Masad: Replit uses an autonomous agent to self-improve prompts via nightly PRs
“We have an agent internally at Replit that is constantly evolving Replit agent. What it does every night, it will look at all the traces of everyone that's interacting with Replit, analyzes them, finds, you know, things that are broken, sentiment issues, error…”
Masad: Designing stochastic AI products requires a roguelike gaming mindset
“And so it was a very important mindset when designing AI experiences because AI is stochastic. And so even the language that we use inside the company, like a, like an agent run, it's sort of like a game run, and so on and so forth. So it's a big part of the c…”
Masad: Replit AI agent runtimes expanded from 2 to 200 minutes
“And so what we're seeing is in agent one, the agent can run for two minutes and then perhaps struggle. Agent two came out in February. It ran for 20 minutes. Agent three, 200 minutes. Some users are pushing it to like 12 hours and things like that. I'm less co…”
Masad: Replit Agent v2 operates autonomously for 10 to 15 minutes
“Agent V two, I would call it almost 3.5. It can work up to 10:15 minutes on its own, but it still needs your input every now and then to test the app and make sure the app is working.”
Masad: AI coding agents frequently break existing features when adding new ones
“Today, Repl.agent often creates a feature and then later on Break, breaks that feature, but also true of cloud code and cursor and all the others.”
Masad: Replit launched the market's first end-to-end coding agent in Sept 2024
“And so we launched in September, 20, 24, and it was the first coding agent on the market where you can just, like, put a prompt, and it will, like, make you an application, it will create a database for you, it will, like, Put it out, deploy it, and scale it f…”
Masad: Replit grew from $10M to $100M ARR in 6–7 months
“And our revenue that year climbed to ten million ARR, and then when we got out of beta and the agent got really good, that's when we went to a hundred million ARR in a matter of six, seven months.”
Nayak: Uses Claude and Replit Agent Mode for Rapid Prototyping
“I currently use actually Claude for prototyping because I actually don't want to deal too much with the very nuanced stuff that Windsor for Cursor has. And so I ended up using Claude and then I've actually been using also Replit agent mode. The reason being is…”
Masad: Replit deployed the first AI agent capable of full-stack app creation
“We were the first agent in the market to be able to Configure services, databases, deployment, and go all the way to deploying deploying an app and it kind of transformed our business just in the past seven months.”
Replit Agent was the first AI programming agent on the market
“So we got to September, we launched it in early access, and at the time, there was no other programming agent on the market, so it was really the first.”
Masad predicts Replit Agent will get 10x better in 2025
“And I feel like it's going to get 10 X better, even this year.”
Replit revenue grew 6x to 7x after launching Replit Agent
“Well, now it's more. Six or seven X.”
Masad: LLM reasoning performance degrades rapidly past 32,000 tokens
“What we found is reasoning over long contacts is actually not very good. It's like, not good at all. Like, once you cross 32,000 tokens, that performance of reasoning, you know, just goes down to hell, like, very, very quickly.”
Masad: Replit built its AI Agent as a multiplayer user
“And actually the way we built it is, is like, it's a multiplayer system. So Repl.it is you know, has real time what we call multiplier coding. And we reused this, the multiplayer system to build the agent. So the agent in the code is structured as a, another u…”
Masad: Replit Agent struggles with database migrations during product iteration
“For example, you know, it's not very good at data database migrations. And so we're trying to fix that. So, you know, a lot of when you're iterating on the product, a lot of the times you're actually you know, changing the structure of the app and that, that r…”
Hu: A YC startup built a Replit Agent clone in hours using o1
“They basically got a version of Replit Agent working with the product. All they had to prompt O-one with was all their developer, actually, some of their developer documentation and some of their code, and they could just prompt it, build me a web app that wri…”
Masad: Use short prompts and iterate instead of overcomplicating initial instructions
“I like short prompts because the way I program is I have like an inkling of an idea and I like to do it in iterative fashion. And so it also depends how, how you think. I think it's easy to overcomplicate things at the at the top. And so I like to see somethin…”
Masad: Replit avoided giving its AI a human name to manage expectations
“We didn't name it Autopilot or Automatic Coder or what have you, or a person's name, because we just want to set the expectation that this is this is something that is going to be trying its best to do what you ask it for but not necessarily always succeeding.”
Masad: Replit Agent begins to struggle after building roughly 10 features
“Let's call it like, you know, 10 features in, the system starts to struggle a little bit.”
Masad: Founders using AI agents will eventually need to understand code
“And then you're still gonna, at some point, have to understand the code and have to figure out how to edit it.”