The most driven founders are problem solvers. Watch their unscripted conversations with the Anthropic engineers.
 |
VOOZH | about |
|
VOOZH | about |
Replit, an agentic software creation platform, uses Claude on Google Cloud's Agent Platform to help anyone build and deploy software applications without coding experience. Its AI agent, powered by Claude, has grown that audience to more than 50 million people.
The most driven founders are problem solvers. Watch their unscripted conversations with the Anthropic engineers.
The most driven founders are problem solvers. Watch their unscripted conversations with the Anthropic engineers.
The most driven founders are problem solvers. Watch their unscripted conversations with the Anthropic engineers.
AI made writing code faster for developers. The rest of software development stayed roughly where it was. Setting up environments, configuring deployments, debugging across files, managing infrastructure: for someone with no engineering background, every one of those steps still sat behind a wall. The gap between "I have an idea" and "I have a working application" stayed wide.
The pattern Replit's leadership kept seeing was that people with business ideas didn't see themselves as software developers, and didn't try to learn. "I'm approached by people who didn't even feel the permission to create software in their life," said Michele Catasta, President and Head of AI at Replit. "They always had an amazing business idea. Now they're doing that 0 to 1, complete in Replit. This is changing people's lives."
Closing the gap meant building an AI agent that could own an entire development cycle: designing schemas, writing frontend and backend code, testing the result, and fixing its own bugs while a non-technical user described what they wanted in plain English. That required a model with the reasoning depth to hold a complex project in working memory for hours at a time.
Replit landed on Claude as the underlying model for its agent in 2024 and has stayed with it across multiple generations of Anthropic models. Agent 4 runs on Google Cloud's Agent Platform, with Sonnet 4.6 handling sustained development work and Opus 4.7 for sophisticated architectural decisions and complex multi-file refactoring.
"We need a high IQ level in our agent," Catasta said. "Models are like the car engine, and we're building the entire car around it ourselves. To remove the barrier of entry to coding, the power of the model is what makes the difference. We made the choice back when Sonnet 3.5 came out, and since then Anthropic has had the best coding models on the market."
With Claude as the engine, Agent 4 manages the full development lifecycle. The agent breaks requirements into tasks, designs database schemas and APIs, implements features across the frontend and backend, integrates third-party services, and tests the result. Sessions run for 6+ hours without human input, a 10x improvement over previous agent capabilities. Application development that used to take days now completes in under an hour.
Claude's vision capabilities power Agent 4's self-testing system. Rather than relying on unit tests alone, the agent watches the application run in a browser, clicks through interfaces, validates form submissions, tests API endpoints, and identifies the root cause of any bugs it finds before fixing them. The approach runs 3x faster and 10x more cost-effectively than computer-use models on the same workflow.
Behind every Agent 4 session is the broader Agent Platform stack: instant deployment to Cloud Run, Cloud SQL databases provisioned automatically, BigQuery integration for analytics, and global delivery through Google's CDN. The platform's security infrastructure lets Agent 4 handle sensitive codebases and credentials for enterprise customers.
The pace of new Anthropic models has shaped Replit's product cadence. Each new release becomes the foundation for a new version of the agent, often shipped the same day.
"It feels like receiving a shiny toy at Christmas," Catasta said of early research previews. "The discovery process of finding out what it's capable of is the most exciting window of time we have before grinding to launch the next version of the agent. The moment Anthropic launched the new model, we launched our new agent the same day. There was an incredible marketing pop."
For Catasta, what was driving the pop stood out from anything Replit had launched before. "Users experience something that I've never seen before in our product," he said.
Replit's natural-language platform now serves more than 40 million users, a population the previous generation of coding tools never reached. Annual recurring revenue has grown from $240M as the AI-native platform has scaled. "We remove all the complexity that makes the job of a software engineer really hard," Catasta said. "That barrier has been completely crushed by Replit."
The user base reaches well beyond traditional developer profiles. Marketing teams build custom analytics dashboards. Sales operations teams create CRM integrations. Some users go further and use Agent 4 to build other agents, describing a workflow in plain English and watching it produce a Slack bot that queries GitHub or a daily digest pulling tasks from Linear.
Teams from Zillow to Duolingo deploy Agent 4 to ship internal tools. "What previously required weeks of developer time now happens in a single afternoon," said Amjad Masad, CEO and co-founder of Replit. "AllFly rebuilt their entire application in days."
The reach is global. One developer in India built their first applications entirely from a phone using Agent 4, earned enough to buy a laptop, and now builds enterprise systems on the platform. Another power user generated more than 36,000 lines of production-ready code in a single Agent 4 session, building a business operating system that integrated eight admin systems with eight client portal systems over roughly 400 minutes of autonomous runtime.
For many of these users, opening Replit is the first time they've built software. "They see the magic for the first time," Catasta said. "They don't even feel that they're writing code anymore. They're just creating an idea they have in their mind and turning it into an application, or a website."
For Catasta, the work is personal. "When I was 16, I was focusing on becoming a good open-source developer," he said. "I love the fact that I'm still working on something I was passionate about as a teenager, and I think this is going to be my life's mission."
