How Replit Is Enabling a New Wave of Million-Dollar Founders | Amjad Masad & Haya Odeh

Replit's twenty-year bet that anyone can build software

Replit spent twenty years turning an open-source side project into a platform where people who are not engineers build real software and earn money from it. Masad's conviction is that coding changed his own life and should be open to everyone, and AI agents finally make that possible. The company competes on interface and context rather than model building, so he expects the next wave of agents to move past the chat box.

Takeaways

• Replit users now build million-dollar businesses. Replit has always produced success stories, from a fitness instructor earning $1,000 in the App Store to a solo entrepreneur who tweeted about $8m. A physical therapist built a 3D body-scanning app after offshore developers made a mess, and a Zillow product manager created millions of dollars of value.
• Prompting has reached its ceiling. Masad thinks the industry has hit the apex of what prompting can do and expects new interfaces and fully autonomous agents. The big labs leave user experience white space, since Claude and Claude Code remain prompt-based. Replit's Canvas and Agent 4 keep a project's whole context in one place.
• Masad's 2023 predictions have all landed. In a 2023 TED AI talk he said software would be solved within a couple of years and sketched vibe coding and parallel agents for the next two. Replit has completed that roadmap. His method is to execute day to day while occasionally predicting the future.
• Individuals and small businesses drive revenue. Revenue is still dominated by individual entrepreneurs and small businesses rather than enterprises. Masad says Replit thinks in traits rather than personas. Odeh argues the same behaviours run from consumers to enterprises, driven by wanting to be heard and by stubbornness.
• Seek pain is Replit's defining value. Odeh created the value, tracing it to army discipline. When people at Replit celebrate shipping something and say it works, she asks what is not working and what sucks about it. The company runs on meritocracy rather than nepotism, and Masad once fired and rehired his brother.

Masad's model roadmap

1. Pre-training drove progress through 2023, and then improvements plateaued.
2. An RL and thinking era followed in 2024 and 2025.
3. The November and December unlock was near-indefinite coherence, as models trained on their own context windows, learned compression and searched across files they stored.
4. Bigger models are coming, since the labs appear to have figured out how to scale pre-training again, and token prices are not falling because of it.

Where Replit finds elite talent