AI slop is good for business if you know what you’re doing

ai and ML

Your irresponsibility is someone else’s opportunity

Amazon founder Jeff Bezos is known for saying, “Your margin is my opportunity.”

Substitute “irresponsibility” for “margin,” and you have a formulation that applies to AI-driven software development.

AI models or coding agents make mistakes, and many of the people using them multiply those errors by asking their AI helper to build software without the knowledge necessary to build actual proper software.

The resulting vibe-coded apps have created a new growth industry – AI slop sanitation. Witness Slopfix, “a team of three senior engineers who refactor vibecoded codebases back to maintainability.”

Konstantin Klyagin, the Lisbon, Portugal-based founder of contract software biz Redwerk and QAwerk, a quality assurance consultancy, told The Register that a growing number of clients are asking for help to fix vibe-coded apps.

Klyagin said he’s been developing software for the past two decades, and with the adoption of generative AI, customers have been seeking help with their vibe-coded applications.

“Back in November,” he said, “we posted on our website that we do vibe code cleanup. And for us, it’s not just reducing the amount of lines of code but also making sure the product is production ready and can serve actual customers, because I don’t think that volume [lines of code] is a fair criterion to consider. Less [code] volume doesn’t necessarily mean more successful software.”

On the surface, Klyagin said, vibe-coded apps tend to look good. But what’s important to understand is how the business logic is implemented and how the app handles arbitrary user behavior.

“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,” he explained.

A common issue, Klyagin said, is code duplication. In an app his company reviewed for a New York-based client, he said, there were duplicate payment paths. So a price mentioned on the front page of the app differed from the number presented during the onboarding flow, he said. 

Then there was a problem with the permission handling. “So you could go straight to the sign up page and to the payment page, skipping the creation of the profile,” he said.

Then there were problems with the forms, which were insufficiently accessible for screen readers. And the test coverage was incomplete.

Technically oriented company founders who create apps, Klyagin said, tend not to need vibe-coding cleanup services. But those who lack software development experience don’t know how to set up the proper architecture and practices for building a maintainable app.

“It’s up to you to set certain restrictions and steer [the AI model] in the right direction towards better software architecture and also infrastructure,” he explained.

Before vibe coding became a thing, Klyagin said, clients asked for code reviews and refactoring, which he described as an integral part of the software development life cycle. Or they might ask for a code review when planning to acquire an application.

Now that people commonly use Claude Code and Codex – although he hasn’t yet seen codebases built with open-weight models – there’s a lot more QA and bug fixing.

Klyagin said his company has been managing the extra work with the help of AI.

“AI makes everyone faster,” he said. “We and our customers are now able to ship more code and more features. Some do it with more discipline and some with less discipline.

“So what we do, we just explain discipline to founders and to enterprises that adopt gen AI for coding. Because you know this discipline is not going anywhere. You still need to specify what you’re building and you still need to test it.” ®