Your AI-built app is breaking. Fix it, rebuild part of it, or start again?
If the data model fits your product, harden what you have: it is the fastest and cheapest path. If one layer such as auth or business logic is wrong, rebuild only that layer. Start again only when the data model itself is wrong, or the product has moved far past the prototype.
Last updated: · Written by Hunchbite
| Harden what exists | Partial rebuild | Full rebuild | |
|---|---|---|---|
| Choose it when | The product and data model are right. The problems are security, deploys and missing safety nets. | One layer is wrong (auth, data model, business logic in the browser), but the screens and flows are right. | The data model is wrong at the core, or the product has moved far from what the prototype does. |
| What changes | Database rules, secrets, authorization, CI/CD, monitoring, backups. | The broken layer is rebuilt on solid ground and the rest is kept. | Everything underneath. The prototype becomes the spec. |
| What stays | Almost all of the code, and all of your users and data. | Screens, flows, design and most integrations. | What you learned: which features and fields users need. |
| Typical time | About 1–2 weeks | Usually 4–8 weeks, scoped after the audit | Scoped like a new build |
| Hunchbite's published price | ₹2L–₹5L (approx. $2,100–$5,300); full hardening up to ₹12L (approx. $13,000) | Scoped after the audit. Major fixes on a taken-over project: ₹5L–₹15L+ (approx. $5,300–$16,000+) | Priced as a new build: a focused MVP is ₹3L–₹6L (approx. $3,200–$6,300), a complex one ₹6L–₹12L (approx. $6,300–$13,000) |
| Main risk | A deeper problem shows up later and you pay twice. | The seams between old and new code, and migrating live data. | Time. Users wait, and the scope grows to match the old app feature for feature. |
Prices are Hunchbite's published fixed-price ranges, quoted after a free audit. USD figures are approximate, at ₹95 = $1. Size moves the price more than the path: a full rebuild of a small app can cost less than hardening a large one. For the full price breakdown, see cost to fix a vibe-coded app.
Harden what exists
Many AI-built apps that are live with users land here. The tool got the product right and skipped the production layer.
Signals
- Users can sign up and use the core flow. The app does what it should.
- The problems are the usual AI-builder gaps: Row-Level Security off, keys in the browser bundle, no staging, no monitoring, no backups.
- One database with a schema that matches the product.
- A change usually works; it just ships by hand and without tests.
Typical scope
Lock down database rules on every table, move secrets and sensitive logic server-side, fix authorization, add a deploy pipeline, error tracking and backups. Hunchbite's security and hardening tier is ₹2L–₹5L (approx. $2,100–$5,300) over about a week. A takeover of a live app is ₹2L–₹6L (approx. $2,100–$6,300) over one to two weeks.
Risk
Hardening fixes what the audit finds. If the audit is shallow, a deeper problem (usually in the data model) can surface later. Ask for an audit that reads the schema, not only a scanner report.
Partial rebuild
The product is right but one layer cannot be trusted. Rebuild that layer and keep everything users see.
Signals
- Access checks run in the browser across the whole app, not on the server.
- Business rules (pricing, permissions, payouts) live in front-end code and must move server-side.
- Part of the data model is wrong: duplicated data, no relations, a table doing three jobs.
- Payments or a key integration fail often and are hard to debug.
Typical scope
Keep the screens, flows and design. Rebuild the broken layer (often the data model and API, or auth) and migrate live data into it. This is usually a 4–8 week engagement, scoped and fixed-price after the audit.
Risk
Data migration and the joins between old and new code. Plan a cut-over, test it on a copy of real data, and keep a way back.
Full rebuild
Less common than founders fear. It makes sense when the foundation, not the finish, is wrong.
Signals
- The data model does not fit the product, and fixing it touches every screen.
- The product has changed so much that most of the prototype no longer matches it.
- The app cannot be run or deployed outside the tool that built it.
- Few users and little live data, so starting again costs little in migration.
Typical scope
Treat the prototype as the spec: the flows and fields users proved they need. Build a clean version on a stack your next engineer can read, then migrate users and data. Hunchbite prices this like a new build: ₹3L–₹6L (approx. $3,200–$6,300) for a focused MVP, ₹6L–₹12L (approx. $6,300–$13,000) for a complex one, plus any data migration.
Risk
Time and scope. A rebuild that tries to copy every old feature before launch takes too long. Rebuild the core, launch, then add the rest.
Bad reasons to rebuild
- “The code is ugly.” AI-written code often looks messy and still works. Ugly is not a reason to pay for a rebuild.
- “A developer said it all has to go” before reading the schema and the deploy. Ask what they read and what exactly is wrong.
- “We want to move off the tool.” Moving the code out of Lovable, Bolt or Replit into your own repository and hosting is part of hardening, not a rebuild.
For software that was not built with AI tools, or for a longer scoring framework, read how to decide whether to fix or rebuild software.
How to decide this week
- Take the fix or rebuild tool (seven questions) to see where you likely stand.
- Get a code audit that reads the schema, the database rules and the deploy. Our AI code audit is free and you keep the report.
- Ask for the verdict in writing: harden, partial rebuild or full rebuild, and what in the code forces it.
- Pick who does the work. Compare a studio, a freelancer and a marketplace, or see agencies that take AI-built apps to production.
Built with a specific tool? Lovable · Bolt.new · Replit · Any AI tool
How do I know whether to fix or rebuild my AI-built app?
Look at the data model first. If the tables match the product and the problems are security, deploys and monitoring, harden it. If one layer such as auth or business logic is wrong, rebuild that layer. If the data model is wrong at the core, a full rebuild is often cheaper than patching. A code audit that reads the schema gives you the answer in writing.
Will a rebuild lose my users' data?
It should not. Both a partial and a full rebuild include migrating users and data into the new version. Test the migration on a copy of real data before the cut-over, and keep the old version until the new one is proven.
Can I keep my screens and design?
With hardening and a partial rebuild, yes: the screens, flows and design stay. With a full rebuild, the flows are kept as the spec and the design can be reused, but the code underneath is new.
How much does each path cost?
Hunchbite's published ranges: hardening ₹2L–₹5L (approx. $2,100–$5,300), full production hardening up to ₹12L (approx. $13,000), a takeover of a live app ₹2L–₹6L (approx. $2,100–$6,300). Partial rebuilds are scoped after the audit. A full rebuild is priced as a new build, from ₹3L (approx. $3,200). The audit is free.
Would you push a rebuild because it is more work?
The audit report says which path we recommend and why, and you keep it whether you hire us or not. If a fix will do, we say so. Get a second opinion if any agency, including us, recommends a rebuild without explaining what in the code forces it.
Who owns the code after the work?
You do. With Hunchbite you own 100% of the code, docs and deployment access, and the repository lives in your account.
Get the verdict in writing
Send us the repository. Our free audit tells you which path your app needs, what it costs, and why. You keep the report either way.
Request the free audit