way kind of, I don't know, psychologically, sociologically, into a state in which I guess the natural assumption for a lot of people is the thing that you, for example, want to train kids to do is follow all the rules. And you could argue that, for example, the K through 12 school system or whatever has gotten kind of more and more focused on that over time.
And it's like, yeah, it's like no, you should actually, and again, especially at unit n equals one, like of your kid. And look, there's something to be had. I just had this conversation with my 10-year-old last night actually. I rolled out the concept of, you know, in order to lead, you must first learn to obey, right? In order to issue orders, you must learn how to follow orders. And you kind of try to keep him with some level of structure in his life and not just pure agency. And so look, some rules are important and so forth, but yeah, look, there's just a huge premium in life on being somebody who is able to fully take responsibility for things, fully take charge, run an organization, lead a project, create something new. And maybe that has been a little bit diminished in our culture over the last 30 years. It's healthy that there's now a term for that that is coming back into vogue. And again, that's how I view AI for kids: okay, AI should be the ultimate lever on the world for a kid with agency to be able to say, okay, I can actually be a primary contributor, right? Whether that's I can be a primary contributor in everything from developing new areas of physics to writing code to being an artist to writing novels, whatever that thing is, I can fully participate in the world, I can really change things. The combination of that idea combined with this technology feels very healthy to me.
What is that quote about? Give me a lever and I'll move the world.
And I'll move the world. Yeah, that's exactly right. Well, so it's actually funny you mentioned that. So the early kind of scientists, including Isaac Newton, were super obsessed with this concept of alchemy, right? Newton developed Newtonian physics and he developed calculus and all these things, but the thing he was really obsessed with was alchemy, which was the thing he could never get to work, right? And alchemy was the transmutation of lead into gold, which meant the transmutation of something that was very common, which was lead, into something that was very rare and valuable, which was gold. And he spent decades trying to figure out this thing called the philosopher's stone, which would be basically the machine or the process that would be able to transmute the common thing into the rare thing, lead into gold. And he never figured it out, and it's incredibly frustrating, nobody ever figured that out. And now we literally with AI have a technology that transmutes sand into thought. [laughter]
Just blew my mind.
Right. The most common thing in the world, which is sand, converted into the most rare thing in the world, which is... right. And so AI is the philosopher's stone. It actually is that, and it's just this incredibly powerful tool. And that's where I get so excited. And again, this is what we're doing with our 10-year-old, which is, all right, a primary thing that we want to make sure to do is to make sure that he knows fully how to leverage and get benefit out of the philosopher's stone, right? Which is to say AI. And that's certainly central to everything we're teaching him. There's this meme going around that Silicon Valley people don't let their kids use computers. There may be a handful of people who are like that. I don't know. I think it's more honestly the other way around, which is the more you're kind of plugged into stuff in Silicon Valley, the more important it is to make sure that your kids actually fully understand this and know how to use it. And that's certainly the mode that we're in, and that's certainly the mode that I would encourage parents to think about.
I did not know your kid was homeschooled. That is super interesting. It's almost a statement on education in today's day. Maybe, is there any thoughts there? Just for folks that maybe aren't in your tax bracket that want to help their kids be successful, maybe homeschooled, maybe not, what advice would you have?
This is the challenge, and again, this kind of goes to your original question, which is education. There's two completely different ways to talk about, think about education. The way that's usually thought about and talked about is kind of at the level of a nation, right? So it's a national level issue, or maybe a state level issue in the US, which is basically how do you educate all the kids? And of course that's incredibly important, and of course you're going to need some level of large scale system like the national K through 12 school system or something like that in order to do that. But then there's this other question, which is at n equals 1 for an individual kid, what can you do with an individual kid?
And so I'll just give you kind of the ultimate answer to that question, which is it's been known for centuries that the ideal way to teach a kid at the unit of n equals 1, by far the ideal way to do it, is with one-on-one tutoring. If you just have an individual kid and the goal is to maximize an individual kid, by far you get the best results with one-on-one tutoring. And this is something that every royal family knew in history. It's something that every aristocratic class knew in history. There's all these amazing examples. Alexander the Great was tutored by Aristotle. He took over the world, right? Many of the great kings and queens and royal families and aristocrats and so forth, over the course of centuries, kind of always had this approach. There's actually also statistical evidence, analytical evidence, that this is correct. There's this massive question in the field of education, which is how do you improve educational outcomes? And basically it turns out it's very hard to improve educational outcomes, except there's one method that always does it, which is called the Bloom two sigma effect, which is there's one method of education that routinely raises student outcomes by two standard deviations and will take a kid from the 50th percentile to the 99th percentile, and that's one-on-one tutoring.
Right. So again, if you go back to n equals one, you have a kid and a tutor and they're in this very tight loop with each other, where the kid is able to constantly kind of be on the leading edge of what they're capable of doing, and they can move incredibly fast and they get kind of correction in real time, you get these better outcomes. But to your question, it's never been economically feasible for anybody other than the richest people in society to be able to provide one-on-one tutoring for kids. AI provides the very real prospect of being able to do that, right? Because obviously now, if you have a kid that's super interested in something, they can talk to an LLM about it and they can ask an infinite number of questions and they can get instantaneous feedback. And in fact, you can even tell an LLM, teach me how to do the following. And you can say, wow, I don't quite understand what you're saying, dumb it down for me a little bit. Okay, now quiz me, do I actually understand this? People can just do this today, right? And so I think there's this massive opportunity for parents in many walks of life, with a little bit of time and focus, to be able to say, okay, my kid's probably still going to go through a traditional education system, but I'm going to augment this with AI tutoring. And of course there's going to be tons of startups, right? And there already are, that are going to try to build all the products and services for this. Khan Academy, on the nonprofit side, has a big push to do this. And so I think the broad answer might be a hybrid approach with schools plus one-to-one tutoring through AI.
There's also this great, you may have heard, there's this great new private school system called Alpha, in which everything I just described is kind of the basis of their philosophy, which is it's a combination of in-person schools and teachers, but it's also heavily based on AI and AI tutoring. And so I think there is a magic formula in here that I think is going to apply much more broadly. And for parents interested in this, now would be a great time to really start to think hard about that and to look at the options.
It's interesting because there's all this concern that young people, jobs are not going to be there for them, AI is replacing them. On the flip side, there's what you're describing here. It feels like people coming into learning today are going to move so fast and learn so much more. Where do you sit on this divide of young people are in big trouble or they're actually going to be the ones winning in the end?
Yeah. So the job substitution, job loss thing is just very reductive. I think it's an overly simplistic model. And again, it goes back to what I said at the very beginning, which is we've actually been in a regime for 50 years of very slow technological change in the economy. So again, like I said, it's at half the rate of the previous era and a third the rate of 100 years ago. And so we're coming out of this kind of phase where we've had almost no technological progress in the economy. We've had remarkably little job churn as a result of that relative to any historical period. And so even if AI ticks up, even if AI triples productivity growth in the economy, which would be a massively big deal, it would take us back to the same level of job churn that was happening between 1870 and 1930. And if you go back and you read accounts of 1870 to 1930, people just thought the world was awash with opportunity, right? At that rate of technological transformation, kids were able to develop new careers in new areas of the economy, building new kinds of products and services. A huge part of everything in our modern world today was kind of invented and proliferated during that period. And so even if AI triples the pace of economic change in the economy, it's going to just translate to a much higher rate of economic growth, it's going to translate to a much higher rate of job growth. And there will be some level of task level and job level substitution that will take place, but that will be swamped by the macro effects of economic growth and innovation that will happen, and corresponding to that there will be hiring booms, quite honestly I think, all over the place.
And then again, go back to the other thing, which is this is all happening in the face of declining population growth and increasingly population shrinkage. And so human workers in many, many countries over the next 10, 20, 30 years are going to be at more and more of a premium, literally because you're going to have shrinking population levels. We don't really want to get into politics particularly, but it does feel like the world broadly is going to reverse course on the rates of immigration that we've had for the last 50 years. It seems to be kind of a broad-based thing happening, with the rise in nationalism, concerns about the rate of immigration, and immigration historically in countries like the US has kind of ebbed and flowed over time based on how the national mood shifts. And so if you combine, in a country like the US or any country in Europe, declining population with less immigration, the remaining human workers are going to be at a premium, not at a discount. And so I think that combination of faster productivity growth, faster economic growth, and then slower population growth and less immigration actually means there's going to be much less of this kind of dystopian no jobs thing. I just think it's probably totally off base.
That is extremely interesting. So what I'm hearing is you're not super worried about job loss. Is the key here that the timing kind of just works out, this population decrease, you know, like all
these kind of have to line up for there not to be this massive job loss with AI?
Yeah. Well, look, if we didn't have AI, we'd be in a panic right now about what's going to happen to the economy. Right? Because what we'd be staring at is a future of depopulation, and depopulation without new technology would just mean that the economy shrinks. Right? So it would mean that the economy itself kind of shrinks over time. The opportunity diminishes. There are no new jobs. There are no new fields. There's no new source of consumer demand for spending on things. And so you would be very worried about going into a period of severe decline or stagnation. And essentially you'd be looking at these very dystopian scenarios of an economy kind of self-euthanizing itself over time. And so you'd be very worried about the opposite of what everybody thinks that they're worried about.
The only reason we're not worried about that is because we now know that we have the technology that can substitute for the lack of population growth and then also for the lack of immigration that's likely. And so I would say the timing has worked out miraculously well in the sense that we're going to have AI and robots precisely when we actually need them, to keep the economy from actually shrinking. And I just think that's fundamentally a good news story.
To get to the mass job loss thing that people are worried about on the other side of things, you'd have to look at far, far, far higher rates of productivity growth. You'd have to look at rates of productivity growth that are 10, 20, 30, 50% a year, something like that, which are orders of magnitude higher than we've ever had in an economy in the history of the planet. It's possible that we get that. I mean, look, I have my utopian kind of temptation along with everybody else.
If AI radically transforms everything overnight, then let's play out the kind of utopian scenario. You get to a much higher level of productivity growth. You get to a much higher level of technological change. Corresponding to that, you'll have a massive economic boom. You'll have massive growth in the economy, and then corresponding with that, you'll have a collapse in prices. And so the prices of goods and services that are, whatever you want to call it, affected by or commoditized by AI, the prices of those goods and services will collapse, right? There'll be price deflation, and then as a consequence of price deflation, everything that people are buying today gets a lot cheaper, and that's the equivalent of a gigantic increase in wealth right across the society.
Think of it this way. This is actually worth talking about because people I think get kind of sideways on this issue. If AI is going to transform the economy as much as the utopians or dystopians or whatever kind of think that it will, the necessary economic calculation of what happens is massive, massive productivity growth. The consequence of massive productivity growth, what that literally means mechanically, is more output requiring less input, right? So you get more economic output for less input, right? So you're substituting in AI for human workers or whatever. And as a consequence, you get this massive boom in output with much lower input costs. The result of that is you get lots of goods and services in all those affected sectors. The result of those gluts is you get collapsing prices, right? The collapsing prices mean that the thing that today cost you $100 now costs you $10 and now costs you $1. That's the equivalent of giving everybody a giant raise, right? Because now they have all this additional spending power. That additional spending power then translates to economic growth, right? The development of new fields. Everybody's materially much better off very quickly.
And then, by the way, to the extent that you do have unemployment coming out the other side of that, it's now much cheaper to provide the kind of social safety net to prevent people from being immiserated, right? Because the prices of all the goods and services that a welfare program has to pay for, they're all collapsing, right? And so the price of healthcare collapses, the price of housing collapses, the price of education collapses, the price of everything else collapses because of this incredible impact that AI is having. And so in this kind of utopian-dystopian scenario that people have, there's no scenario in which everybody's just poor. In fact, it's quite the opposite, which is everybody gets a lot richer because prices collapse, and then it's actually much easier to pay for the social safety net for the people who, for some reason, can't find a job.
And so maybe we end up in that scenario. I mean, the kind of optimistic part of me says, yeah, maybe AI is that powerful and maybe the rest of the economy can actually change to accommodate that and maybe that'll happen. But the result of that is going to be a much better news story than people think it's going to be.
And again, everything I've just described, by the way, is just a very straightforward extrapolation on very basic economics. I'm not making any bold predictions with what I just said. This is just a straightforward mechanical process that plays itself out if you have higher rates of productivity growth, which are necessarily the result of higher rates of technological growth. And so, to be clear, I think we're looking at a world that's not radically transformed the way that maybe the utopians think that it will be or the dystopians think it will be. I think it'll be more incremental, for reasons we can discuss. But I think that incremental process is overwhelmingly going to be a good news process. And then even if it's much faster, it's also going to be a good news process. It'll just be a good news process in the other way that I described.
I love hearing optimism and good news. I will also add that I was researching you ahead of this chat, and you've been right so many times about where the world is heading. That's why I'm especially excited to talk to you. I'll give you a short list. I imagine there are many more things. Okay. So, one, you were right about the web and web browsers becoming important. You were right about software eating the world. Check. In 2011, you said that in 10 years we're going to have 5 billion people using smartphones. And I believe the actual number ended up being six billion.
You also had this debate with Peter Thiel that I came across where you were debating whether technology stops progressing or if new technology will continue to emerge. And you were arguing there is progress, progress will continue. And he was like, "No, I think we're done with cool technology." You were right. I imagine there are many more things you were right about. So again, I love hearing your predictions because I feel like they're actually going to turn out to be correct.
So, I should start by saying I've been wrong about tons of things, but I buried those out back behind the shed.
Delete them from the internet. No web browser can discover them.
Yes, I have them nuked out of the internet archives so they're never seen again. So I'm wrong plenty of times also. But yeah, I mean, look, some of those I got right. By the way, I will say on the Peter one, I've come much more around to Peter's point of view.
I would probably argue that one quite a bit differently today than I did, and I would give his view a lot more credit. And it actually goes to the kind of conversation we just had, which is the real form of what Peter was arguing was we have lots of progress in bits, right? But we have very little progress in atoms, right? And that's the real core of what he was arguing. And I think I was a little bit, I don't know, missing that or kind of glossing that over a little bit, because I was so focused on making sure people understood no, there actually is still progress happening in bits. But I think a lot of his critiques around the lack of progress in atoms is real.
And again, this goes back to this thing of, and he's talked about this for a long time, in the last 50 years there has just been very little technological innovation in most of the economy. There's been very little technological innovation in particular in anything involving atoms. There's been very little real-world technological change. The built world is just not that different today than it was 50 years ago. And again, if you compare and contrast 1870 to 1930, it was a dramatically different world. If you contrast 1930 to 1970, it was a dramatically different world. If you contrast 1970 to today, it's not that different, right? And look, you just see that. You could just walk around and it's just like, oh yeah, there's a bunch of buildings that were built in like 1960, right? And there's a bridge that was built in like 1930, and there's a dam that was built in like 1910, and there's a city that was founded in 1880, and like, what have we done, right? Where are new cities? Where are new dams? Where's the California high-speed rail? Like, what's going on here? And so I think he is right about a lot of that.
Again, this is also why I think that AI is not going to have as rapid an impact. It's not going to be, again, this kind of utopian or dystopian view of everything changes overnight. I think it just kind of can't happen because of the reasons that Peter articulates, which is there's so much about how the world works that's basically just wrapped up in red tape: bureaucratic process, rules, restrictions, the politics, by the way, unions, cartels, monopolies. There are all these structures in the world that are kind of economic or political or regulatory structures that basically prevent things from changing.
And so, let's take a great example, like AI's impact on the healthcare system. By rights, AI is going to have a dramatic impact on the healthcare system, and in very positive ways. But large parts of the medical system today are cartels, right? And so the doctors are a cartel, and nurses are a cartel, and hospitals are a cartel, and then there's this push to nationalize all the healthcare systems, and then you've got a government monopoly, right? And guess what cartels and monopolies don't like: they don't like rapid change, right? And so you show up as a kid and you're like, "Wow, I've got this new technology to do AI medicine." And they're like, "Oh, well, does it threaten doctors' jobs? Well, in that case, we're going to block it." And I think a lot of consumers, by the way, I see this in my life and you'll probably see this in your life also, which is ChatGPT is almost certainly a better doctor than your doctor today, but ChatGPT can't get a license to practice medicine, right? So it can't substitute for a doctor. It can't prescribe medications, right? It can't perform procedures, right?
And so, anyway, Peter I think was very articulate, and has been for a long time, on: no, there are actually real structural impediments in the economy and in the political system that we have that actually prevent rates of change that are anywhere near the rates of change that people had in the past. And you can maybe say optimistically, maybe the presence of the new magic technology of AI causes us to revisit a lot of these assumptions for the first time in decades, to really say, okay, is this really the world we want to live in? Don't we actually want to get to the future faster? So maybe that would be the optimistic view.
It's time to build. Somebody famously said. In my calendar, I actually have that as my block when I start to work. It's time to build. That's my block in the morning of the day. Thank you for that.
Okay. I love the way you go from just macro to just n of one. And I want to go to n of one. A lot of the listeners of this podcast are product managers. They're engineers. They're designers. There's a lot of founders, but there's also a lot of non-founders. There's a lot of people building product that aren't founders, and obviously a lot of people are worried about where their career is going. Is one of these roles going to disappear? Is one of these roles going to do really well? How do I stay up to date? You're close with a lot of teams, a lot of product teams. What's your sense of just the future of these three very specific roles? Product manager, engineer, designer.
This I think is a really funny question. So these three roles in particular obviously are kind of the central roles for building tech companies. So, the way I've been describing it is, you know the concept of the Mexican standoff, right? Which is the movie scene where the two guys have guns pointing at each other's heads.
And then if you watch John Woo movies, he loves to do the three-way Mexican standoff where you've got a triangle of people, and of course it's a John Woo movie, they've got guns in both hands.
So each is aiming at the other two.
Yeah. And you've got this kind of standoff situation. And so the way I've been describing this is there's a Mexican standoff happening between those three roles, between product manager, designer and coder. Specifically the following, which is every coder now believes they can also be a product manager and a designer, right, because they have AI. Every product manager thinks they can be a coder and a designer. And then every designer knows they can be a product manager, right, and a coder, right? And so people in each of those roles now know or believe that with AI they don't need the other two roles anymore, right? They can do that because they can have AI do that. And then of course there's the real irony, which is all three of them are going to realize that AI can also be a better manager, right? So they're going to end up aiming the guns up the org chart. But that's probably the next phase.
And what I think is so fascinating about this Mexican standoff is they're actually all kind of correct, I think, right? Which is AI is now actually a really good coder. It's actually now a really good designer, and it's also a really good product manager, right? It's actually good at doing all three of those things, or at least doing a lot of the tasks involved in those three jobs.
And so again, this goes back to this kind of idea of the superpowered individual, where if I'm a coder, I mean, step one is I need to make sure that I really understand AI coding and what that means and what how coding is
going to change in the future. That I need to, specifically, how to go from being a coder who writes code entirely by hand to being a coder who orchestrates a dozen instances of coding bots. There's a change in the actual job of coding itself, which is happening right now. But the other part of it is, okay, how do I become that superpowered individual? How do I become a coder that also then harnesses AI so that I can also be a great product manager and I can also be a great designer, right? And then the same thing for the product manager, which is how do I make sure that I can now use coding tools, how do I make sure I can also do AI-based design. And the same thing for the designer, which is how do I use AI to also become a coder and also become a product manager.
And then what you get is maybe those individual roles change, like maybe those are not anymore sort of stovepipe roles the way that they have been for the last 30 years or whatever. But what happens is the talented people in any of those roles become superpowered and they become good at doing all three of those things. And then those people become incredibly valuable, because then those are people who can actually build and design new products right from scratch, which is the most valuable thing. And so I think that's the opportunity.
So I love this answer. So what I'm hearing is essentially, if you're amazing at any of these three roles, you will do well.
Number one, if you're amazing at these roles, that's great, but also part of being amazing at these roles is also being able to fully harness the new technology, right? So if you're a master coder today and you don't ever get to the point where you figure out how to use AI to leverage your coding skills and do more, right? Like at some point you are going to hit an issue, right?
Here's another way economists talk about this, which is there's the concept of the job, but the job is not actually the atomic unit of what happens in the workplace. The atomic unit of what happens in the workplace is the task. And the way the economists think about it is a job is a bundle of tasks. And everybody wants to talk about job loss, but really what you want to look at is task loss, right?
Tasks changing. The classic example of task changing was, once upon a time, executives never used typewriters or personal computers themselves, right? If you were a vice president of a company in 1970 or whatever, you did not have a typewriter or computer on your desk typing things. You had a secretary who you dictated memos to, right? And then there was this change where emails started to show up. And what would happen was the job of the secretary changed from sending out letters with stamps on them to sending or receiving emails with the other admins. And then the secretary would print out the email and bring it into the executive's office. And the executive would read the email on paper, scrawl the reply, and give that message back to the secretary, who would go back and type it into the computer on his or her desk and send it as an email.
Fast forward to today, none of that happens. Now executives just do all their own email. They still have secretaries or admins, but they're now doing different tasks. They're travel planning and orchestrating events and doing all these other things that the great admins do. And then the task set, ironically, of the executive has expanded to do actually more of the clerical work themselves, actually sit there and type their own memos, which again, 50 years ago they never would have done.
And so the executive job still exists, the secretary job still exists, but the tasks have changed. And I think that's a great example of what's going to happen in coding: the tasks are going to change. Same with product management, the tasks are going to change. Designer tasks are going to change. And so the job persists longer than the individual tasks, and then as the tasks change enough, then that's when the jobs change.
And so at the level of the individual, you kind of want to think of it like, okay, I have this job, the job is a bundle of tasks, I need to be really good at making sure that I can swap the tasks out, right? I can really adapt, use the new technology, get really good at AI coding, for example. And then you want to kind of add skills. I can also get really good at design. I can also get really good at product management because I've got this new tool. So you want to kind of pick up more and more scope as you do that. And then, 10 years from now, is your job title coder, or coder, designer, product manager, or is it just "I build products," or is it just "I tell the AI how to build products"? Whatever that job is called, who even knows what it's going to be, but it's going to be incredibly important, because the people doing that job are going to be orchestrating the AI. And so that's the track that the best people are going to be on. And I think that's the thing to lean hard into.
I think people aren't fully grasping, just specifically, software engineering and how much that is changing. Like it's pretty clear we're going to be in a world soon where engineers are not actually writing code, which I think a year ago we would not have thought. And now it's just clearly where it's heading. It's like there's going to be this artisanal experience of sitting there writing code, which is so crazy, how much that job is going to change.
Yeah. So again, here I go back, and again, pardon maybe the history lesson, but I go back to coding. So you may know, do you know the original definition of the term calculator? Do you know what that referred to?
No.
It referred to people. Right. So back before there were electronic calculators or computers or any of these things, the way that you would actually do computing, the way that you would do calculating, like the way an insurance company would calculate actuarial tables or the military would calculate, I don't know, whatever, troop logistics formulas or whatever it was, the way that you would do it is you would actually have a room full of people. And by the way, in these big rooms you could have hundreds or thousands or tens of thousands of people doing this. You'd have somebody at the head of the room who was responsible for whatever the mathematical equation was, and then they would parcel out the individual mathematical calculations to people sitting at desks who were doing them all by hand, right? And that job title was, those people were calculators, right?
And so we've gone from a world in which you literally have people doing mathematical equations by hand. Then we got the first computers. The first computers, of course, didn't have programming languages, right? They only had machine code, right? So the first computers were programmed with ones and zeros. And so the task of the programmer became do the ones and zeros, and then that became punch cards, and there are still people kicking today whose job as a programmer was to deal with the punch cards. And then you got actually this big breakthrough, which was called assembly language, which was basically the way to do machine code but with some level of English kind of added to it, and then the best programmers did assembly language. And then when I was coming up, it was higher-level languages like C that compiled into machine code, and that's what programmers did.
And then I still remember when scripting languages, we developed JavaScript at Netscape, and then Python took off and Perl and these other scripting languages. Scripting languages took off in the 2000s, and there was this big fight in the technical community, which is: is scripting real programming or not? Right? Because it's kind of cheating, right? Because real programmers write code that compiles to machine code, and real programmers do memory management themselves, and they do this whole craft of writing C code, and these JavaScript or Python programmers are doing this kind of lightweight thing, and does it even really count as coding? And of course the answer is yes, it very much counted, and now most coding is done with the scripting languages, right? You see my point: the scripting languages have abstracted away like five layers of detail underneath that people used to do by hand and they don't anymore. And then, to your point, AI coding is the next layer on that. AI coding actually abstracts away the process of actually writing the scripting code, right?
And so in one sense this is a really big deal for all the obvious reasons, but on the other hand it's like, okay, this is the next layer of the task redefinition under the job of programmer, right? Now what's the job of the programmer? To your point, it's not necessarily to write the code by hand. But if you talk to the world's best programmers today, what they'll tell you is, "Oh, my job is I'm sitting there and I'm orchestrating 10 code bots," right? Coding bots that are running in parallel, right? And literally they sit there and they shift from browser to browser or terminal to terminal, and their day job now is kind of arguing with the AI bots, trying to get them to write the right code, right? And then debug it and fix the problems and change the spec and do all these things. And so now the job of the programmer is to argue with the coding bots, but if you don't know how to write the code yourself, you don't know how to evaluate what the coding bots are giving you, right?
And so, you asked about the 10-year-old. Our 10-year-old is super into computers and super into programming. He's using Claude and ChatGPT and Copilot and all these things. And by the way, he loves coding. He's on Replit all the time doing vibe coding, doing games. It's hysterical, right? Because it's a 10-year-old basically who spends two hours at dinner arguing with an AI for fun, right? But what I'm telling him is, no, look, you need to still fully understand and learn how to write and understand code, because the coding bots are giving you code. If it doesn't work, or if it's not doing what you expect, or it's not fast enough or whatever, you need to be able to understand the results of what the AI is giving you, right? In the same way that somebody who's writing scripting language code does need to understand ultimately how the microprocessor works. And so again, it's kind of this upleveling of capability where you actually want the depth to be able to go down and be able to understand what the thing is actually doing, even if you're not spending your day actually doing that by hand.
And again, I look at that and I'm like, okay, now programmers are going to be 10 times or 100 times or a thousand times more productive than they used to be, right? And that is overwhelmingly a good thing. The tasks are definitely changing. The nature of the job is changing. But are human beings going to be involved in the coding process and overseeing the AI coding and all that? And the answer is of course, absolutely, 100%. No question.
So you're in the camp of still learning to code, still a valuable skill.
Oh yeah, totally. Well, again, if you want to be one of these super... Look, if you just want to put yourself on autopilot, like, "I can't be bothered and I'm just going to have AI write the code and it's going to generate whatever it does and that's fine," if the goal is to be a mediocre coder, then just let the AI do it. It's fine. The AI is going to be perfectly good at generating infinite amounts of mediocre code. No problem. It's all good. If the goal is "I want to be one of the best software people in the world and I want to build new software products and technologies that really matter," then yeah, you 100% want your skill set to go all the way down to assembly and machine code. You want to understand every layer of the stack. You want to deeply understand what's happening at the level of the chip, right, and the network and so forth. By the way, you also really deeply want to understand how the AI itself works, right? Because people who understand how the AI works are clearly able to get more value out of it than somebody who doesn't understand how it works, right? I mean, you're always more productive if you know how the machine works when you use the machine. And so yeah, the superpowered individual on the other end of this that wants to do great things with the new technology, yes, you 100% want to understand this thing all the way down the stack, because you want to be able to understand what it's giving you, right? And when something doesn't work or when something isn't right, you want to be able to really quickly understand why that is.
By the way, again, this goes back to education. AI is your best friend at helping you learn all that, right? Because it's like, oh, I need to understand, I don't know, this isn't fast enough. I need to figure out as a coder how to do a different approach to memory management or something. And you can be like, well, I don't quite know how to do that. Okay, AI, let's spend 10 minutes. Teach me how to do this, right? Teach me what this all means, right? So all of a sudden you have this incredibly synergistic relationship with the AI where it's also helping you get better at the same time as doing a lot of work for you.
By the way, I was going to say, I was a big Perl programmer. I was an engineer for 10 years and that was my language of choice.
I don't know when you were doing it, but do you remember, at least early on, did you ever hit this where coders were looking down their nose at you, being like...
For sure, for sure. It's like, "This is so slow, it's not going to scale, what are you spending all your time on this thing?"
Yeah, exactly. And of course, again, it was sort of this thing where they were sort of correct, which is at the beginning it wasn't fast enough or whatever. By the end, they were definitely wrong, right? Which is it got much better, much faster, and it swept the world. Most coding today happens in scripting languages. And then by the way, along the way, the people who really understood the scripting languages and the people who understood all the lower-level systems, they were the ones who were able to actually make the scripting languages work really well, right? And so that was a great example
of this kind of adaptation. And then again the result of that was a far higher number of people writing code with scripting languages than were ever writing code with lower level languages. And I think this will just kind of be a more dramatic version of that.
I love that Perl was designed by a linguist. I don't know if you remember that part, and that's what made it so nice to code with.
Well, that's funny because of course it was so notorious for being impossible to understand. So how ironic.
Yeah.
This episode is brought to you by Datadog, now home to Eppo, the leading experimentation and feature flagging platform. Product managers at the world's best companies use Datadog, the same platform their engineers rely on every day to connect product insights to product issues like bugs, UX friction, and business impact. It starts with product analytics where PMs can watch replays, review funnels, dive into retention, and explore their growth metrics. Where other tools stop, Datadog goes even further. It helps you actually diagnose the impact of funnel drop-offs and bugs and UX friction. Once you know where to focus, experiments prove what works. I saw this firsthand when I was at Airbnb, where our experimentation platform was critical for analyzing what worked and where things went wrong. And the same team that built experimentation at Airbnb built Eppo. Datadog then lets you go beyond the numbers with session replay. Watch exactly how users interact with heat maps and scroll maps to truly understand their behavior. And all of this is powered by feature flags that are tied to real-time data so that you can roll out safely, target precisely, and learn continuously. Datadog is more than engineering metrics. It's where great product teams learn faster, fix smarter, and ship with confidence. Request a demo at datadoghq.com/lenny. That's datadoghq.com/lenny.
Coming back to this kind of triad, the other element that I hear more and more of is the skill of taste and design and user experience. It feels like that's a very hard skill to learn, and to me tells me design is going to be much more valuable in the future.
Yeah, that's right. And again, this is a great example. So again, the task level of like design the perfect icon, right, is going to be like, all right, the AI's going to do that all day long. It's going to give you a thousand icon designs. It's going to be great. Like it's going to be fantastic. And there will still, by the way, be some level of human icon design or whatever, but AI is going to get really good at that. But like, what are we trying to do? The kind of capital D design of like, all right, what is this thing for? And how is this going to function in a world of human beings? And is this going to make people happy when they use it? Is it going to make people feel good about themselves? Is it going to fit into the rest of their life? Is it going to, I don't know, challenge them in the right way? All these kinds of higher level questions that the great designers have always thought about. The job of designer, right, will involve much more of those higher level, more important components, and then again with AI doing a lot more of the underlying tasks. And so one way to think about it is, you think of, I don't know, the world's best designers, Jony Ive or whatever, you could be like, "Wow, if I'm a designer today, if I'm a 25-year-old designer and I aspire to be Jony Ive in a decade, all of a sudden I have a new path that I can use to get there, because Jony Ive did everything he did without AI." Now a young designer can be like, "Wow, if I really harness AI, in a decade I'm going to be the best designer the world's ever seen, because it's not just going to be me. It's going to be me plus being so super empowered by this technology to be able to do so much more. And then so much more of my time and attention is going to be able to be focused on these higher level things that most designers never get to." And I think that's going to be another great example of that.
So maybe what I'm hearing here is kind of this T-shaped strategy of, if you want to be successful in any of these three roles, be very, very, very good at that specific role, product management, engineering, design, and then get good enough at these other two roles.
Well, so I think that's great. I think that's really, really relevant. And then, you know, Scott Adams unfortunately just passed away, which is a real tragedy, but I referred for years to Scott Adams. He had this famous kind of career advice he would give people, which I think makes a lot of sense, which dovetails with what you're saying. He used to say, look, "I could have been a pretty good cartoonist, or I could have been pretty good at business, but the fact that I was a cartoonist who understood business made me spectacularly great at making Dilbert," right? Because even the world's best cartoonist who didn't understand business could have never written Dilbert. And then the world's best business people who didn't know how to do cartoons couldn't have done Dilbert. It took somebody who actually had both of those skills to be able to make Dilbert, right? Which is one of the most successful cartoons in history, right? And so the way Scott always described it was that, from a career development standpoint, the additive effect of being good at two things is more than double, right? The additive effect of being good at three things is more than triple, right? Because you become a super relevant specialist in the combination of the domains.
And you see this all over the economy. I'll give you an example: Hollywood, just Hollywood as an example. There are a lot of writers who can't direct a movie, and they can be very successful writers. There are a lot of directors who can't write a movie. They can be very successful directors. But the superstars in the entertainment industry are the people who can write and direct, right? And they have a term for those. They call those auteurs, right? And those are the people who are the real creative forces that move the field.
And by the way, Hollywood, actually it's really funny, I've been spending a lot of time talking to Hollywood people about AI. Hollywood has the same Mexican standoff going right now that we described in tech, except in Hollywood, for example for film, it's the director, the writer, and the actor, right? Because the director is now thinking, wow, I don't need the writer anymore because the AI can write the script, and I don't need the actor anymore because I can have AI actors. The writer is saying, I don't need the director because AI can direct the movie and the AI can do the actors. And the actor is saying, I don't need either one of these guys. I can have the AI direct the thing, I can have the AI write the thing, and I'm just going to show up and do my performance, right? And so it's the same kind of triangular configuration. And again, what's great about it is they're all correct, right? Each person in each of those three fields is going to be able to expand laterally and pick up those additional skills. And then as a consequence, you're going to have more people who can write and direct, or write and act, or direct and act, or do all three.
And I think, to your point, your T-shaped thing, I think that's going to be true basically across the entire economy. And if you think about the T configuration, it's like, yeah, the breadth, the top of the T, is how many individual domains are you familiar enough with to be able to use the AI tools to be able to do really good work. And then this part of the T is how deep can you go in at least one of those domains so that you really, really deeply know what you're doing. But if you're super deep on coding and you can use AI to do design and you can use AI to do product management, right, that's your T right there. And you're a triple threat at the top of the T, but with this level of technical grounding underneath that. And at that point, again, you're the superpowered individual. You're going to be able to just perform feats of magic, for example, in terms of designing and building new products, that people in my generation couldn't have even dreamed of. And so I think that this is a universal kind of theory that could apply across the entire economy.
I'm going to invent a new framework right now. Okay, forget the T framework. I'm picturing an F sideways or an E, where there's two or three, I don't know, downward parts. And so what I'm hearing is get good at at least two.
I think that's right. I think that's right. Yeah. The combination. Yeah.
My friend Larry Summers had a different version of the Scott Adams thing, which is he used to tell people, he said the key for career planning is, he said, "Don't be fungible," right? And he's an economist, and so that was economics speak, and what that means essentially is don't be replaceable. Don't be a cog, right? And what that meant was don't just be one thing, right? So if you're, quote unquote, again, just a designer, just a product manager, just a coder, then in theory you can be swapped in or out. But if you have this E or F, laying on its side kind of thing, and if you have this combination of things that's actually quite rare, then all of a sudden you're not fungible. Not only are you not fungible, you're actually massively important, because you're one of the only people in the world who can actually do that combination of things. And yeah, your ability now to become one of those people is just titanically enhanced with AI as compared to anything we've ever seen before.
This is so interesting because I've worked with people that are good at these two skills, and they were always called unicorns at the company. "She can code and design. Oh my god." And what I'm hearing here is this is what you need to become. You need to become really good at at least two things there. I think you used the term smokestack or something, where it's like PM over here, engineer, design. And what I'm hearing here is you need to get good at at least two of these skills. The silos of these two roles are disappearing.
That's right. That's right. And again, I can't overstress the following for anybody listening to this. The thing about AI that I think people are just not getting enough benefit out of yet is it will teach you. Like, this is amazing. There's never been a technology before where you can ask it, teach me how to do this thing, right? It's one of these things where there's so much focus on figuring out how to use a large language model, like, okay, what am I going to try to get it to do for me? Right, which is of course very important, but the other side of it is, what can I get it to teach me how to do, right? And it's just as good at that, right? And so again, this level of latent superpower, people who really want to improve themselves and develop their career should be spending every spare hour, in my view, at this point talking to an AI, being like, "All right, train me up. Superpower me. Train me how to be, you know, I'm a coder, train me how to be a product manager." It will happily do that. It knows exactly how to do that. "Make me problems, make me assignments, then evaluate my results," right? And it will do that just as happily as it will do work, quote unquote, for you.
Two tricks I've heard along those lines. One is to watch the output, what the agent is doing and thinking as it's doing the work. So if you're not an engineer, just sit there and watch it think and make decisions. And it's almost become this layer on top of learning to code: learning to see what the agent is doing and thinking, because that teaches you about architecture. And the other, a couple podcast guests have mentioned this: when you get stuck and then you figure out how to unstick yourself, you ask it, "What could I have done differently? What could I have said that would have avoided this error in the first place?"
Yeah, that's right. That's right. Yeah. Look, on that first one, and again this is what I'm doing with my 10-year-old. Yeah, this is a really good point. So if you ask an AI, write me this code, and it comes back and it doesn't work right, if all you know is a single function, I asked and it gave me back something that's not good, what do you even do with that, right? You don't understand why it gave you that result. Do you even understand what to tell it to try to get it to do something different? But to your point, if you actually watch what it's doing, and then you have the grounding, kind of that leg of your E or your F, if you have that grounding, then you can be like, oh, I see what it's doing, I see where it made the mistake, I see where it went sideways. And then all of a sudden you're able to intervene and able to say, no, no, that's not what I meant, do this other thing, right? And so again, this is a big part of having the actual kind of synergistic relationship, is that you understand. And by the way, look, everything that we're saying right now also is the same as if you're working with human beings, right? If you and I are colleagues and I ask you to do something and you come back with something completely different, I do need to understand what was happening in your head, right, in order to be able to give you feedback, right? If I just tell you, oh, that's wrong, nothing happens. I need to actually understand. I need to have theory of mind, right? I need to understand what you were thinking in order to really give you the right feedback. And again, the great thing with AI is AI will happily sit there and explain all day long why it's doing what it's doing. It'll happily critique itself. By the way, this has a very fun thing where you can have one AI critique the other AI, right? Which is another thing, which is you have one AI write the code, you have another AI debug the code, and so you can actually play the AIs off against each other and get them to argue with each other. And yeah, these are all the kinds of skills that are going to become, I think, incredibly valuable.
I think people call those LLM councils. Yes.
They're talking to each other.
Yeah, that's right. That's right.
I do feel like if I were, like, I have
No design background. I've always wanted to design. I've always wanted to be a great designer. It feels like that's the hardest one to learn of all these three by just watching and talking, right? Because there's a lot of exposure hours, as folks have used this term. Just, how do you learn to be a great designer? That feels like that's going to be really hard and valuable.
So, my true confession is I've always kind of wanted to be a cartoonist, but I have no art skills, but as we're talking, I'm like, it might be time. [laughter]
Their time has come, Marc.
Yes.
I want to pivot to founders, maybe your bread and butter. You spend a lot of time with the most cutting-edge, AI-forward founders. I'm curious what you see them do, how you see them, some way they operate that's maybe blowing your mind about how the future of starting a company looks, how the future of AI-forward companies looks.
Yeah. So, this is a great, very topical topic that's all playing out in real time right now on the leading edge. So, I think there's like three layers of it, and see if this makes sense. I think there's like three layers of it. I think layer one is they're thinking, all right, how does AI redefine the products themselves?
Right. And this is kind of the time-honored kind of thing that happens at technology transitions, and this is kind of what a lot of venture capital is based on, which is, okay, there's a new technology that comes out, and maybe it's the personal computer or the iPhone or the internet, or now it's AI, and it's like, all right, is this a new capability that gets added to existing products, right? So all of a sudden you've got, I don't know, an existing software business, and now you've got your PC version of it, and now you've got your iPhone version of it, and you just kind of keep on going, and the new technology kind of gets added into the mix. It's kind of another ingredient into an existing formula. And of course a lot of new technologies are like that, right? I don't know, when flash storage came out or something, it didn't really redefine the software industry, because people just went from using hard disk to using flash storage or something.
But when the internet came out, basically old-school on-prem software, for the most part, not entirely, but a lot of it died and just got replaced by web software, right? And so sometimes you get the kind of, it's additive to an existing thing. Sometimes you get the, actually it redefines an entire product category, redefines an industry. In many cases the companies themselves turn over.
And so there's sort of this question, and an example, you just mentioned Nano Banana. So a great example is there are these businesses, like just take Adobe. Photoshop has built, whatever, a 40-year franchise in image editing. Okay, is AI sort of a feature now that gets added to Photoshop to be able to do AI-based image editing? Or do you just stop editing images entirely because you're using Nano Banana and all your images are just being generated, and it's just easier to have AI generate a new image than it is to try to edit an old one? So I think there's many areas of tech in which that question is being asked, and the answers I think will vary by domain. But obviously, as a venture firm, we're betting hard on many of these categories being totally reinvented, and a lot of the best founders are trying to figure out how to do that. So that's kind of AI changing the definition of the product.
I think the next layer is actually a lot of what we've already talked about, which is AI changing the jobs. And so it's a lot of what we've already talked about, but like, okay, if I'm a founder of a company and I have room in my budget for 100 coders, how do I get those coders to be super-empowered AI coders, not the kind of coders I used to have? And if they're super-empowered AI coders, then does that mean, do I still need the hundred? Maybe now I only need 10. Or does that mean I still want 100, but now they're doing 10 times more, right? And so, as you know, a lot of the best founders are working on that right now.
And then I think the third shoe to drop hasn't quite dropped yet, but it's kind of the big one, which is like, all right, the basic idea of having a company, right? Does that change? And again, here you've got this concept of the superpowered individual, which is like, okay, can you have entire companies where basically the founder does everything, right? Because what the founder is doing is overseeing an army of AI bots. And there's kind of this holy grail in our industry that's been running for a long time, which is, can you have the one-person billion-dollar outcome? And we've had a few of those over the years. Bitcoin is probably the most spectacular example, with Ethereum right behind it, which wasn't quite one person, but a very small team. You had kind of Instagram and WhatsApp that had very big outcomes with very small teams. Every once in a while you get one of these things where something hits and you just have a very small number of people associated with it. But that said, most software companies obviously end up with huge numbers of employees. And so I think some of the most leading-edge founders are thinking, okay, how do I reconstitute the very definition or idea of having a company? And can you have a company that's literally basically just all AI? And if you're doing anything in the real world, that's hard, but if you're doing software, that seems like it might be feasible in some cases.
And then there's the ultimate example of that, which is, can you have autonomous AI economy stuff happening, where you have AI bots on the blockchain or something that are basically out there functioning as a business and making money, and literally where the AI does all the work itself and just issues me dividends? And so maybe that's the final outlier result. We have a few founders who are chasing that kind of thing. So I would describe that as kind of the ladder that the best founders are on.
Super interesting. This whole idea of a one-person billion-dollar company, I think it depends on your definition of what this is, like an outcome I could see. Having run my newsletter as one person with some contractors, there's so many little annoying things that I have to deal with, with just support tickets and issues and bugs, and it's hard for me to imagine actually a one-person billion-dollar company, even if AI is handling so much of your support, because there's just so many random edge cases that I'm just constantly filling out forms. And so I guess it depends on, do you have contractors? Does that count? What does it mean to be a one person? But I just can't see that happening.
Yeah. I mean, look, Bitcoin's Satoshi pulled it off.
But like, the open source community, does that count? I don't know. I guess it counts. Okay.
Yeah. Exactly. Right. So, yeah. And I would say I don't propose to have answers here, but more just, many of the smartest people I know are thinking hard about this.
Yeah. What do you think about moats? A big question constantly in AI, the fact that everything's changing. Just, what's your guys' thesis on moats in AI? Is that even a thing? Do you care?
My experience with really big technological transformations, and of course I kind of lived this directly with the internet and I saw this happen, is the really big technological transformations take a long time to play out, and there's all of these structural implications that just kind of cascade out over time. And then there's this rush to judgment up front where people kind of say, "Oh, it's therefore obvious that XYZ. It's therefore obvious that this kind of company is going to be the company of the future, not that kind. It's obvious that this incumbent is going to be able to adapt and this other one isn't. It's obvious that there's economic opportunity in this kind of startup and not in these others. It's obvious that the moats are going to be in this area of the technology but not in this other area." And what everybody does is they kind of state those things with just an enormous amount of self-assurance, where they really sound like they have all the answers. And then what happens is these ideas kind of saturate the media, right? Because the media naturally prizes definitive answers over open questions, because, like when CNBC is booking guests, they want a guest who's going to come on and say, yes, this is the way it's going to be, X, not like, I think that's a really good question and let's debate it from eight different angles.
And what I found is if you look back on those predictions a few years later, and you can do this, by the way, if you pull up coverage of the internet from like 1993 through like 1997, or for that matter even through like 2005 or 2010, and you look at the kinds of confident statements people were making in the first 10 or 15 years, I would say almost all of them were wrong, [laughter] and generally quite badly wrong.
And so I think with massive, there's going to be a massive amount of technological change. It's going to be, I don't know, five or six layers of structural change that will play out over time. And again, we've talked about a lot of this, but the implications on what are the definitions of products, what are the definitions of companies, what are the definitions of jobs, what are the definitions of industries, how does this play out at the national level, how does this play out at the global level, how does this, by the way, intersect with politics, how does this intersect with unions, how does this intersect with war, what's China going to do? And so there's just a tremendous number of unknowns, a very, very large number of unknowns, and I think it's just really, really dangerous to prejudge these things.
And so I'll just give, I'll just run this as a thought experiment, see what you think on this, but it's like, are AI models themselves defensible? Is there a moat on AI models? And on the one hand, you'd be like, wow, it certainly seems like there is or should be, because if something takes billions of dollars to build, and you need this incredible critical mass of compute and data, and there's only a certain number of engineers in the world that know how to do this, and they are getting paid like NBA stars, and then these companies have to deal with all these crazy political issues and press issues and reputational stuff and regulatory and legal, all of that translates to, okay, probably at the end of this there's going to be two or three companies that are going to end up with, I don't know, whatever, 50/50 or 40/30/30 or 90/10/1 or whatever it is, market share, and then they're going to have whatever profitability they have, and it's going to be kind of a classic oligopoly. Or maybe one company's going to definitively, it'll be a monopoly. And by the way, those outcomes have happened in software many times before, and so maybe that will be the outcome.
The other side of it is, if you had told me three years ago, in kind of the Christmas of ChatGPT, that within basically a year to a year and a half there would be five other American companies that would have basically exactly capable products, and then there would be another five companies out of China that would have exactly capable products, and then there would additionally be open source that was basically the same, I would have been like, wow, the thing that seemed like it was black magic all of a sudden has become commoditized really fast. Which, by the way, is exactly what happened, right? Within a year of GPT-3 coming out, there were open source GPT-3s running on a fraction of the hardware, right? That were available for free. And then now you've got, fully in the game, Google and Anthropic and xAI and Meta and all these other companies, and then DeepSeek and Kimi and all these other Chinese companies.
And so even at the level of LLMs or AI models, you can squint and make that argument either way.
By the way, same thing at the level of apps, right? One school of thought is apps are not a thing because the model's just going to do everything. But another way of looking at it is, no, actually adapting the model as kind of the engine into a domain involving human beings, where you need to actually have it fit for purpose to be able to function in the medical industry or the legal industry or whatever, or coding, no, you actually need, the application level is actually going to matter enormously, and maybe the LLM's commoditizing, maybe the value goes to the apps. And again, you can kind of squint either way on that one, and I know very smart people who are on both sides of that argument.
And so my honest answer on this is I think we're in a process of discovery over time. The way I think about this structurally is it's a complex adaptive system. The technology itself provides one of the inputs. The legal and regulatory process is another input. Actual individual choices made by entrepreneurs matter a lot. The economics matter a lot. Availability of investor capital varies over time; that matters a lot. And this is a complex system, and so we actually don't know the outcomes on this yet, and we need to basically be open to surprises at the structural level of what happens.
And of course, as a VC, this is very exciting, because it means we should kind of make bets along every one of these strategies and see how this plays out. And I'd just say, there may be one particularly brilliant, I don't know, hedge fund manager or something who has this all figured out, but I guess I would say if they exist, I haven't met them yet.
[laughter] So what I'm hearing here is don't over-obsess with moats at this
point, because we have no idea what it'll end up being. And as much as it may feel like, okay, there's no way OpenAI will lose this lead, clearly we're seeing a lot of competition. The GPT wrapper point is really great. It's such a derogatory term. I don't know, a year ago it was just like, you're just a GPT wrapper. Now it's like, those are the companies that are the biggest companies, fastest growing companies in the world.
Yeah, well, it's a little bit like, I don't know, I mean, even just like with, you know, if three years ago was the holiday of ChatGPT, this last month or whatever has been the holiday of Claude, particularly Claude Code, right, for coding. But it's pretty amazing, because it's like, okay, there was Claude, which is obviously a great accomplishment, but then there's Claude Code, which is an app, right?
It's a Claude wrapper.
Right? It's, you know, an agent harness. And then they did this amazing thing where they came out with, was it Coworker?
Cowork.
Cowork. And remember they said Cowork, which is a... Claude Code wrote Cowork in a week.
Yeah. A week and a half. Yep. 100%.
Well, and there's two ways of looking at that, which is like, wow, obviously that's really impressive that Claude Code was able to build Cowork in a week and a half. That's great. That's amazing. The other way to look at it is Cowork was developed in a week and a half. [laughter] Like, how much complexity could there be? How much of a barrier to entry can there be in something that was developed in a week and a half? And then again, it's this push and this pull thing where it's like, wow, it's incredibly functional, incredibly valuable, and people all over the world every day now are like, wow, I can't believe what I can do with this, it's like the most magical product ever. But at the same time, it took a week and a half, right? And so every other model company, you'd have to expect, is sitting there being like, okay, obviously we need to build an agent harness, and then obviously we need to build a Cowork thing for regular people. I'm not even saying I know anything, but just like, obviously they're all going to do that, right? And so how defensible is that? And in six months, and we've seen this happen before, is Claude Code going to get lapped the same way that GitHub Copilot got lapped? The history in the last three years has been everything that looks like it's the fundamental breakthrough gets basically replicated and lapped very quickly.
Like, many of the smartest people I know in the field, when I really kind of talk to them, get a couple drinks into them, they're like, "Yeah, basically, one theory is there really aren't any secrets among the big labs." Like, the big labs kind of all have the same information and they kind of have all the same knowledge, and they lap each other on a regular basis, but there's not a lot of proprietary anything at this point. And again, evidence of that is DeepSeek came out of left field and basically was a re-implementation of a lot of the ideas of the American big labs, and had some original ideas of its own. But like, wow, it wasn't that hard for basically a hedge fund in China to do it. And so how much defensibility is there?
But on the other side of it, you've got, wow, all these big labs are now paying individual engineers like they're rock stars, and they're incredibly bright and creative people, and maybe there's a dozen nent ideas in any one of these labs that are actually going to be a huge breakthrough that's going to be hard to replicate.
And so again, it's just like, I think we just need to... I don't know, my view is I need to put a big discount on my forecasting ability on this one. For me, it's much less interesting to try to say, okay, as a consequence, industry structure in five years is going to be X, the big winner in the category is going to be company Y, the big product killer app is going to be... This is to say, I don't think I can predict that. I think a much, much better use of my time is being very flexible and adaptable at a time like this.
So with all this in mind, do you feel like there's something you're paying attention to more to help you decide, okay, this is where we want to place our bet? Or is the answer essentially the strategy you guys have, which is place a lot of bets? You guys raised the largest fund in history. Is that the way you win in this world?
Yeah. So for us, we obviously have a very, very deliberate strategy. [laughter] One way to think about this, you remember the Peter Thiel formulation? He said there's a two-by-two: there's optimism and pessimism, and then there's determinate and indeterminate, right? And he always argued that Silicon Valley is characterized by too much of what he calls indeterminate optimism, right? And what he always described he meant by that is basically, I think the way he would describe it is an indeterminate optimist who thinks the world is going to be better but can't explain how, right? Like, some combination of things is going to happen to make the world be better, even if we don't know what those things are. And I think he, at least historically, would say that risks at least being just wishful thinking or delusional thinking, and what the world needs more is determinate optimists, which are people who are like, no, the world is going to be better because I'm going to do this specific thing, right? And he would sort of maybe say VCs are indeterminate optimists, and then he would say, for example, Elon is the determinate optimist, where it's like, no, I'm going to build the electric car, and I'm going to do solar, and then I'm going to do Mars, right? And these very concrete things.
I think there's a lot to Peter's framework, but the way I would describe it is, maybe if he and I disagree with part of that, it would be I think the indeterminate optimism is a stronger phenomenon than at least I think he's historically represented it as. And I would put myself firmly in the indeterminate optimist category, and that's the strategy that we have at a16z. And the reason for that is, hopefully it's not so much wishful thinking. It's more, no, the indeterminate optimism of venture capital, or the indeterminate optimism of a16z or Silicon Valley, is actually very specific, which is: there are these extremely bright and capable people, like Elon and many others, who are founders, right, and product creators, right? And each of those individual people is a determinate optimist. Each of them individually has a very strong view of what they're going to do. But the great virtue of the capitalist system, the great virtue of the American economy, the great virtue of Silicon Valley is we don't just have one of those, and we don't just have 10 of those. We have a hundred and a thousand and then 10,000 of those. And the way to optimize the outcome is to have as many of those as possible be as good as possible. Run as hard as possible.
And then just the nature of the future is like, we just don't know all the answers, and that's okay. And the right way to deal with that is to run as many experiments as possible and have as many smart people try to do as many interesting things as possible. And so yeah, I would put myself firmly on the side of the indeterminate optimist.
I'm wondering if the answer to the question of what you look for now more and more is this determinate optimistic founder. Yeah. That has this massive ambition and is actually working on achieving it.
Yeah. Yeah. No, that's right. That's right. I mean, look, the founders need to be determinate optimists. They need to have a very specific plan. Now, look, the critique always, the critique from the founders is, oh, you VCs have it easy, because you don't actually have to commit, right? You don't have to make the bed you lie in. You can place multiple bets. You can operate a portfolio. You should have a lot more sympathy for us as founders, because we only get to make the one bet. And there's truth to that. You know, the counter-argument on that is the founders get to run their companies. We don't. [laughter] We don't get to put our hand on the steering wheel.
And so the great virtue of being a determinate optimist is you actually get to single-mindedly execute against that goal. And look, in the long run, who does history remember? History remembers Henry Ford, right? Not whoever was the seed investor who seeded Ford Motor Company and 10 other car companies that failed, right? And so the determinate optimist, the founder and the company builder and the engineer, I mean, these are the people who actually do the thing and deserve 99.99999% of the credit. But having said that, I do think there is a role for having some indeterminate optimists in the background helping along the way and helping keep the whole cycle going.
Do you think about AGI in shifting your investment thesis? Like, as we approach AGI and hit AGI, as an investor, how do you think about your investment thesis changing?
Yeah. So I've always kind of struggled with the concept of AGI, because at least, well, there's those defined terms, which is where I kind of struggle with it, which is there's the prosaic definition of AGI and then there's, I don't know, the cosmic definition. And the way I would describe it is, well, let's start with the cosmic one. So the cosmic one is basically it's the singularity, right? And so AGI is the moment where you enter the singularity, which is to say where the world fundamentally changes and the rules of the old world are gone. We're now operating in a new domain. And then the kind of full definition of singularity is it's a world in which human judgment is no longer really relevant, because you get this self-improvement loop. The AI is improving itself and it's sort of racing, so-called takeoff scenarios. You can see this takeoff thing where the AI is improving itself and the machines are making decisions so much faster than people, and people are just sitting there watching the machine do its thing. And I don't really think we live in that world. Whether you could call that utopian or dystopian, I don't think we're lucky or unlucky enough to live in that world.
We could debate that, we can talk about that more. But the prosaic definition of AGI that at least I think the industry participants have kind of converged on, and tell me if you agree with this, is it's when the AI can do every economically relevant task as good as a...
The way the co-founder of Anthropic put it is like a basket of the most valuable economic tasks. So it's like 10, 15, not every single economically valuable task.
Okay. Got it. Yeah. So it's maybe even a slightly reduced definition. And by the way, we're clearly getting close to that, if we're not already there.
And so on that one, I kind of feel like the cosmic one overstates what's going to happen, and then I kind of feel like the kind of AGI definition that you just gave, I think it kind of understates what's going to happen. Like, it's almost too reductionist. And the reason for that is I don't think there's any reason to assume that human skill level is the cap on anything, right? And so the way we say AGI, the definition you gave, the definition I gave, it's always kind of relative, in comparison to a human worker, right? And it's like, I don't know, human skill level caps out at a certain point, but that's because of the inherent biological limitations of the human organism, right? I'll give you an example: human IQ, kind of what they call fluid intelligence, or the sort of g factor of fluid intelligence. IQ, I think, in humans as a species, tops out around 160, right? Where at like 160, it's like Einstein level, Einstein, Feynman IQ,
in terms of IQ. Like, it just tops out at 160. The 160 IQ people are the ones who come up with new physics. There's only a small handful of those. Generally speaking, when we run into somebody in the world who's incredibly smart, who's like a best-selling author, or one of the world's best, I don't know, research scientists, or one of the world's best doctors, or whatever, it would be probably 140 is kind of the IQ that you're looking for there. If you're looking for a really good lawyer, it's probably 130. If you're looking for a really good line manager in a business, it's probably 110. If you're looking for an accountant, like a small business accountant who's good at doing the books for small businesses, it's probably 105, right? And so the kind of scope of the ability of the human organism to do intellectually impressive things, it's sort of that 110 to 160 is kind of the spectrum. And good news is there's a lot of those people running around, but there's not that many at 140, 150, 160. But that's just the limitations of what can fit in here,
right? And it's like, there's no theoretical limit on where this goes if you release the limitations of human biology, right? And so, you already have people running these experiments to do human-equivalent kind of IQ for existing AI models. And by the way, existing AI models right now are kind of testing around the 130, 140 level, which means they're going to get to the 160 level, and they're arguably on the mass high starting to get to the 160 level now. But I think we're going to have AI models relatively quickly that are going to be like 160, 180, 200, 250, 300, by the way.
And I think that's great, right? I feel as great about that as I do about the fact that we occasionally get an Einstein, right? It's like, would the world be better off or worse off with more or fewer Einsteins? And the answer is, of course, the world would be better off with more Einsteins. And of course the world would be better off with machines that have IQ like Einstein or greater than Einstein. But I think the IQ of the machines is going to exceed that of the humans. I think that's really good. And then the performance, again, it goes back to the AI coding thing that is happening. The performance against tasks is going to get better also. Like I
I think, you know, this is where Linus Torvalds in particular is like, yeah, okay, like this thing is starting to generate better code than I can. Okay, so now we're going to have AI coders that are actually better coders than the best human coders. I think that's great. I think we're going to have AI doctors that are better than the best human doctors. I think we're going to have AI lawyers that are better than the best human lawyers, which actually is going to be very interesting to see, which we can talk about, which I think is also great. And so like I don't think there's a... I think we're used to living in a world where we just don't understand how good good can get because we've been capped by our own biology, and we're going to get to experience what it's like when you have the capability at your fingertips that's actually better than human in these domains.
And so, you see what I'm saying, which is like I think this idea of like human equivalent is just going to be like a footnote. It's like, oh yeah, that was just on Tuesday, you know, in 2026 is when they hit that, and it kind of didn't matter because the next question was like, okay, what do we get to do in a world in which we actually have machines that are better than that, right? And so I think this is going to be much more of an exploratory process for actually exceeding human capability than it's going to be any sort of particular singular singularity moment or whatever that just happens to coincide with the human threshold.
200 IQ. I just like that frame of reference is such a mind-expanding way to think about just how fast and how smart these things are going to get, and quickly.
Well, I don't know if you have this experience. I have this experience all the time. Well, two experiences I have all the time. One is just like I know I ought to be able to do this, but like I just can't, like it's going to take too long. You know, I want to write this thing or I want to like whatever. I want to have this theory on this thing or I have a plan or whatever. And it's just like I don't have the eight hours or, by the way, the eight weeks or the eight years, right? And like I just don't know enough yet, and I can't do the math in my head, and my memory isn't perfect, and like I can't remember, and I read, you know, you get interested in something, you read 10 books, and then you're like, I forgot almost everything that I just read. Like I wish I could retain it all, but I can't. It's just like I sort of live in this kind of state of like endless frustration. So it's like if I could just be smarter than I was, like I'd be so much better at what I do, but I'm not. So there's that. And I don't know how often you have this, but I have this on a regular basis. It's just like, you know, because of what we do, like, I know a bunch of people who I know for sure are smarter than I am. And I know it because when I talk to them, I just find myself at a certain point, you know, it's like for the first half of the conversation, I'm just taking notes the entire time. And for the second half of the conversation, I'm just like, like, me. Like this person is just smarter than I am, and they're just outthinking me, and they're going to keep outthinking me, and I just can't, and I'm just like, all right, god damn it, like I gotta go home and I gotta like have a drink because I'm just not, you know, I'm just not whatever that is. I'm not that. And so we're just so used to having those limitations that the idea of having machines that work for us that don't have those limitations, I just think that's much more exciting than people are giving you credit for.
Oh man, I could talk to you for hours, Marc. I'm thinking to close out the conversation, I want to ask about your media diet and your product diet. You just talked about books, reading 10 books. I think you famously read constantly. I saw an interview with you where you're just like, AirPods changed my life, I'm just listening to audiobooks now all the time. So, in terms of media diet, what are you reading? What are you paying attention to these days in terms of, I don't know, podcasts, newsletters, blogs, things like that? And then any books in particular?
Yeah. So, what I read is basically, I would say, basically three categories of things. So like in terms of general media, it's basically, I always describe it as I have like almost a perfect barbell strategy, which is I read X and I read old books, right? So it's basically either like up to the minute, what's happening right now, or it's like a book that was written 50 years ago that has stood the test of time, and then, you know, presumably there's something timeless in it. And then it's sort of everything in the middle I'm always much more skeptical about. And in particular it's kind of what I already said, which is I think if you go back and you read old... nobody ever does this. It's actually really funny. Nobody ever does this. There's no market for it. But if you go back and you read old newspapers, and by the way, you can do this, just read last week's newspaper, right? I guess we're taping on Friday. So read last Friday's newspaper, right? And just go back and read it and be like, "Oh my god, like none of this happened." Like none of what they predicted played out the way that they said that it would. None of this turned out to actually be that relevant or correct. Like they didn't understand, like, you know, by the way, they had no view of what was going to happen this week, that they couldn't know, and so they were making predictions and forecasts and so forth based on not having any information, but it's like, wow, like, you know, none of this happened. Like I wish I had never read this, like, oh my god. And then, you know, it's kind of the same thing with magazines, like go back and read old magazines, and just like the level of, you know, just the endless numbers of predictions that they make. Yeah. And kind of, you know, the problem with, you know, newspapers, at least they're going day-to-day. The thing with magazines is like it's a week or month, you know, kind of long cycle, and so by the time an article even hits publication, it's often out of date. So I just have like a big problem with kind of everything in the middle. And so it's either of the moment or timeless.
But then, yeah, you mentioned like newsletters. I mean, so the other thing, and you know, this is maybe obvious, but I think it's probably still underrated, which is the actual practitioners in the field who are actually creating content, I think probably is still like dramatically underrated. And I think this is a huge part of like the Substack phenomenon and the newsletter phenomenon and the podcast phenomenon, is like direct exposure to the people who are actually principals in the field, who actually know what they're talking about, is probably still dramatically underrated. And I think again the reason for that is like we're used to being in this mass media kind of culture in which basically everything is mediated, right? Everything got filtered through like TV interviews or like newspaper interviews or magazine interviews, and, you know, obviously now more and more it's just, no, you actually want like smart people who are actually working on something explaining themselves, and then you have new kinds of intermediation like podcasts that kind of open that up for people to make that possible. And so, yeah, like domain practitioners are, you know, really great. I mean, just to state the obvious, in AI, you know, it's obviously your stuff, but also like, you know, Lex, you know, the fact that like Lex Fridman can have, you know, the world's leading, or, you know, whoever, any of you guys, you know, there's a small handful of you guys who have access to these people. You can have the world's, you know, kind of leading experts in the domain actually show up.
And by the way, look, the critique always is, you know, people talk their book, like if I'm running a startup or whatever, I'm just selling, and there's always a little bit of that. But it's also, you know, my experience is people love to talk about what they do, and, you know, they fundamentally like want to express what they do, and they want to explain it, and they want people to understand it, and everybody kind of enjoys that, and they get to contribute to kind of human knowledge by doing that, and they get ego gratification by doing that. And so I think there's just actually tremendous amounts of alpha in listening to the world's leading experts in the space who actually just like show up and talk about what they're doing. And of course like the world is awash in that today in a way that it wasn't as recently as 10 years ago. So, yeah, I do as much of that as I can too.
And there's also just this culture in tech, Silicon Valley in particular, of sharing, of not trying to keep these secrets. Everyone on LinkedIn is always like, how is this free? Like it's just the way it works.
Yeah. Somebody said Silicon Valley is a company town, but the company is Silicon Valley.
Right. And again, at the level... this goes again, there's one of these great n equals one... at the level of n equals one as somebody, you know, and I've run startups before, run companies before, at the level of n equals one of like running a company, that's just a giant pain in the butt, because, you know, your secrets are walking out the door and your employees are walking out the door and the whole thing sucks. But, you know, the other side of it is you also benefit from that, right, because you get to hire people with all these skills and experiences, right, and you're in this ecosystem that adapts, right, and channels talent and skill and knowledge and people into the new fields. And so, you know, there's kind of the push and pull of that at the level of just being an individual CEO. At the level of just being in the ecosystem, to your point, like, yeah, it's an absolutely magical phenomenon.
And by the way, like, you know, for all the issues in Silicon Valley, you know, I think AI... I did the comment once, AI is the ninth major technology platform in the history of Silicon Valley, right? That, you know, Silicon Valley is still called Silicon Valley. We haven't made silicon here in decades, right? We used to actually, you know, it's called Silicon Valley because they used to make chips, right? They used to have the actual fabs in Silicon Valley, and they designed them and they made the chips. And so that was, you know, wave one, starting in the 19... actually no, actually more or less like wave three or whatever, but like, you know, that was when the area was named, like in the 1950s. But now we're on like wave nine, right? And the company town phenomenon, where the company is the industry, like, again, the indeterminate optimism: nobody had to sit and plan and say, okay, in the 1990s Silicon Valley is going to do the internet, in the 2000s they're going to do the smartphone, in the 2010s they're going to do the cloud, in the 2020s they're going to do AI. It just, right, the indeterminate optimism of the ecosystem, the flexibility of the ecosystem meant that Silicon Valley could morph into all these categories. And again, maybe a testimony to indeterminate optimism.
This reminds me of the meme of how we're all just wrappers over sand. Everything we're building is just wrapper, wrapper, wrapper, wrapper.
The wrapper thing is hysterical. Yeah. I'm a software company now. I'm a chip wrapper, right? Yeah. I'm a business application. I'm a database wrapper.
Yeah, exactly. I'm a sand... Yeah. You and I are, we're all now sand wrappers. [laughter]
Sand wrappers.
Perfect.
Okay, one more question. Along the media diet, I asked your partner Ben Horowitz, the Z in a16z if people don't know him, what to talk to you about. And he said that you're really into movies these days.
Yeah.
And so I don't know any movies. Any movies you're really into these days? Any movies you've absolutely loved recently?
Yeah. So the movie that blew my socks off last year, which I think is the best movie of the decade for sure, and maybe of the last like 15 years, is this movie. Unfortunately, it's one of these things, not a lot of people have seen it, but I would highly encourage it. It's called Eddington.
Not heard of it.
Have you not heard of it? Okay. So you're going to really enjoy it. So I won't spoil too much of it. So at the surface level, the following spoils nothing, at the surface level, it's set in a small town in New Mexico called Eddington, which is a small town of about 600 people. And there's a sheriff who's played by Joaquin Phoenix, who's like an old crusty, basically right-winger. And then there's a mayor played by Pedro Pascal, who's basically a young hip progressive. And then the movie starts, I think, in March of 2020. And so it starts when COVID first hits, and then as it plays out over the next few months, it intersects and it sort of extends into the summer of 2020. So, you know, kind of the George Floyd moment and then the protests and riots and kind of everything. So sort of the convergence of COVID and then all the BLM stuff. And then there's a third kind of element to it, which is there's a company, which is basically a loosely disguised version of Meta if you read the backstory of it, which is building an AI data center on the outskirts of town. So they kind of pull that in as sort of a thing that looms larger and larger over time.
And then the thing it really is great at is it really shows, you know, this is a small town in New Mexico, and so everybody in the town gets kind of fully wrapped up in all the COVID stuff, and they get fully wrapped up in all the BLM stuff, and they get fully wrapped up in all the like, you know, tech anxiety stuff, but they're all experiencing it basically through the internet, right, which is, you know, what actually happened, right? And so the reason I love the movie so much is, one, it's the first movie that directly grapples with 2020, with what happened in 2020, and it just like fully engages and grapples with all the dynamics that were playing out in the country. But the other reason is it's the first movie that does a really good job of showing what it was like, especially in that era, to live in a world in which there were things happening in the real world and people were kind of experiencing events online, you know, in a way that was very central in their lives, right? And so it does a really good job of pulling in like smartphones and social media in a way that movies really, really, really struggle with. And then the whole thing comes together in an incredibly entertaining way. And so, and I won't even say I completely agree with the movie or whatever, and I think the director of the movie and I would probably disagree about a lot, but he really tries hard to like really grapple with like what it's actually like to live like a human being in the 2020s in America, in a way that I think many other filmmakers who are very talented have just been very scared of touching. And this guy, for some reason, he's just like, "Yeah, I'm just going to find all the third rails and
I'm just gonna like grab them."
I [laughter] can see why that's your favorite movie of the year.
It's great. It's great. It's great. Everybody should see it.
Oh, man. Okay, final question I want to ask about your product diet. Are there any products you use that maybe are less known that you love that you want to recommend? You can, you know, mention products you're investors in if you use them constantly.
I mean, we have so many that it's really hard to, you know, I always feel it's like, you know, who's your favorite children? So, it's really hard to pull out specific ones. But I'll talk about a few. I mean, I'll just give observations.
So, one is my 10-year-old. My 10-year-old right now is 100% obsessed with Replit. And by the way, it was not from me. Do you have kids?
I do. I have one two-and-a-half-year-old.
Two and a half. Okay. So, you haven't run into what I'm running into now, which is whatever it is you do is not cool, right? Like at two and a half, whatever daddy does is like the coolest thing in the world. I can tell you by the time he's 10, whatever you do is like deeply uncool, right? And I'm highly aware of that. And so like if I mention, oh yeah, we work on XYZ, he's like, okay. But when he discovers something, then it's cool. Or when his friends tell him about it, it's cool.
And so he, through no interference on my part, discovered Replit about three months ago and discovered vibe coding and is like completely obsessed with vibe coding games and all kinds of things, and like literally will do it for hours. And so I'm seeing that phenomenon play out, which is super fun. That's one.
Two is I am just completely in love with all the AI voice stuff. I think it's just absolutely amazing, hysterical. My favorite party trick at dinner parties now is to pull out Grok with Bad Rudy, which, if you've seen it, is a foul-mouthed raccoon avatar in the Grok app. So I think that's super fun. We have this company Sesame that went viral last year for these just incredibly, you know, intimate, emotional kind of voice experiences. So I think the voice stuff is fantastic.
I'm also super fascinated by all the voice input stuff. And so, you know, most recently that company recently sold, but I think like the pendants, the wearables, like all that stuff is going to be big. The Meta glasses. I think there's going to be a whole wearables revolution here.
I love the voice input stuff. There's this app on my phone now called Wispr Flow, which is voice transcription, which works like staggeringly well. It's like a voice transcription function, but you can actually talk to the AI model while you're doing voice transcription. So it kind of understands when you're telling it, no, no, I want bullet points over there and I want this and that. And it understands that you're not telling it to type in the words "I want bullet points." It just actually understands that you want bullet points. And so that's a great example of a super useful thing. And so I think the voice mode stuff is going to be really great.
Subscribers of my newsletter get a year free of Replit and Wispr Flow. So there we go. What's the most memorable thing your son built with Replit?
Oh, well, so he's gotten super into Star Trek. And so so far he's like writing Star Trek simulators.
So like all the, you know, by Next Generation they actually had—
Next Generation. Okay. I was going to ask which.
Well, we actually like them all. We watched the new Starfleet Academy last night, which is actually quite good. We watched the original, we watched them all, but it was in Next Generation where they actually developed an actual design language for the computers, because if you watch the original series, they just had basically knobs with lights, and they just were like, you know, around on set trying to pretend they were doing it. But by Next Generation, they actually had a UI design language.
So one of the fun things you can do with vibe coding is you can say, give me a Star Trek: The Next Generation user interface for whatever this or that. And it actually uses the—I'm a nerd now—they call it the LCARS design language, and it'll actually build you like Star Trek: The Next Generation British using that design language, but with your choice of, like, a Star Trek game, for example. And so he's going crazy for that kind of thing.
That sounds extremely delightful. You guys should open source or release that. Marc, like I said, I could talk to you for hours. You've got things to do. Anything you want to leave listeners with before we wrap up? Anything you want to double down on or just leave listeners with?
Yeah, so a couple things. So, one is we got super lucky last week. Packy McCormick wrote the best piece ever written about us, actually, which he released, and so it's the best explanation of what we do and how we think. And so I would definitely recommend that. And then, you know, we have a great team of folks now. We're putting a lot of effort ourselves into video, into content, and so I definitely recommend our YouTube channel, which I think has a lot of great stuff and is going to be very exciting in the next year.
Awesome. We'll link to that. I think it's just youtube.com/a16z, something like that. And you guys have great stuff. Marc, thank you so much for being here.
Awesome. Thank you for having me. I really appreciate it.
Bye, everyone.
Thank you so much for listening. If you found this valuable, you can subscribe to the show on Apple Podcasts, Spotify, or your favorite podcast app. Also, please consider giving us a rating or leaving a review, as that really helps other listeners find the podcast. You can find all past episodes or learn more about the show at lennyspodcast.com. See you in the next episode.
Article published · Updated