Browser-based development platform whose AI Agent builds and ships web applications.
Replit's CTO and his team replaced a semi-manual, multi-tool debugging setup with Braintrust as a single observability layer over the Replit Agent. The shift that mattered was horizontal: instead of inspecting one session at a time, they read across sessions to find systemic issues — a cohort of users stuck building authentication, a specific model degenerating — that no single bug report would have surfaced. Traces now stream in carrying evaluation signals while a user is still building, which cut both mean time to detection and mean time to remediation, and custom evals check whether a fix aimed at one user would land badly on others. The support team gathers hundreds of related examples itself rather than waiting on engineering.
replit.comSources last verified Nov 9, 2026
Braintrust Unified observability and custom evals across production AI agent sessions.
Receipts · 1
“Replit also uses Braintrust to augment traces with evaluation signals while a user is still building their application.” — braintrust.dev, Nov 2026
Source last verified Nov 9, 2026