Building an Uber-style app from scratch with AI can consume millions of tokens once retries, schema rewrites, and billing edge cases are included. This page quantifies those costs and compares them with a pre-built architecture workflow.
Phase-by-phase token breakdown
| Phase | Scratch | With Boilerplate |
|---|---|---|
| Auth + sessions | 2.1M | 120k |
| Schema + migrations | 1.7M | 80k |
| Routing + middleware | 2.4M | 140k |
| Billing + entitlements | 3.6M | 220k |
| CI/CD hardening | 900k | 50k |
Bottom line
Scratch path estimate: 10.7M tokens. Boilerplate path estimate: 610k tokens. At common coding-model rates, that delta alone often exceeds the cost of the boilerplate.
Skip infrastructure loops with MyAppTemplates and spend your AI budget on differentiating features.
Frequently asked questions
- How reliable are these token cost estimates?
- Estimates are based on repeatable phase budgets and retry multipliers, then calibrated with observed generation logs.
- Why does retry tax dominate total cost?
- Most spend comes from re-generating architecture glue code after errors or mismatches, not from first-pass feature prompts.