In March 2025, Amjad Masad told TBPN that “learning to code is a waste of time,” and X piled on. Eighteen months later, a lot of the people who dragged him are prompting instead of typing.
Now he’s going after college. Andreessen Horowitz launched the Horowitz Andreessen Academy last month as a tuition-free alternative to college, and Masad sat down with its co-founder Gagan Biyani and a16z’s Erik Torenberg on the a16z podcast to talk about what school should look like for kids growing up with AI.
He argued that school beats curiosity out of kids and that starting a company too young is “premature optimization.” He also thinks the shift that hit coding is coming for machine learning next.
His core belief is that the job of young people is “to be heretics,” and college does the opposite.
I went through the full conversation and pulled out the 10 ideas worth your time.
If Masad is right that prompting beats typing, the skill to build this week is prompting Astra well. Astra now runs your computer for you and builds apps or slides from a single prompt.
1. Why College Manufactures Conformists (And Why That’s the Point)
School was never built to produce heretics. Masad thinks that’s exactly the problem.
“The utility of young people is to question deeply held beliefs, to be heretics. And in today’s college, it’s actually the opposite. It kind of beats any outside thought out of them.”
College tells you the path is settled. Grades, degree, job, house.
Masad’s point: that path stopped paying off. You can follow every instruction and still end up unemployed with debt.
An institution built to produce compliant graduates keeps producing them, even after the job market it was built for disappears.
“You can do really well in college and you can still be unemployed with a huge debt.”
Every founder reading this broke out by ignoring an instruction, not following one better.
2. The Missing Middle Between College and a Term Sheet (Nobody Is Building For It)
Masad sees a binary trapping ambitious teenagers on both sides of it.
“There’s this huge unfair choice young people are given today. It’s either go to a 4 year school, homework and tests and grades and all of that, or start a startup and someone will give you $50,000 tomorrow.”
Both paths force commitment before a teenager knows what they’re committing to. College locks you into a curriculum. A pre-seed check locks you into a cap table before they’ve even shipped anything.
What Masad wants instead: 2 years of low-pressure exploration, funded by grants instead of equity. No expectation that exploring becomes a company.
That option barely exists. Most founders either grind 4 years of a system they didn’t choose, or skip straight to fundraising before they’ve found the idea worth funding.
3. Premature Optimization Is Killing Founders Before Year One (The Programming Bug Behind It)
He borrows a line from software engineering to explain why so many young founders burn out on ideas they never wanted.
“A lot of starting startups is premature optimization. And I see it in a lot of kids now where they’re part of a group and a lot of their friends are raising a lot of money, and it just becomes the motivation is envy.”
Premature optimization, in code, means tuning a function before you know if it matters. Applied to a career: locking into a startup before you know if the idea has any actual pull on you.
He sees it in kids chasing the same unicorn headlines their friends are getting, then scrambling to raise money for an idea they never actually wanted.
Masad’s own story ran the opposite way. He tried not to start Replit. It exploded into a company anyway, because the idea had enough gravity that avoiding it stopped being an option.
That’s the signal. If you have to talk yourself into an idea, it’s premature. If the idea won’t leave you alone, that’s the one to build.
4. The $50K Check That Locks Founders In Before They Have an Idea
He draws a straight line from VC term sheets to the way the music industry signs teenage artists.
“VCs go to Stanford students and give them, like, $1 million, not even $50K. $50K is like $1 million pre-idea. And it’s like the moment you start your idea, you already have a cap table. I think that’s sort of like the music industry.”
The comparison works because the mechanism is the same. Money shows up before the person has anything to protect.
Whoever’s offering the check sets the terms. The founder has no leverage yet.
Masad credits YC and Paul Graham‘s SAFE with softening this. He still calls pre-idea funding “somewhat predatory,” because the commitment happens before the founder can evaluate it.
If you’re an investor writing pre-idea checks, the founder on the other side usually can’t tell a fair term from a bad one. That asymmetry is the whole deal.
5. The Kid Taking Investor Calls From a Bathroom Stall (And Why the School Should Be Proud)
This is the image Masad keeps returning to.
“There’s this kid, Eric, who has to take investor calls from the bathroom because he’s hiding in the school. They should be, you know, coaching him through it, enabling it in some way.”
A student building something serious inside a school that won’t acknowledge it isn’t a discipline problem. It’s a resourcing failure.
The computer lab, the compute, the bio lab. All of it sits unused by the one student who could put it to work.
Masad’s version of college flips the incentive. Early traction gets you more resources, not more scrutiny. A kid with a live deal should get a demo day, not a disciplinary meeting.
He says this behavior used to be rare among students and is becoming the norm. What hasn’t caught up is permission.
The gap between “not allowed” and “actively encouraged” costs a young builder a lot of value.
6. He Got Canceled in 2025 for Killing Code. He Was Right a Year Later.
His most public prediction was also his most mocked, at least for a while.
“I don’t know if you remember, like 2025, I was canceled for saying you shouldn’t learn how to code anymore. And now the same people that were pointing against me, they’re like, oh, I don’t read code anymore. I just prompt the code.”
The reversal happened fast. A claim that sounded reckless in 2025 was consensus a year later, without Masad changing his position at all.
“It’s amazing how many things I’ve been wrong about, or right about, 5 or 10 years early.”
The lesson isn’t “trust every contrarian take.” It’s that backlash tells you nothing about correctness.
7. Coding Wasn’t the Last Shift. Machine Learning Is Next.
Masad thinks the collapse that hit coding is about to hit ML research.
“We’re at a similar moment when it comes to machine learning. I’m working on something at Replit to bring that to the world, where anyone can become a machine learning researcher, or any business owner who has data can find your own model to solve a problem.”
His evidence is personal, not theoretical. He got into model training to play chess with no ML background, found current models are surprisingly bad at chess, and ended up deep in model training within months.
Same pattern as Replit Agent before “vibe coding” had a name. The tooling gets good enough that the skill floor drops. The label shows up after the fact, once everyone’s already doing it.
If that repeats, “who can train a useful model“ stops meaning ML researchers and starts meaning anyone with a dataset and a question.
“You can have agents kind of function as the glue within these organizations, and have the bureaucracy be this invisible thing that’s happening by machines in the background. We call this idea a self-driving company. Eventually the company will run itself.”
Medium tried Holacracy. GitHub tried no managers. Both blew up.
Masad’s read: the failure was technological, not philosophical. Coordination without hierarchy needs something to actually coordinate.
He’s testing it inside his own company now, agents handling the bureaucratic layer so people can work on what actually interests them.
He’s upfront that he doesn’t have this fully figured out. Rare, for someone claiming to have solved the future of work.
9. Why Every Student Can Now Get a Personalized Curriculum (Without a Custom Syllabus)
Education has wanted this for decades. Masad argues the excuse just expired.
“If you are going to learn coding, I don’t want you just to take a class on Python. Go build something you’re super excited about in Python, use it as a tool to learn the language. And the great thing about today with AI is every single student can go and do that on their own.”
The old objection was logistical: no teacher can hand-build a syllabus for every student in a class. AI removes the constraint.
A student doesn’t need a custom curriculum if they can prompt their way through whatever they’re building and pick up the skill along the way.
Masad calls the underlying idea project-based learning, then immediately rejects the label. It shouldn’t need a buzzword. It should be the default way anyone learns anything.
The barrier was never pedagogical theory. It was the cost of building one syllabus per student.
10. Stop Sending Teenage Hackers to Prison. Send Them to the NSA.
Masad starts from a specific case: a teenager facing 2 years in prison for hacking a government system, after learning to hack on Roblox.
“It’d be really cool to have some kind of sandbox for these people to play in and then discover talent and then funnel them into cybersecurity jobs. We need cybersecurity talent right now. Why are you putting them in jail?”
The instinct to test boundaries and the skill to break into systems are the same raw material behind both criminals and elite defenders. Prison burns the talent. A sandbox with mentorship redirects it.
Masad’s own on-ramp was a site called Hack the Site, a sandbox with vulnerabilities built in on purpose, the kind of thing he now wants schools running deliberately instead of leaving to chance.
Not leniency for its own sake. A labor-market argument: the country needs cybersecurity talent, and it’s currently incarcerating a pipeline of it.
The gap between “kid who broke something” and “kid who could defend something” is usually a mentor, not a prosecutor.
The Heretic’s Playbook
Masad’s thesis in one line: the systems built to prepare young people for the old world are actively disqualifying them from the new one.
College optimizes for compliance. Early VC money optimizes for lock-in. Neither optimizes for what actually produces outliers: unsupervised curiosity with genuine resources behind it.
▫️ Founders: Don’t let envy set your timeline. Pre-committing to an idea because your friends are raising money is premature optimization. Wait for the idea with enough gravity that avoiding it isn’t an option. ▫️ Investors: Pre-idea checks carry more leverage than they look like. If you’re writing one, be honest about whether the terms survive the founder actually understanding them in a year. ▫️ Operators: The self-driving company thesis is testable in your org today. Find one coordination task a human relays and hand it to an agent instead. See what breaks. ▫️ Everyone else: If you’re mentoring a teenager testing boundaries, the choice isn’t punish or ignore. Build the sandbox. That’s the third option.
The 4 Principles to Steal
Wait for gravity, not envy. An idea worth building keeps pulling at you without external pressure. If it needs a friend’s funding round to feel urgent, it’s not ready.
Fund exploration, not commitment. A grant with a stated expiration does what a term sheet can’t: 2 years of runway without locking someone into an outcome they haven’t chosen.
Replace bureaucracy with agents, not managers. The failed flat-org experiments needed a coordination layer that didn’t exist yet. It exists now.
Redirect boundary-testing instead of prosecuting it. The instinct behind a security breach and the instinct behind a security career are the same one, depending entirely on who builds the sandbox first.
Every one of these argues against the default setting, not for a specific alternative.
If this breakdown saved you from a bad five-year bet, send it to one founder or investor who needs it.
Who told you he was right? Himself? Whatever people will say to get attention now.