Skip to content
ClaudeClaude

The Problem Solvers: Michele Catasta at Replit

At 16, Michele Catasta set out to make software open to everyone. What followed was a decade of research at Stanford and Google X in training state-of-the-art models, all in service of that goal. He joined Replit as a pioneer in agentic building: the place where anyone can take an idea and build it in natural language, with no coding experience required. Over 40 million users are doing exactly that, many discovering for the first time that they have permission to build. None of it works without the right model. Replit chose Claude in early 2024 for its coding capabilities, and that decision has evolved into a partnership. Anthropic brings Replit in on early research previews, they build against what's coming, and when a new model ships, Replit ships the same day. The Problem Solvers is a series from Anthropic speaking to founders about how they're solving problems, and why they build with Claude. Discover more from the founders building at the frontier: https://claude.com/problem-solvers

Michele Catastaguest
May 27, 20262mWatch on YouTube ↗

At a glance

WHAT IT’S REALLY ABOUT

Replit’s AI agent democratizes software creation using Claude models

  1. Replit aims to let users with no coding skills describe an idea in natural language and watch it turn into an app or website.
  2. Catasta says model capability is the key differentiator for an AI coding agent, comparing the model to an engine that determines how powerful the overall product can be.
  3. Replit chose Claude—starting with Sonnet 3.5 in early 2024—because it consistently led on coding performance needed for a “high-IQ” agent.
  4. Early access to Sonnet 3.7 in early 2025 enabled Replit Agent v2, and launching it the same day as the model release created a notable marketing surge.
  5. Replit’s value proposition is removing software engineering complexity so more people can go from “zero to one” and build products they previously felt unable or unqualified to create.

IDEAS WORTH REMEMBERING

5 ideas

Model quality sets the ceiling for an AI coding agent.

Catasta frames the foundation model as the “engine,” implying that UI/UX and tooling matter, but the agent’s reasoning and coding capability ultimately constrain what users can successfully build.

Replit’s strategy is to make coding feel invisible.

The product goal is that users don’t experience the act of programming; they experience “creating an idea,” with natural language as the main interface to produce working software.

Claude was selected for “high-IQ” coding performance.

Replit landed on Claude (notably Sonnet 3.5) because they believed Anthropic led the market in coding models, which directly improved agent reliability and outcomes.

Tight coordination with model releases can amplify product launches.

Replit leveraged early access to Sonnet 3.7 to ship Agent v2 immediately upon the model’s public launch, creating a marketing “pop” from simultaneous announcements.

Early model previews are a high-leverage experimentation window.

Catasta describes preview access as a rapid discovery phase—testing what the model can do and quickly productizing new capabilities before competitors catch up.

WORDS WORTH SAVING

5 quotes

Replit is the place where you go with an idea in mind, no coding skills whatsoever, express in natural language what they want to build, and then they see the magic for the first time.

— Michele Catasta

We made coding so accessible to everyone, they don't even feel that they're writing code anymore.

— Michele Catasta

We need high IQ level in our agent. I love to use this analogy where models are like the car engine and we're building the entire car around it ourselves.

— Michele Catasta

It feels like receiving a shiny toy-

— Michele Catasta

That barrier has been completely crushed by Replit, and I'm approached by people that didn't even feel the permission to create software in their life.

— Michele Catasta

Natural-language app buildingReplit Agent and Agent v2Claude Sonnet 3.5 and Sonnet 3.7 adoptionModel-as-engine product philosophyLowering barriers to software engineeringUser growth and product virality“Zero-to-one” creation for non-coders

High quality AI-generated summary created from speaker-labeled transcript.

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.

iOSAndroidClaudeChrome