A code review for the app AI wrote? That's the job.

A jeweller's loupe resting on an opened clockwork mechanism, a few loose springs beside it

Somebody on your team opened an AI tool and built an app the company now leans on. For a while it carried real work. Now it breaks in ways nobody can explain, and the person who prompted it cannot tell you what half the code does.

Pentangle audits and refactors apps built with AI tools: we read the code before we say a word about it, close the security holes, and leave you with software an engineer can maintain.

How does a vibe coded app become tech debt?

A vibe coded app becomes tech debt the way most of them start: an analyst who knows a little code opens an AI tool and builds a stopgap. The stopgap works, so it spreads, until the business leans on it. An AI tool wrote code nobody read, and now every change is a gamble, because nobody can say what a change will touch. The app still runs. What you have lost is the ability to change it safely.

No one did anything wrong here. But an app nobody has read carries risks the older kinds of tech debt did not: you cannot say who can reach your customer data, whether the quote math holds in every case, or what else stops when it goes down. A new engineer cannot safely touch it either, because there is no spec, no history, and no one to ask. This is the tech debt AI-written code leaves behind, and it compounds just like the rest. Your analyst's stopgap fix. Our day job.

What we usually find in an app built with an AI tool

Most of what we open was purpose built: a real problem somebody had, solved with the tools they had in hand. It usually is not architecturally optimized, and we often find user experience issues, missed conventions, security holes, and performance issues. But there is a solution to a problem in there, and we respect it. We make sure credit is given where it is due: we like people who roll up their sleeves and do the work. We are those people too; we just carry some know-how in this field.

A cracked machine casting mended with a riveted brass patch plate, the machine still standing

What the code audit includes

The audit is a working engineer's read of your whole app: the code, the data, the logins, and how the app sits inside the business. You get a written account of what is sound, what is risky, and what we would change and why, with the work put in order. Every audit covers:

  • Security: who can sign in, what each account can reach, and where keys, passwords, and customer data live
  • Stability: the recurring bugs, the failures nobody can reproduce, and what happens when a request fails
  • Data: whether what the app stores can survive more users, more orders, and more years
  • Maintainability: whether an engineer who was not in the chat can change the code without guessing
  • The call on each piece: keep it, fix it, or replace it, with our reasons in plain words

The account is written so you can hand it to your board or to your own engineers.

How the cleanup works

The work runs in a fixed order: hear the problem, read the code, report what we found, then fix only what you approve.

  1. You tell us what the app does, what it should do, and where it hurts. If the person who prompted it is still around, we want them in that conversation.
  2. We read the code as it stands.The audit changes nothing; this step is reading, tracing, and testing.
  3. You get the written account: what is sound, what is a risk, what we would change and why.
  4. You decide.If we think a piece should change, we say why, and the decision stays yours. Your original path gets built after we have said why we would change it.
  5. We refactor in small steps: a first working version, then a constant feedback loop with the people who use the app.

And when the fix should reach further than the app itself, its numbers handed to the systems around it instead of a person retyping them, that is the same practice: your systems finally talking to each other.

What does the security review catch in a vibe coded app?

The security review traces every way into the app and every way out of it: who can sign in, what each login can reach, where the keys and secrets live, and what the app gives away when something fails. It covers a customer-facing web app and an internal tool the same way, because an app can run clean every day and still be wide open. The review asks the questions daily use never does:

  • Access: which accounts exist, what each one can reach, and which leftover logins still have a way in
  • Secrets: where API keys, passwords, and customer data sit, and who else could read them
  • Failure: what an error shows the outside world, and what happens to the data in mid-crash
  • Input: what the app does when a form or a request feeds it something it did not expect

Fixing what the review finds does not freeze your company out of its own tool. Get the app locked down without locking out the team.

Will you tell us to scrap it and start over?

No verdict comes before the code has been read. The service is auditing and refactoring: keep what works, rebuild what is dangerous underneath it, and keep the app standing while that happens. Where a piece truly cannot be saved, you get the reason in plain terms, and the decision stays with you, piece by piece, never as one take-it-or-leave-it rewrite.

They told you to start over and never opened the code? Here the reading comes first. And when the app that worries you long predates AI tools, handwritten code grown over years, that is its own practice: modernizing the code you already have.

Who reads your code

We read it ourselves.

A player piano roll unwinding across the brass tracker bar that reads it

The people in your code are the people doing the work: Pentangle's founder, based in the United States, and specialists we trust, brought in as the work needs them. We have built software for whatever industry walked in, and run technology as a CTO.

We look at the whole business before we look at the code, asking even the naive-sounding questions, because an app only makes sense next to the process it serves. A career spent translating between the people who run operations and the people who write software matters here too: a vibe coded app is a business process and a codebase tangled together, and both ends need reading. And when the audit raises questions bigger than one app, who owns technology, what gets built in what order, the engagement can grow into a fractional CTO who knows a patch from a fix.

Common questions about vibe coding cleanup

The title started as a joke online, and the problem underneath it is real: apps an AI tool wrote that nobody has read, carrying real work. Whatever name it travels under, vibe code fixer, vibe code cleaner, cleanup specialist, what you are buying is a code audit and a careful refactor. At Pentangle it sits on the services list in plainer words: auditing and refactoring apps built with AI tools.

Yes. Code is step-by-step instruction, and it can be read whether a person or a model wrote it. What the code cannot tell us is the business rule it was meant to carry, which is why the audit starts by hearing the problem from your people before the repository gets opened. We have built and reworked software for whatever industry came in for years; AI-written code is read the same way, line by line.

The audit cannot break anything: it is reading, tracing, and testing, and it changes nothing. When repairs begin, they come as a first working version, then a constant feedback loop with the people who use the app, so each change is seen by the people it touches and the next one waits on their word.

Nobody can honestly price a repair on code they have not read, so the audit comes first, scoped as its own piece of work. What it finds becomes a list of projects, put in order, each one priced and approved by you before anything is built. You decide piece by piece; there is no single take-it-or-leave-it number for a rewrite you did not ask for.

No. An AI tool got you an app that carries real work, and that speed is worth keeping. Cleanup makes the speed safe: the code gets read, the risky parts get rebuilt, and the structure gets set so the next AI-assisted change has something solid to land on.

Tell us what the AI built

Send the short version: what the tool was, what the app does, and what it did recently that worried you. We will read it, come back with plain questions, and set up the audit from there. The app is already carrying real work; the job now is to get the app fixed without killing the vibe.