How Amjad Masad Saw Vibe Coding A Decade Early
Y Combinator
Future thinking and early code 0:00
Amjad Masad says AI pushes people through denial, depression, and then acceptance. He describes Replit as an iceberg: the visible part is typing English and getting code back, while the hidden part is deployment and infrastructure. He says the company did not start by calling itself vibe coding. It began by watching what users actually did, and by 2024 many users were already generating 50 to 60 percent of their code with autocomplete.
Watching users finish tasks 2:31
He describes seeing new users copy code from ChatGPT into Replit, then copy errors back into chat, then paste fixes back again. That showed him the real problem was not just an API, but automating the whole loop. The system had to understand errors, keep trying, and stop only when the program worked. He says Replit showed that LLMs can run to completion and can provision infrastructure such as databases and Redis.
Security and guardrails 15:30
Competitors may move fast and build insecure apps, but this team keeps guard rails in place. That can make the product slower or more expensive, yet enterprises like the caution. The visible part looks simple: you tell it in English what to do and it writes code. Underneath, there is deployment, sandboxing, kernel work, and a file system built to be transactional and reversible.
Agents and enterprise 16:30
The old work on collaborative coding carried straight into agent use. The agent could act like another user on the same project, and the company had already split its teams around infrastructure, deployment, glue, and the IDE. Editing was hard in 2024 because models missed line numbers, wrote lazy unfinished code, and needed robust looping. Early users still paid because it was far better than outsourcing.
Future making 30:31
Masad says Replit works because it keeps moving with the future. He thinks companies should not make one big announcement and stop there. They should let customers show the next step, then keep shipping. He points to the shift from one agent to 10, then 100, then 10,000, which cannot run on a MacBook and will push work into the cloud. He compares this to Apple: the iPhone looked sudden from outside, but inside it was built from years of earlier thinking. For him, the same pattern explains vibe coding and the way Replit helps people use AI instead of being crushed by it. A telecom hackathon showed that one hour with Replit changed anxiety into optimism for most people.
AI-generated summary. It can be wrong or incomplete - check anything that matters against the original.

