Replit CEO Amjad Masad on What Young People Should Learn in the AI Era

Summary

• Amjad Masad’s core thesis is that the social function of young people is heresy — “the utility of young people is to question deeply held beliefs” — and today’s college does the opposite: “it kind of beats any outside thought out of them.” The old bargain is dead: “You can do really well in college and you can still be unemployed with a huge debt,” and the civilizational cost is losing the clear-minded outsider who asks what’s beyond the city walls, per his “City and the Stars” parable.
• The most thesis-relevant disclosure: Masad says AI is now capable of doing machine learning itself — “RSI” / “cursive self-improvement,” “vibe research,” or “autoresearch” — and Replit is building a product on it. A late-life chess hobby led him to discover LLMs “actually suck at chess,” to train his first model, and to conclude ML is at the same inflection coding hit when Replit Agent came out in 2024, before “vibe coding” was coined. He’s now working at Replit so “any business owner who has data can fine-tune a model.”
• Masad calls current teen-founder funding predatory: VCs are handing Stanford students “a million dollars, not even 50K…” pre-idea, locking kids into cap tables like music-industry contracts. Kanye’s critique of record labels is his analogy; SAFEs help but “it’s still somewhat predatory,” and a lot of startup formation is “premature optimization… the root of all evil” — envy-driven pre-commitment to ideas founders don’t care about. His counter-signal: the right idea “kind of exploded into a startup on its own” — he says he actively tried not to start Replit.
• Gagan Biyani’s Academy is positioned as the missing middle between a rigid four-year degree and a $50K teen startup check: two years of exploration with “no grades, no tests,” where Silicon Valley brings resources to students rather than forcing side quests underground. He wants the investor track “way underutilized” for teenagers — grants yes, ten-year commitments at 18 no — and argues a lower-stakes structure sidesteps the troublemaker problem: “You wanna build Facemash, go for it.”
• Masad predicts the nature of the firm changes next: Holacracy and GitHub’s no-managers experiments “blew up in spectacular ways” because of a technology problem AI may now solve. Agents become the invisible bureaucratic glue, “AI can sort of judge your impact,” and he’s already running Replit toward a “self-driving company” that “will run itself.” Track record note: he was “canceled” in 2025 for saying people shouldn’t learn to code — a year later, he says, his critics say, “I don’t read code anymore.”
• Biyani quantifies Silicon Valley’s trust premium from his Oakland politics side quest: a verbal $50K nonprofit pledge lands somewhere between 20% and 50% of the time; in Silicon Valley, he estimates 75% to 100%, depending on who is involved — and that gap compounds because you can execute six weeks before the check clears. His transfer thesis: apply that trust to students and “they’re gonna figure out the next version of Holacracy that actually works.”
• On values, Masad lands closest to Aristotelian virtue ethics: principles should evolve as the world changes, but remain grounded in a small set of core truths. He argues consequentialism is especially difficult as “we’re entering the singularity” and “you can’t really predict the future.” The “autistic genius” role may once have paid off, but “Silicon Valley is no longer this renegade thing… it’s part of the establishment”; graduates must eventually “be in front of Congress,” not just speak in-group language. He criticizes Dwarkesh’s post, which Bernie Sanders read, for failing in his view to create more understanding among outsiders.

Deep dive

1. College should manufacture heretics; instead it deletes them

• Masad opens with “The City and the Stars”: an AGI-governed utopia where people compress their memories and retrain for life, until a glitch produces a genuinely new person who asks, “What’s out there? What’s in the stars?” His point: it took a glitch to get a clear mind capable of questioning deeply held beliefs — and college today does the opposite. “It tells you you’re stupid. We’ve figured everything out… get a job and be a perfect citizen.” But “that world doesn’t exist anymore.”
• Even Silicon Valley gets indicted: “every culture is somehow a monoculture, even the most radical cultures like Silicon Valley.” The job of the young is to try impossible things, because occasionally technology has changed and the impossible quietly became possible.

2. The ideal school: non-prescriptive, attendance-free, staffed with unconventional people

• Masad’s redesign of 18–22 has three planks: freedom to meander (Jobs dropping into Reed’s calligraphy class, which resurfaced years later as the Mac’s fonts — “connecting things backwards”); space to build when obsession strikes (his own college banned you after six missed classes); and people quality — “if you’re around conventionally minded people, eventually you’ll be conventionally minded.”
• Biyani’s Berkeley version: many semesters with three jobs, often working instead of attending class, then taking time off for midterms. What was odd behavior 20 years ago is now becoming the student norm — “and the thing that’s weird is that it’s not encouraged yet. It’s still not allowed.”
• All three converge on the same inversion: colleges should bring resources to students with traction, not force side quests underground — Torenberg cites a kid taking investor calls from the school bathroom who should be getting a demo day or other support instead.

3. Project-based learning becomes practical with AI

• Masad’s contrarian history claim: learning things without knowing their utility “is a very modern invention, a very weird thing to have done in society.” He’s explicit that he’s not motivated by pure knowledge: “I’m motivated by skills — what skills can it give me in order to achieve what I wanna build?” Torenberg’s basketball version: a stats class taught through the salary cap and player analytics would have actually landed — the Alpha School model.
• Biyani says educators always agreed on customized curricula but it was “practically impossible”; with AI, “it’s actually impossible to imagine why you wouldn’t do that.” Applying this to the Academy, he says coding students should build something they’re excited about in Python rather than merely take a Python class, and every student can now do that without a bespoke syllabus.

4. The $1M pre-idea check is the music industry all over again

• Masad’s warning: much of teen entrepreneurship is “premature optimization… the root of all evil” — kids motivated by envy of unicorn friends, pre-committing to ideas they don’t care about. He’s heard VCs give Stanford students “a million dollars, not even 50K… pre-idea,” so that “the moment you start your idea you already have a cap table.” His analogy is Kanye on record contracts; SAFEs, credited to YC and Paul Graham, help, “but it’s still somewhat predatory.”
• The counter-signal to look for is gravity. “I tried not to start Replit, like actively not to start it, but it kind of exploded into a startup on its own… it’s almost like, I guess I have to do it.”
• Biyani’s structural fix: only a small number of teenagers should be raising money; give the rest two pressure-free years where “Silicon Valley will come to you,” with grants explicitly framed as temporary.

5. From chess hobby to “vibe research” to the self-driving company

• Masad’s best example of curiosity compounding: he got into chess late, discovered that LLMs such as ChatGPT “actually suck at chess” despite rising ARC-AGI scores, trained his first LLM to play, and realized AI can now do machine learning. He calls this RSI or “cursive self-improvement,” and “vibe research” or “autoresearch,” which he thinks is Karpathy’s term. He sees machine learning at a similar inflection to coding when Replit Agent came out in 2024, and he’s building a Replit product so anyone with data can fine-tune models. “When I started playing chess, what are the hops you need to make to understand that’s gonna lead me to build a new business line? It’s just impossible.” Even meme subcultures pay: “I learn something about markets… and I have alpha.”
• His firm-structure corollary: Holacracy at Medium and GitHub’s no-managers era “all blew up in spectacular ways,” but “it was a technology problem.” Agents could become invisible bureaucracy, AI could sort of judge impact, and humans could gain true autonomy — he’s already changed how he runs Replit toward a “self-driving company.” Biyani, hedged but certain about the direction: “I genuinely don’t have an opinion about where it’s gonna go, but I know it’s gonna change… a lot faster than you think.”
• Receipts on being early: Masad was “canceled” in 2025 for saying people shouldn’t learn to code; he says some of the same people who opposed him now say, “I don’t read code anymore, I just prompt.”

6. Trust economics, redeemed troublemakers, and virtue ethics for the singularity

• Biyani’s number from Oakland politics: after a verbal nonprofit donation commitment, the money lands somewhere between 20% and 50% of the time; in Silicon Valley, he estimates 75% to 100%, depending on who is involved. Trust compounds, since you can make six weeks of progress before the check arrives. The Academy imports this by trusting students, reducing institutional weight, and making it easier to let them try things like Facemash. Masad’s addendum: kids learning to hack on Roblox — he has discussed a talent sandbox with Roblox founder and CEO Dave — could be funneled into cybersecurity jobs or even government work rather than sent to jail. Biyani grew up using hackthesite.org and would point students toward school and corporate bounty programs.
• Masad extends the discussion into discernment: knowing when to respect elders and when to disrupt comes from reps, trusting students to fail and learn, and understanding asymmetric upside versus downside. He coins “meta retard maxing” for knowing when to follow instinct and when it will drive you into trouble.
• Biyani also frames the Academy as human cultivation, not just technical training: interpersonal skills are an “absolute pillar,” and the goal is to produce positive influences in a positive-sum, high-trust culture where people grow as adults over two years.
• Masad’s moral philosophy, arrived at by reading widely and rejecting utilitarianism: virtue ethics — act so you’d be proud “regardless of who’s watching… regardless of the impact,” while allowing principles to evolve and keeping core axioms “as small as possible.” The kicker: “you can’t be a consequentialist or utilitarian when we’re entering the singularity,” because that requires predicting the future. The “autistic genius” role may once have paid off, but he says the Valley is now “part of the establishment”; he cites the deleted Sequoia SBF post and criticizes Dwarkesh’s blog post, which Bernie Sanders read, for failing in his view to create more understanding among outsiders.
• What impresses Masad in an 18-year-old, in closing: “they dove really deep in something, they can teach me something” — depth signals persistence and intrinsic motivation — paired with respect and likability, because young-and-smart people can easily tip into thinking everyone is an idiot. “The most endearing thing is you can nerd out on something obscure and interesting.”