Chapter

Pick the right tool — so you don't build production on a prototype engine (agent vs one-shot builder)

Know what kind of tool you're holding. Two families:

• One-shot app builders — Lovable, Bolt, v0, Replit, Base44. Describe an app in a text box; they generate a full stack (often React + Supabase) and deploy to a URL, managing the infra for you. The loop is fast and addictive — superb for blank-canvas prototypes, MVPs, and non-technical founders. (Several run on Claude under the hood; Lovable is a scaffolding/deploy layer on the same intelligence.)
• Coding agents — Claude Code (also Cursor, Aider). A terminal-native agent that works inside your real repo — your git, tools, tests, stack. No capability ceiling, you own the code, it makes targeted diffs (not full-file rewrites), runs commands, opens PRs.

The dividing line is prototype vs production and code ownership. One-shot builders plateau at CRUD on a fixed stack; apps that outgrow them are rebuilt, not refactored — there's no clean path to a production codebase, plus vendor lock-in.

A coding agent has the higher ceiling but demands engineering discipline (git, tests, architecture) in return.

The pragmatic play: prototype on a one-shot builder to validate, then rebuild on Claude Code once it has traction — or start in Claude Code if you're technical and it's meant to live in production. This guide is about the second mode: owning and operating the result, not shipping a demo.