Contact Book vs vibe-coding your own personal CRM

The first eight hours with Cursor / Claude Code / Cline get you a stunning amount: a Next.js app, a Postgres schema, contact list, contact detail, basic CRUD, a CSV import endpoint that mostly works on the file you tested. By Sunday afternoon you have screenshots that look indistinguishable from a real product.

What's missing isn't features. What's missing is the hundred small productionisations every shipped product carries: rate limits, audit logs, soft delete, real auth flow, password resets, mail bounces, suspended state, GDPR data export, GDPR data delete, content-security policy, third-party trackers absent, two-factor auth, refund flow, plan switching. None of this is fun; all of it matters.

Three weeks in you're debugging a cron worker that didn't fire because Vercel's serverless cold-start timed out. Two weeks after that you're rebuilding the import flow because Google's CSV format changed shape. Three months in you're chasing a soft-delete bug that lost six contacts because you and your AI assistant misunderstood each other about whether is_deleted defaulted to false.

None of these moments is a tragedy on its own. Together they add up to a very specific outcome: the side project quietly stops being maintained around month four. The CRM either drifts into staleness or gets abandoned for a hosted tool. The right answer isn't don't ever build - it's don't build the thing whose value is operational, only the thing whose value is creative.

If you love building, the productive move is the opposite of replicating us: use Contact Book for the boring, durable layer (storage, reminders, hosting, GDPR) and script on top of it with our MCP + CLI + drop-in clients. We ship the boring middle. You build the interesting edge - the AI agent that drafts your follow-ups, the export that pipes into your obsidian vault, the slack bot that nudges you about overdue people. That's where vibe-coding actually shines.