Vibe-coded apps are spawning a cleanup industry, QA founder says
Konstantin Klyagin, the Lisbon-based founder of contract software firm Redwerk and QA consultancy QAwerk, told The Register that a growing number of clients now come to him to fix apps built with AI coding assistants rather than to build new ones. Klyagin, who says he has been developing software for two decades, said his company posted on its website back in November that it offers vibe code cleanup, and that the work is not just about trimming lines of code. He argued that code volume is not a fair measure of quality and that less code does not necessarily mean more successful software. What matters, he said, is whether the business logic holds up and how the app handles arbitrary user behavior, including validation and security. A common issue he flagged is code duplication: in one app his company reviewed for a New York-based client, duplicate payment paths meant the price shown on the front page did not match the price shown during onboarding. He also described a permission-handling bug that let users skip straight to the sign-up and payment pages without creating a profile, plus forms that were not accessible to screen readers and incomplete test coverage. Klyagin said technically oriented founders who build their own apps tend not to need this cleanup work, while founders without a software background often do not know how to set up architecture and practices for a maintainable product; steering the AI model toward better structure, he said, is on the person directing it. He noted that before vibe coding, clients already asked for code reviews and refactoring as a normal part of the development cycle, or ahead of acquiring an application. Now that Claude Code and Codex are in common use, he said, there is a lot more QA and bug-fixing work, though he has not yet seen codebases built with open-weight models come through his company. He said his own team uses AI to keep up with the added workload, and that the core of what his company sells clients is discipline: specifying what is being built and testing it, a habit he said AI adoption has not made unnecessary. The article also mentions Slopfix, described as a separate team of three senior engineers who refactor vibe-coded codebases back to maintainability, as another example of the same trend.
Key facts
- Konstantin Klyagin, founder of Redwerk and QAwerk, says clients increasingly hire his company to clean up vibe-coded apps rather than build new ones.
- Recurring problems he cites include code duplication, weak permission checks, and forms that are not accessible to screen readers.
- In one client app, duplicate payment paths meant the price shown on the front page did not match the price shown during onboarding.
- A separate outfit called Slopfix, a team of three senior engineers, is named as another company doing the same kind of vibe-code cleanup work.
- Klyagin says Claude Code and Codex are now common in the apps his team reviews, though he has not yet seen ones built with open-weight models.
Why it matters
The piece captures a service industry forming around a byproduct of AI coding: apps built quickly with AI assistants but without the practices that keep software maintainable, secure and accessible. Klyagin's account suggests demand for this cleanup work is steady enough to be a standing line of business rather than a one-off favor.
Who it affects
Founders and companies who built products with AI coding assistants without a software engineering background are the ones described as needing this help. Contract development and QA shops like Redwerk, QAwerk and Slopfix are the ones positioning themselves to do the fixing.
How to use it
Klyagin's framing is that AI coding tools need direction: someone has to set restrictions and steer the model toward sound architecture and infrastructure rather than trusting default output. His prescription for avoiding the bugs he describes is the same discipline that predates AI coding: specify what is being built, and test it.
How solid is it
The account rests on one named source describing his own company's work and one anecdote about a New York-based client, without figures for revenue, client count or how fast the cleanup business is growing. The Slopfix example is mentioned only in passing, with no detail on its size beyond the stated team of three or its relationship, if any, to Redwerk or QAwerk.
Risks and caveats
The specific bugs described, duplicate payment paths, permission bypasses, inaccessible forms and thin test coverage, are attributed to unnamed client apps and are not independently verified. The article does not say which AI tools were used to build the flawed apps in question; Claude Code and Codex are mentioned only as tools developers commonly use now, not as the tools behind those specific bugs.
“It's not just how it was coded and what it was supposed to do in the best case scenario, but also how validation works, if there are security problems.”
— Konstantin Klyagin, founder of Redwerk and QAwerk