a16z · September 23, 2026

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

Amjad Masad says a teenager who raises money for an idea they barely care about is doing premature optimization. Two years spent following curiosity pays off better. The Replit CEO made the case to Erik Torenberg and Gagan Biyani on the a16z channel in September 2026. He backs it with his own path from a chess hobby to a new Replit product.

Speaker
Amjad Masad
Where
a16z
Published
Length
47 min

Amjad Masad: Co-founder and CEO of Replit. Interviewed by Erik Torenberg, with Gagan Biyani.

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

Key points

  1. 1

    Young people are there to question the rules

    Masad opens with the novel The City and the Stars, where a glitch produces the one person who asks what lies outside the city. He says that is the job of the young in any society. Today's college does the opposite and trains students to follow. You can do well in college and still end up jobless with a large debt, so the old bargain is broken.

    “And I think ultimately the utility of the young people is to question deeply held beliefs”
    Amjad Masad, a16z, 03:34
  2. 2

    Starting a startup at 18 is premature optimization

    Masad borrows the programmer's rule that premature optimization is the root of all evil. He sees teenagers raise money out of envy of friends who built unicorns, then commit to an idea they do not care about. He says he tried not to start Replit until it pulled him in. Investors who hand Stanford students a million dollars before they have an idea strike him as "somewhat predatory."

    “I think a lot of starting startups is premature optimization”
    Amjad Masad, a16z, 17:37
  3. 3

    Follow the itch, even into chess

    Masad took up chess late, wrote programs to solve positions, then trained a model because ChatGPT makes up chess moves. Along the way he saw that AI can now do machine learning work, the way agents learned to code around Replit Agent's 2024 launch. The hobby became a Replit product that lets a business owner with data train a model. No plan would have routed him from chess to a new business line.

    “I realized that AI is actually, you know, capable of doing ML.”
    Amjad Masad, a16z, 22:18
  4. 4

    Agents fix what killed holacracy

    Holacracy at Medium and GitHub's no-managers experiment both failed. Masad says the cause was missing technology. Agents can now run the bureaucracy in the background and judge impact, so people can work on what interests them. He has started running Replit as what he calls a self-driving company, and he admits the idea is not fully formed.

    “I think it was a technology problem. Like I think now like you can have agents kind of function as the glue”
    Amjad Masad, a16z, 24:50
  5. 5

    The learn-to-code fight is over

    Every Replit meeting room carries the name of a computing pioneer, to show the company respects the past and still means to change it. In 2025 Masad was attacked for saying people should stop learning to code. A year later, he says, the same engineers tell him they no longer read code. They prompt for it.

    “I was like canceled for for saying you shouldn't learn how to code anymore”
    Amjad Masad, a16z, 28:09
  6. 6

    Go deep enough to teach him something

    Asked what makes an 18-year-old stand out, Masad puts depth first: one project followed all the way through. Most people get bored and fall back on the usual path, so depth signals persistence and drive from within. He pairs it with respect for the other person, since smart young people are quick to decide everyone else is an idiot.

    “I think the first thing is that they dove really deep in something they can teach me something”
    Amjad Masad, a16z, 44:28

How it compares

How this recording lines up with what Amjad said before and with other operators Gavel cites.

Agrees with

In an earlier Y Combinator interview, Masad said liberal arts and generalist skills gain value as AI takes over technical tasks. Here he says the same about school: take the class outside your field and read widely. The new twist is his hiring signal, which rewards going very deep on one project.

Amjad Masad on Y Combinator: The Future of Software Creation

Builds on

Masad's earlier line is that once anyone can build software, the bottleneck moves from building to deciding what to build. Here he moves the same argument one layer down. AI can now do machine learning, so he wants any business owner with data to train their own model on Replit.

How Replit Went From $10M to $100M ARR In Just 9 Months

Breaks from

Gavel's Levels of AI Autonomy page sets Masad against Cursor's Michael Truell, who says vibe coding is not enough for professional work and programming stays valuable. Masad gives no ground here. He says the engineers who attacked him in 2025 now tell him they no longer read code.

Amjad Masad's Levels of AI Autonomy

Related

More talks

All talks