Two quarters of vibe-coding
First posted on LinkedIn.
After two quarters using vibe-coding tools like Replit across the org, this is the pattern emerging:
What teams are actually building
Replacing SaaS — narrow internal tools that retire paid seats or heavyweight workflows.
Net-new tools — dashboards and utilities built directly by teams closest to the problem.
Escaping spreadsheets — apps replacing Google Sheets used as both database and UI.
Micro-sites — fast, local experiments for engagement and reward distribution.
What we’ve learned
- Replit works best as a thin, opinionated UI/workflow layer on top of existing systems.
- Apps that never touch internal data rarely compound in value.
- Velocity is high, but without standards, we risk fragmentation, manual steps, and hero-maintained tools.
What we’re trying next
- Wider read access, tighter boundaries: enable more internal datasets to be safely consumable by Replit apps, with scoped, auditable, and time-bound access.
- Ephemeral by design: enforce purge and retention rules so Replit remains a UI and workflow layer — not a shadow data store.
- Controlled write paths: narrow, traceable write access to internal APIs to remove CSV uploads and manual back-syncs.
- Sustainable ownership: patterns so tools aren’t coupled to their original builder.
- Meta-prompting: prompts that generate prompts, codifying how common classes of tools are built.