"Can we turn on payments next Monday?" It's the question we get most from founders who've built an app in Lovable, Bolt or Cursor over a few weekends. The demo works, a handful of early users love it, and nobody has written down what "ready" actually means.
That's the vibe coding production gap. The tools are very good at screens and flows that work for one happy user clicking through a preview. They're much weaker on the parts customers never see: which rows each user can read, where the keys live, what your server does when a payment webhook turns up twice, and how you'd get data back after a bad migration.
None of the nine fixes below needs a rewrite. Most take hours or days once someone knows where to look.
Why the preview hides the risky parts
You describe a feature, the tool writes code until the preview looks right, you move on. Row Level Security, environment separation and backups never show up in a preview, so nobody prompts for them and they don't get built.
The public research lines up with what we find in reviews. Escape, a security vendor, scanned more than 5,600 publicly available apps built on vibe coding platforms and reported over 2,000 vulnerabilities, more than 400 exposed secrets and 175 instances of exposed personal data, including medical records and bank account numbers. The write-up, published on 29 October 2025, names Supabase tokens sitting in frontend JavaScript bundles and missing Row Level Security as recurring causes.
A better model won't close the gap on its own. Veracode's 2025 GenAI Code Security Report, released on 30 July 2025, tested more than 100 large language models and found AI introduced security vulnerabilities in 45 percent of cases. You need a deliberate review before launch, not a cleverer prompt.
The vibe coding production checklist: 9 fixes before launch
Do these roughly in order. The first four protect customer data, the next three protect revenue and trust, and the last two keep you in control once you're live.
- Lock down data access. On Supabase, enable RLS on every table in an exposed schema (
alter table ... enable row level security) and write a policy per operation. Supabase's Row Level Security guide is blunt: a table in an exposed schema without RLS is readable and writable by any role with a grant on it. Test it properly. Sign in as user A, grab an ID that belongs to user B, and try to fetch it. - Get secrets out of the browser. Open your deployed site, search the built JavaScript for
sk_,service_roleand your AI provider's key prefix. Publishable keys are meant to be public. Secret keys, service role keys and OpenAI or Anthropic keys aren't. Move those calls into a server function and keep keys in your host's secret store, which is the pattern the OWASP Secrets Management Cheat Sheet describes. Anything that was ever shipped or committed gets rotated. - Split development, staging and production. If experiments and customers share one database, you'll lose customer data eventually. Supabase's managing environments guide uses separate projects for staging and production, with schema changes written as migrations (
supabase migration new) and deployed by CI. Do the same with Stripe, your email provider and your AI accounts: test keys in staging, live keys only in production. - Turn on backups and actually restore one. Check your plan. Supabase's backups documentation says Pro, Team and Enterprise projects are backed up daily, and recommends free tier projects export their own data regularly. Restore a backup into staging before launch. If you've never done it, you don't know it works.
- Finish the boring account flows. Demos skip password reset, email verification, session expiry and account deletion. Walk through each one. Check that a user removed from a team loses access immediately, not when their token expires an hour later.
- Make payment handling idempotent. Stripe's webhook documentation says live mode deliveries are retried for up to three days, events aren't guaranteed to arrive in order, and the same event can be delivered more than once. Verify the
Stripe-Signatureheader, store processed event IDs, and return a 2xx before doing anything slow. Then runstripe triggeragainst staging and replay a few events to prove it. - Authenticate your sending domain. A sign-up confirmation in the spam folder looks like a broken product. Google's email sender guidelines require SPF or DKIM for all senders, and SPF, DKIM and DMARC for anyone sending 5,000 or more messages a day to Gmail accounts. Send from your own domain, and allow for DNS propagation before launch day.
- Test the paths that make money. Forget 100 percent coverage. You want automated tests on sign-up, sign-in, the core workflow and checkout, run on every change. AI tools regenerate whole files, so something that worked on Tuesday can be gone by Thursday without anyone touching it on purpose.
- Add monitoring and name an owner. Capture frontend and backend errors, add an uptime check, and send alerts to a person's phone rather than a shared inbox. Agree now who answers when checkout fails at 9pm on a Friday.
What each fix protects
| Fix | Protects | Typical effort |
|---|---|---|
| Data access rules | Customer data, your reputation | Hours to a few days |
| Secrets management | Your accounts and your bill | Hours |
| Environments | Production data during changes | One to two days |
| Backups and restore test | Recovery from mistakes | Hours |
| Account flows | Support load, account security | One to three days |
| Payments and webhooks | Revenue accuracy | One to two days |
| Email authentication | Activation and receipts | Hours, plus DNS wait time |
| Core path tests | Release confidence | Two to five days |
| Monitoring and ownership | Time to detect and fix | Hours |
The effort column comes from our own work on small apps. It's a rough guide, not a quote. More integrations, more user roles or regulated data all push it out.
Where launches go wrong anyway
A pattern we see often: the founder asks the same AI tool to "make the app secure". It adds a nicer login screen and reports success, while the database still answers anyone holding the public key. You can't check security from inside the tool that wrote the code. Someone has to try to break it from outside.
The second trap is patching straight into production because there's no staging. Every fix risks a fresh outage, and when something does break, nobody can say which change did it. Set up environments first and every later fix gets safer.
The third is treating launch day as the finish line. Change is cheap with AI tools, so apps move fast after launch, and without tests and monitoring each change is a gamble. We go further into this in why AI-built apps work in the demo and break in production.
How JRC Software takes AI-built apps to production
Week one of our AI-Built App to Production service is a readiness review. We read the code, check every table's policies, pull the keys out of your live bundle and try them, look at your hosting, Stripe and email setup, and restore a backup if you have one. You get a ranked list: must fix before launch, can wait, fine as it is.
Then we fix the launch blockers in your existing code base and set up staging, backups and monitoring. You keep the stack and the code, and you can keep building with your AI tools afterwards. We'll also tell you what not to do, which is usually "don't rebuild this" and "don't launch payments before the webhook handler is idempotent". If security worries you most, start with the Supabase, API key and auth mistakes we see most.
If you'd rather talk it through first, book a discovery call with our team in Melbourne or Sri Lanka.
Not sure where your app stands? Get your free app launch-readiness check and we'll tell you which of these nine items need work before you go live.
Frequently asked questions
What does production ready mean for a vibe-coded app?
It means the app protects customer data, survives mistakes and can be changed safely. In practice: access rules on every table, no secret keys in the browser, separate staging and production, backups you've restored at least once, tested payment and account flows, authenticated email, automated tests on core paths, and monitoring with a named owner. A working demo gives you none of these by default.
Can I launch a Lovable app without rewriting it?
Usually, yes. Most Lovable, Bolt and Cursor apps we review have a workable structure and need targeted fixes, not a rebuild. The common work is tightening database policies, moving privileged calls server side, adding environments and backups, and testing anything that touches money or accounts. A rebuild only makes sense when the core data model can't support what the business needs next.
How long does it take to get an AI-built app ready for launch?
For a typical small app with sign-in, a database and payments, the fixes in this checklist often take one to three weeks of focused engineering. More integrations, user roles or sensitive data stretch that. A readiness review at the start gives you a concrete list and a realistic timeline before you commit to a launch date.
Is the Supabase anon key safe to expose in my frontend?
Yes, but only if every table it can reach has Row Level Security turned on with sensible policies. The public key is designed to live in the browser. Without RLS, anyone who copies it from your site can read or change your data. Secret and service role keys are different: they can bypass RLS and must never reach the browser.
Vibe Coding
Get your free app launch-readiness check
Answer a short checklist and we'll tell you what to fix before launch.
Start the free checkSources
- Escape: Methodology: How we discovered over 2k high-impact vulnerabilities in apps built with vibe coding platforms
- Veracode: AI-Generated Code Poses Major Security Risks in Nearly Half of All Development Tasks
- Supabase: Row Level Security
- OWASP: Secrets Management Cheat Sheet
- Supabase: Managing Environments
- Supabase: Database Backups
- Stripe: Receive Stripe events in your webhook endpoint
- Google: Email sender guidelines