Evaluating and improving Replit Agent at scale
- Featured
- Agentic AI
Date
Tue 19 May
Time
12:30 - 12:45 CEST
Location
Tech Stage
About this session
Building AI agents that work reliably at scale is one of the hardest challenges in applied AI. In this talk, Michele explores how Replit approaches evaluation and continuous improvement of Replit Agent: the AI-powered coding assistant used by millions of builders worldwide. From designing meaningful evals to closing the feedback loop between real-world usage and model improvements, Michele shares the frameworks, lessons learned, and open questions from operating an AI agent at production scale.
