Context
Erik Torenberg talks with Amjad Masad, founder and CEO of Replit, and Gagan Biyani, co-founder and CEO of the Horowitz and Andreessen Academy. The question is what education should look like for people growing up with AI. It matters to PMs because the same ideas apply to how you hire young builders, how you run a team, and how you keep curiosity alive inside a company. Amjad speaks as a founder who hires young talent. Gagan speaks as someone building a new school that has no grades and no tests.
The Big Idea
The most valuable thing young people bring to society is their willingness to question what everyone else takes for granted, so education should give them freedom, trust and real projects instead of grades and hoops.
Amjad's point is that traditional college trains people to follow. AI now lets any student build real things without a custom syllabus, so the missing piece is permission to go deep on what pulls them.
Key Insights
Young people exist to question beliefs
- Claim: Amjad says the utility of young people is "to be heretics", meaning people who question deeply held beliefs.
- Contrast: he says college often does the opposite. It tells students the answers are settled and that they should learn "our way" to get a job, a house and a car.
- Why it is weaker now: he says that path no longer works, since you can do well in college and still end up unemployed with a large debt.
- Blind spots: even Silicon Valley is a monoculture with its own blind spots. Young people with a clear mind are how those blind spots get found. Every so often one of their impossible tries works, because the technology has changed.
Steve Jobs and meandering
Amjad points to Steve Jobs at Reed College. Jobs could not afford it, dropped out, and kept dropping into classes anyway. One was calligraphy. He did not know why it interested him. Later, when the Mac was being built, the proportionally spaced fonts and font choice traced back to that class. Amjad's takeaway is that meandering, which modern society sees as useless, is what makes college valuable. A single class outside your world can change your direction.
Room to build beats attendance
Amjad wants space to work when he is deep in a project. His own college banned students who missed roughly six classes. Gagan had the opposite luck at Berkeley. Attendance was rarely required, so he held three jobs in a semester and planned around his midterms.
- Both note that today's students often work this way, but colleges still do not encourage it.
- Gagan argues schools should bring extra resources to a student whose crazy idea is taking off.
- He cites a student who takes investor calls from a bathroom because he is hiding it from the school.
Learning by building, not by syllabus
- Start from interest: Gagan says he asked his math teacher what the point of the lesson was. If she had tied it to basketball stats and the salary cap, he would have cared.
- Academy approach: if you want to learn Python, do not just take a class. Build something you are excited about in Python and use it to learn the language.
- What changed: with AI, each student can do this alone without a customized syllabus. They mostly need the confidence to hear "yes, you are allowed to do this".
- Amjad's view: project-based learning should be the standard, not a buzzword. He says he is not motivated by pure knowledge. He is motivated by the skills that help him build what he wants.
Startups can be premature optimization
- The false choice: Gagan says teens are offered a four-year school with tests and grades, or a startup with $50,000 tomorrow. He wants a middle path of about two years to explore with no pressure.
- Amjad's frame: in programming, "premature optimization" means tuning code that does not matter yet. He says much of starting a startup is like that.
- The bad motive: he sees kids whose friends are raising money and becoming unicorns. The drive becomes envy, and they commit to an idea they do not care about.
- The better signal: when you find the right idea it has so much gravity you cannot avoid it. He says he tried not to start Replit, and it grew into a startup on its own.
- Investor behavior: both call it somewhat predatory when investors give students large checks before an idea exists. Gagan compares it to a young artist locked into a contract they do not understand.
Curiosity feeds building
Amjad says he oscillates between an itch to build and an itch to read and explore. Each feeds the other. His recent example:
- He got into chess late, and started writing programs to solve it.
- He asked why ChatGPT is bad at chess if models are nearly AGI. That led him to train a model to play chess, which he had never done.
- Along the way he decided AI can now do machine-learning research. He compares it to 2024, when agents first became able to code.
- He is now working on something at Replit so that anyone with data can find a model for a problem. He says no one could have planned those steps in advance.
Following and leading both matter
- Gagan says you need to be disagreeable enough to have interesting ideas, but not so disagreeable that no one can work with you.
- Amjad adds that you should respect what a company has already built, and still feel free to change a lot of it.
- His line for Replit: "we respect the past, but we're unapologetically going to change it". He says he was mocked in 2025 for saying people should not learn to code, and now people who mocked him say they do not read code either.
Trust is the real advantage
Gagan volunteers in local politics and says it taught him how much a foundation of trust is worth.
- Nonprofit: a verbal $50,000 pledge lands in the bank maybe 20 to 50 percent of the time.
- Silicon Valley: the same promise lands about 75 to 100 percent of the time, depending on who is talking.
- Effect: you can act on the promise and make weeks of progress before the check arrives. Without trust, you spend that time anxious.
- For school: trusting students to follow their own threads, even when it looks crazy, is what the Academy wants to offer.
Mental Models & Frameworks
The City and the Stars
Erik recalls a science-fiction story of that name. An AGI governs a city and has solved all human problems. People compress their memories and retrain, so nobody is truly new. A glitch produces one truly new person. He asks what is beyond the city borders, and it turns out to be the one who finds new things. Amjad uses it as a model for young people. It takes a person with a clear mind to question a belief everyone else stopped seeing.
Premature optimization applied to startups
Borrowed from programming, where tuning code that does not matter yet wastes effort. Applied to careers, it means committing to a company or idea before you know what actually drives you. Use it as a test: is your motivation curiosity, or envy of peers who raised money? If it is envy, you are probably optimizing for a path you have not reached.
Discernment comes from reps
- The problem: rules and heuristics depend on context, so students cannot simply be handed the right rule for when to trust elders and when to disrupt.
- Gagan's answer: discernment is built through repetitions. Give students more chances to choose a path, succeed or fail, and learn.
- Where it matters most: where the upside or downside is asymmetric, since that is where you need reps most.
Principles as few axioms
Amjad describes himself as closest to virtue ethics: act in a way you would be proud of whoever is watching. He says the world changes so fast that rigid rules break, and outcome-only reasoning breaks because the future cannot be predicted. So keep a small set of ground truths that you hold dearly and let the rest evolve.
Trade-offs & Nuance
Professionalize early versus explore
- Against early startups: most people benefit from a couple of years to explore without pressure, and only a small number of teens should raise money.
- For early startups: Gagan says some people at 16 or 17 should start a company, and grants are fine if they are clearly temporary.
- The weigh-up: the concern is a ten-year commitment made at 18, not starting early as such.
Freedom versus rules
Gagan does not want a world without limits. The Academy will have rules against things like harassment or being cruel to someone in public, and a culture of trust and positive-sum behavior. His argument is that with no grades or tests there is less weight for students to push against, so many trouble-making impulses lose their target. Asked about someone building something like Facemash, he said he would defend the right to try it, and still say afterward that it was a weird idea.
Following is a skill too
Amjad says students spend all of high school and much of college learning to follow, which is not more useful than listening to their itches. He still agrees that knowing how to follow, work with people and navigate systems matters if you want to lead. The goal is both.
Common Mistakes
Mistake: Pre-committing to an idea
Founders who raise money before they care about an idea lock themselves into a path they have not chosen. They also take a cap table before they have a product. The better route is to follow curiosity until an idea has enough momentum that you feel you have to build it.
Mistake: Punishing side quests
- Colleges that count attendance or expel students for missing class punish the exact students who are busy building.
- Hiding an idea from the school, like taking investor calls in a bathroom, shows the school is seen as an obstacle.
- The fix Gagan describes is to support those students with resources, coaching and demo days.
Mistake: Treating troublemakers as criminals
Amjad describes a teen who hacked a government system and faced roughly two years in prison. That teen said he learned hacking on Roblox. Amjad suggests sandboxes to channel this talent, like bounty programs, and hiring these kids into cybersecurity instead. Gagan adds that some young hackers are doing things for reasons that are fine, and more permissiveness could create more opportunity.
Practical Application
Hire for one deep thread
When Amjad meets an exceptional 16 to 20 year old, the first thing he looks for is that they dove deep into something that can teach him something.
- Ask candidates to tell the story of a project they took all the way through.
- Look for persistence and motivation that had no immediate reward.
- Check the basics of respect and likability too, since strong opinions without interpersonal skills make people avoid working with you.
Give people project time
Gagan's Python example translates to any team. Instead of assigning a course or a sample task, let a new hire learn a tool by building something they care about with it. AI means they do not need a customized exercise, just permission and confidence.
Tie the why to every task
Gagan says a teacher who connected math to basketball would have hooked him. When you hand someone a task, explain what skill it builds or what outcome it serves, since Amjad says pragmatic people are motivated by skills, not pure knowledge.
Check your own motivation
Before starting a company or a big bet, ask whether curiosity or envy is driving you. Amjad's test is whether the idea has such gravity that you cannot avoid it.
Build trust into the process
Amjad's and Gagan's examples suggest making trust the default. Let people follow threads that look strange, set a few clear rules, and let reps build judgment, instead of relying on approvals and grades.
Questions to Consider
- If you were 18 today with AI tools, what would you build to learn your craft, and what would you skip from a traditional course?
- Where on your team do attendance-style controls, like required approvals, punish people who are busy building something useful?
- Are you working on a company or project because of curiosity, or because peers who raised money made you feel behind?
- What would change on your team if a verbal commitment from a colleague carried the same weight as a signed one?
- When you last hired someone young, did you ask what they had taken all the way through, or only what credentials they had?
Bottom Line
Amjad argues the point of education, and of hiring, is to protect young people's habit of questioning assumptions. Give them freedom, real projects and trust, delay the pressure to professionalize, and pair curiosity with the interpersonal skills to work with others.
Concepts to Explore
Self-driving company
Amjad says he has changed how he runs Replit around this idea, though he says his thoughts are not fully formed. Past experiments to change how firms work, like Holacracy at Medium and GitHub, blew up, and he thinks the cause was technology. Agents could act as the glue, handling bureaucracy invisibly while people work on what interests them and AI judges their impact.
Auto research
Amjad also calls it vibe research: AI doing machine-learning research, which he credits Karpathy with naming. He argues it is at the same stage coding agents were in 2024, and is building toward it at Replit.
Case Studies Mentioned
Steve Jobs and the calligraphy class
Jobs dropped out of Reed College because he could not afford it, but kept sitting in on classes. Calligraphy did not seem useful at the time. Years later it shaped proportionally spaced fonts and font selection on the Mac. The lesson Amjad draws is that unplanned learning pays off in ways nobody can predict.
Amjad's chess detour
Amjad picked up chess late, wrote programs to solve it, then trained a language model to play it after noticing that ChatGPT plays badly. That path led him to see that AI can do machine-learning research and to start a new line of work at Replit. The lesson is that following an itch with no plan can lead to a business idea no roadmap would have found.
Tools & Products
| Tool / Product | What it does | Why it was mentioned |
|---|---|---|
| Replit | Online platform for building software with AI agents | Amjad's company, where he is working on letting anyone become a machine-learning researcher. |
| Alpha School | School that builds lessons around student interests | Gagan says he is excited about how it uses natural interests to teach what students need. |
| Roblox | Online game platform | Amjad notes that many young people learn programming, and hacking, there. |
People to Follow
Amjad Masad
Founder and CEO of Replit. He is known for building tools that let people create software with AI and for taking strong contrarian positions.
Gagan Biyani
Co-founder and CEO of the Horowitz and Andreessen Academy, a new school with no grades or tests that focuses on projects and interpersonal skills.
Notable Quotes
"Ultimately, the utility of young people is to question deeply held beliefs, is to be heretics." (Amjad Masad)
"We respect the past, but we're unapologetically going to change it." (Amjad Masad)
"We don't have grades. We don't have tests." (Gagan Biyani)
