AI-native IDE workflows are not "let the model write everything." They're compressed senior loops: explore → diff → verify → ship.
My Cursor stack
- Project rules — encoding conventions (commercetools patterns, SEO requirements, Node 18)
- MCP servers — browser, Postgres, GitHub without context switching
- Agent for breadth — refactors, test scaffolding, content migrations
- Human for judgment — architecture, security, product tradeoffs
What AI accelerates on real client work
| Task | AI role | My role |
|---|---|---|
| Coveo integration boilerplate | Generate adapter | Review ranking + analytics |
| n8n workflow JSON | Draft nodes | Secure webhooks + idempotency |
| Portfolio / marketing copy | Draft + SEO pass | Truth-check claims |
| LangGraph nodes | Scaffold tools | Define state + guardrails |
Discipline beats vibe coding
"Vibe coding" demos well; production FinTech and commerce need verification gates:
npm run buildbefore every success claim- No secrets in prompts or commits
- Diff scope control — one feature per PR mentally even when solo
Takeaway
Cursor is a force multiplier for senior engineers, not a replacement. The portfolio you're reading was built with AI assistance and human review — same model I use for Wizly, Revenued, and Rich Cookies deliveries.
