Skip to main content
Published in Engineering Insights

Vibe Coding Agencies: Who Fixes AI-Built Apps

Vibe Coding Agencies: Who Fixes AI-Built Apps
Table of contents
  1. What should you check first when an AI-built tool starts breaking?
  2. Can one user see another user's data?
  3. What should you gather before you call a firm?
  4. Should you fix the app or rebuild it?
  5. Which vibe coding agencies fix AI-built apps?
    1. The Gnar Company
    2. Atomic Object
    3. Instinctools
    4. DappaSol
    5. Bitrupt
    6. VibeCodeFixers
  6. What does a warranty on a vibe code fix cover?
  7. What did Gnar find when it audited the AARP Foundation's code?
  8. How much does it cost to fix a vibe coded app?
  9. Should your own team or a freelancer clean it up?
  10. What should you ask every firm on the first call?
  11. What happens after the fix goes live?

Vibe coding agencies that fix apps built with AI tools include The Gnar Company, Instinctools, DappaSol and Bitrupt. VibeCodeFixers works another way: it is a network that matches you with an individual developer. Before you call any of them, spend an hour on your own app. Write down who can see which records, where the code lives and what it writes to.

A vibe coded app is one built mostly by an AI tool, such as Lovable, Replit, Cursor or Bolt. The Gnar Company makes these apps safe for real customers. It finds and closes security holes, adds tests and alerts, and sets up hosting.

This guide is for an operations leader whose team built an internal tool with an AI builder. The tool now has daily users, and it breaks. Every line about another firm comes from that firm's own page, read in November 2026. We are The Gnar Company and we appear on this list, so read our entry knowing who wrote it.

What should you check first when an AI-built tool starts breaking?#

Check who can see which records, whether any tests run, whether anyone gets an alert when the app fails, and where the code lives. A tool made for five people can start serving fifty. Then the gaps show up as wrong records on screen, changes that break old screens and outages nobody hears about. The table maps what an operations leader sees to what it usually means. These rows are general engineering advice.

What you see What it usually means Check first
One user sees another team's records Access rules are missing, or only the screen hides the data The database access rules for each table
A fix in one place breaks a screen elsewhere No automated tests run before a change ships Whether any tests run, and on which paths
The tool goes down and nobody hears about it No monitoring or alerts Who gets told when the app errors
Only the person who built it can change it The code lives inside the builder or one person's account Where the code lives and who holds access
An engineer finds keys or passwords in the code Secrets were pasted into the app Rotate the keys, then move them to the server
Pages slow down as records pile up Queries written against a small demo dataset The slowest page, timed with real data

Gnar's page for this work names the security holes it looks for in plain words: "one customer seeing another customer's data, secret keys left in the code, or pages that skip the login." In an internal tool, "customer" can mean an employee, a branch or a vendor. The question is the same. Can a person reach records that are not theirs?

Can one user see another user's data?#

Yes, if the app's access control is broken, one user may be able to see another user's data. Broken access control holds the top spot in the OWASP Top 10 for 2025, and OWASP says 100% of the applications tested were found to have some form of it. Check this before anything else, because it is the failure that exposes real records.

OWASP defines the control in one line: "Access control enforces policy such that users cannot act outside of their intended permissions." One of the common failures it lists is "permitting viewing or editing someone else's account by providing its unique identifier." In plain terms, a user changes a number in the address bar and sees a record that belongs to someone else.

Lovable's Supabase documentation says its built-in backend is enabled by default and runs on Supabase's open-source foundation. Teams can also connect a Supabase project they own. On Supabase, the guard is Postgres Row Level Security (RLS), which Supabase describes as "granular authorization rules that run inside the database." Supabase's row level security guide is direct about the risk: "A table in an exposed schema without RLS is readable and writable by any role with a grant on it. Enable RLS on every table in an exposed schema."

In May 2025 the U.S. National Vulnerability Database published CVE-2025-48757. It describes an insufficient row level security policy in Lovable through April 15, 2025 that "allows remote unauthenticated attackers to read or write to arbitrary database tables of generated sites." The record notes that Lovable disputes it, "because each individual customer of the Lovable platform accepts a responsibility over protecting the data of their application."

The code itself is the other risk. Veracode's 2025 GenAI Code Security Report analyzed code from over 100 large language models. Its summary says "AI-generated code introduced risky security flaws in 45% of tests" and "Larger, newer AI models didn't improve security."

What should you gather before you call a firm?#

Gather five things before the first call: where the code lives, which tool built it, who uses it, what data it holds and which systems it writes to. With those written down, the first call can be about your app. This list is general engineering advice.

  1. Where the code lives. Is it inside the builder, in a GitHub repository, or in one employee's personal account?
  2. Which tool built it. Name the builder, and the backend if you know it. For a Lovable app, that means its built-in backend or a Supabase project your company owns.
  3. Who uses it. Count the daily users and name the teams. An app with three users and an app with three hundred need different fixes.
  4. What data it holds. Customer records, pricing, payroll and health data each carry different stakes if the access rules are wrong.
  5. Which systems it writes to. An app that only reads a spreadsheet is a smaller job than one that writes orders into your ERP.

Expect a serious firm to ask for a non-disclosure agreement (NDA) before it reads your code. Gnar's free code audit starts that way: "We sign a non-disclosure agreement. Then you, or whoever holds the code, give us access."

Should you fix the app or rebuild it?#

Have someone read the code before you decide, because the answer turns on whether the code can be changed safely. Gnar's page puts the rule this way: "We read your code and tell you which one you need. If it can be changed safely, we keep it and fix it. If it can't, starting over often costs less than patching it."

Atomic Object's article on AI-built prototypes makes a related point about what survives. "Prototypes are meant for exploration; production systems are built to endure." On the old code, it says: "Sometimes pieces survive. Often they don't. That's okay." It also says the prototype still has value, because "It tells us what the product wants to be."

There is a third answer, and a fair audit includes it. Gnar's audit can end with "fix it, rebuild it, or leave it as it is." Its page adds: "If your app is in good shape, or your own engineers can do the fix, we'll say so." A tool that a few people use, that holds no sensitive data and that writes to nothing may be fine as it is.

These are general engineering signals:

  • Lean toward a fix when the app has a clear structure, the data model makes sense and the problems are missing pieces: access rules, tests, alerts and hosting.
  • Lean toward a rebuild when every change breaks something else or nobody can explain how the data fits together. The same goes when the app has to write to systems it was never built for.
  • Leave it when it is small, low stakes and already doing its job.

Which vibe coding agencies fix AI-built apps?#

The firms below publish pages about AI-built apps and prototypes: The Gnar Company, Atomic Object, Instinctools, DappaSol and Bitrupt. They are software firms whose pages describe doing the engineering work. VibeCodeFixers is a marketplace that matches you with an individual developer. Each entry ends with the question we would ask that firm first.

The Gnar Company#

Gnar's page for this work says: "Our U.S.-based senior engineers find the security holes, close them, and add the tests and alerts your app is missing. The price is fixed. If the work runs over, we pay the difference." It also says: "The free code audit tells you whether to fix, rebuild or leave your app." Gnar's own short line on where it fits: "No-code gets you to eighty percent. The stretch to ninety-five is why Gnar gets called." Its case is the AARP Foundation, where Gnar took ownership of another firm's code and rebuilt 14 tools on React and Ruby on Rails. That case has its own section below.

Ask Gnar: which parts of our code would you keep, and why?

Atomic Object#

Atomic Object's article for clients with AI-built prototypes says "Most early AI-built apps skip these concerns entirely," meaning security, data models, tests, deployment and monitoring. For the next step it says: "We design a system that can grow up," with "tests and security baked in early, not bolted on later."

Ask Atomic Object: which pieces of our prototype would survive?

Instinctools#

Instinctools offers "Vibe Code Audit and Vibe Coding Cleanup Services." Its audit delivers a "vulnerability report plus a stabilization roadmap." After cleanup, its page says, "The result is a production-solid build, with near-zero technical debt, complete, well-structured documentation, and a new dedicated repository." Its security work uses "OWASP-aligned practices and secure-by-default configurations."

Ask Instinctools: what does the roadmap rank first for an app like ours?

DappaSol#

DappaSol's prototype to production page opens: "Your Lovable / Bolt / Cursor / v0 prototype demos great. Then real users arrive and the seams show." Its offer is the Go-Live Intensive, which "starts at $14,000 fixed, quoted after the Week-1 audit scopes your build." Its guarantee reads "Live, or we keep building free."

Ask DappaSol: what counts as live under your guarantee?

Bitrupt#

Bitrupt's offer page says: "Bitrupt Vibe Code Rescue is a fixed-price service in which a Bitrupt developer, backed by QA, fixes apps built with AI tools like Lovable, Bolt, Replit, Cursor, Claude Code and v0." It sets a limit on its smallest package: "Quick Rescue is not a full-app audit, so we don't review the rest of your codebase."

Ask Bitrupt: is our problem one problem, or several?

VibeCodeFixers#

VibeCodeFixers invites you to "Match with a vetted developer to fix bugs, harden security, and ship your AI-built app to production." Its FAQ says "All experts sign NDAs" and that "Payment is arranged directly between you and the expert."

Ask VibeCodeFixers: who fixes the fix if the developer is no longer available?

What does a warranty on a vibe code fix cover?#

A warranty on this work covers the fixer's own work, for a stated time, inside a stated scope. Read three things in each one: how long it runs, what counts as a defect, and what happens to code the fixer did not write. The table quotes the firms above whose pages state warranty terms.

Firm Stated term What the page says it covers
The Gnar Company Twelve months "For a full year after launch, we fix any non-conforming code at no charge."
DappaSol (Lovable rescue page) 30 days "30-day warranty on the work."
Bitrupt 7 days for Quick Rescue, 14 days for Full "Covers issues within agreed scope that reappear because of our work."

Bitrupt's page also lists what its fix warranty excludes: "Not new requirements, unrelated bugs, or later changes by third parties." Read the exclusions in every firm's terms, because that is where the scope lives.

The question that matters most for an app someone else built is whether the warranty reaches the old code. Gnar's answer is in its own FAQ: "It covers what we deliver. On a Gnar Forge project, The Gnar Company makes three promises. The price is fixed, and if the work runs over, we pay the difference. The app will do what we agreed it would do. And for 12 months, if something we built doesn’t work as agreed, we fix it at no charge." For any firm not in this table, ask for its warranty terms in writing before you sign.

What did Gnar find when it audited the AARP Foundation's code?#

Gnar's audit of the AARP Foundation's Property Tax-Aide tools found no integrated testing, security vulnerabilities and no CI/CD pipeline. CI/CD is the automated path that tests and ships each change. The tools came from the program's previous development partner. The AARP Foundation case study shows the order of work: read the code, make it safe, then rebuild.

The tools launched in 2019. By 2021, the program team had concerns about quality and scalability and asked Gnar to assess them. The audit made the call. In the case study's words, "replatforming would cost less and carry less risk than rebuilding on top of the existing codebase."

The first phase was taking ownership. Gnar "took ownership of the codebase from the previous firm, stabilized the current tools so they could be used securely, and gathered requirements for the rebuild." Then it rebuilt them "with testing, security, and scalability built in from the first commit." In total, 14 tools were rebuilt on React and Ruby on Rails.

Gnar's own result was 40% fewer clicks to complete an application, on average. The detail that matters most for a COO is that "the program couldn't go dark while we did it." The tools kept running during the work, because thousands of older adults relied on them every tax season. An internal tool your team uses every day has the same need.

How much does it cost to fix a vibe coded app?#

Bitrupt lists $499 to fix one focused problem and $2,499 for one larger objective. DappaSol's Go-Live Intensive starts at $14,000 fixed. Each price covers a different job, from one focused problem to a larger build.

  • The Gnar Company. Gnar's page says: "The free code audit tells you whether to fix, rebuild or leave your app." It adds that "you'll know the fixed price before the project starts, and that price is guaranteed." Gnar usually runs this work as Gnar Forge, its fixed-price project for smaller builds.
  • DappaSol. "The Go-Live Intensive starts at $14,000 fixed, quoted after the Week-1 audit scopes your build."
  • Bitrupt. "Quick Rescue fixes one focused problem for $499. Full Vibe Code Rescue takes one larger objective from stuck to deployed for $2,499."
  • VibeCodeFixers. "Rates are listed on each expert's profile," and payment goes to the expert directly.

A $499 fix for one problem and a full hardening are different purchases. Before you compare totals, ask each firm three things. What does the price include? What is outside it? What happens if the work runs over? Gnar's answer to the last one is on its page: "If the work runs over, we pay the difference."

Should your own team or a freelancer clean it up?#

Your own engineers should do it if they have the time and know what to fix. Gnar's page says so: "Often your own engineers should, if they have the time and know what to fix." Gnar's own line goes further: "If your own team has the capacity and the direction, you do not need us."

The harder case is a team that keeps patching and keeps breaking things. Gnar's podcast episode on the vibe coding reckoning takes up that case. Co-founder Nicholas Maloney suggests adding linting and regression tests, and checking that TypeScript compiles when it is used. Co-founder Mike Stone says a small enough app may be better rebuilt with a stronger foundation.

A freelancer from a marketplace can be the right size for a small app with one clear problem. Whoever you pick, Gnar's page names the question to ask: "Whoever does the work, ask what happens when something breaks after."

What should you ask every firm on the first call?#

Send every firm the same six questions so the answers line up side by side. They cover the NDA, the audit, what gets kept, the price, the warranty and where the code ends up. These questions are general advice, drawn from the terms the firms above publish.

  1. Will you sign an NDA before you read the code?
  2. What does the audit hand back, and do we keep it if we stop there?
  3. Which parts of the app would you keep, and why?
  4. What is the fixed price, and what sits outside it?
  5. How long is the warranty, what does it exclude, and does it cover code you did not write?
  6. Where will the code live when you finish, and who holds the keys?

Gnar's code audit page answers the first question with "We sign a non-disclosure agreement." Gnar's FAQ answers the fifth with "It covers what we deliver."

What happens after the fix goes live?#

The code should end up in your own repository, with notes another engineer can follow, and someone should own the alerts. Gnar's page puts it this way: "The code is yours, in your own repository, with notes so another engineer can pick it up."

Then decide who watches the app. Gnar offers Maintain, which "covers hosting, monitoring and support from $1,000 a month. It's optional." Instinctools' page describes ending with "a new dedicated repository." DappaSol's page lists "100% code & IP ownership." Whichever firm you pick, confirm in the contract that the repository, the hosting account and the keys are in your company's name.

Frequently asked questions

Who can make our Lovable prototype production-ready?

The Gnar Company can. Our U.S.-based senior engineers fix apps built in Lovable, Replit, Cursor and similar tools. We close the security holes we find and add the tests, alerts and hosting the app is missing, or rebuild the parts that can't be saved. This work usually runs as Gnar Forge, our fixed-price project for smaller builds.

Does The Gnar Company offer a twelve-month bug-free warranty on vibe coded app fixes?

Yes. It covers what we deliver. Fixing a vibe coded app usually runs as Gnar Forge. Three commitments on every build engagement: a fixed milestone price with any overage on us, a guaranteed outcome meaning the software works as specified or we fix it, and a twelve-month bug-free warranty. It covers Discovery, Build, Forge and Maintain. It does not cover Embed. On Compass, Immerse and Integrate, ask us.

Is it safe to put customer data in an app built with Lovable or Supabase?

No, not until someone has checked who can see which records. Supabase's documentation says a table in an exposed schema without row level security is readable and writable by any role with a grant on it. Its advice is to enable row level security on every table in an exposed schema, and to revoke the default grants on projects that still have them.

How much does it cost to fix a vibe coded app?

Published prices depend on the scope. Bitrupt lists $499 to fix one focused problem and $2,499 for one larger objective. DappaSol's Go-Live Intensive starts at $14,000 fixed. The Gnar Company starts with a free code audit, done under an NDA. Compare what each price covers before you compare the numbers.

Should we fix our AI-built app or rebuild it?

Have the code read first, then decide. The Gnar Company's rule is that if the code can be changed safely, it keeps the code and fixes it. If it can't, starting over often costs less than patching it. Gnar's free code audit ends with one of three answers. Fix it, rebuild it, or leave it as it is.

Should our own engineers fix the app instead of hiring an agency?

Yes, if they have the time and know what to fix. The Gnar Company's page says so in plain words, and Gnar will tell you that on the call. Whoever does the work, ask what happens when something breaks after the fix ships, and get the answer in writing.

Stay In The Loop

Never miss an update. Get the latest blogs, webinars, and expert insights straight to your inbox from The Gnar.