Replit Introduces Safer “Vibe Coding” with Separate Dev & Prod Databases
This comes after a Replit user revealed how an agentic AI assistant within Replit destroyed months of work.

Replit has launched a beta feature to separate development and production databases, making its “vibe coding” environment much safer for users building live applications.
This comes after a Replit user revealed how an agentic AI assistant within Replit destroyed months of work by deleting the company’s entire production database — and then lied about it.
📢 We're excited to announce the beta launch of separate development and production databases for Replit apps, making it safer to vibe code with Replit.
— Replit ⠕ (@Replit) July 21, 2025
Starting today, we’re rolling out this feature for new Replit apps — gradually reaching all Replit users and apps over the…
Previously, Replit apps used a single database for both work-in-progress and production data—a setup that, in light of recent mishaps, posed significant risk.
Now, new apps will get two distinct databases:
- Development: Used for testing and iteration during code changes
- Production: Reserved for live user data
During deploys, Replit’s agent identifies schema modifications in the development database and prompts users to approve or reject changes before applying them to production. This ensures that potential issues like accidental deletions or corrupt schema updates don’t affect live systems.
"This is the next step in making Vibe Coding even more secure on Replit. This feature adds to other security enhancing features for databases users already have access to like 'Point-in-time Restore' which allow users to securely rollback any changes to their databases," Replit said in a blog post.
The company plans a phased rollout: the feature is now in Beta for new apps, then will be extended to all users and existing apps, with automatic migration in coming weeks. No user action will be required for the transition. Replit sees this as a critical step toward better AI-assisted development.
Comments ()