Amjad Masad Details Replit's Nightly Self-Improving Agent That Reads Every Trace and Ships Its Own A/B Tests

SaaStr·low signal

Describing Replit's internal setup, Masad said an agent runs every night reading all user traces, finds what's broken, generates a pull request with prompt changes, ships it as an A/B test, and loops back autonomously — with constant model swaps feeding the improvement cycle. It's a concrete blueprint for a self-improving product loop, closely mirroring the 'systems that learn run-over-run' pattern many solo builders are chasing. Treat the exact framing as illustrative rather than a dated announcement.