Summary of Key Points
Recently, a new business idea has gained popularity: Slopfix, a team consisting of three senior engineers, specializes in helping companies clean up AI-generated code that is “functioning but difficult to maintain.” This type of code is often created when companies use AI to quickly develop prototypes. However, as projects grow, the resulting code can become redundant, have messy architectures, and making minor changes can cause multiple issues to arise. This business has sparked intense debates among developers—some see it as a genuine need, while others argue that it’s a “pseudo-demand.” Research data also supports the idea that 22.7% of problems in AI-generated code persist for a long time (even more than nine months), highlighting the reality of this cleaning service, although the exact methods for doing so are still being explored.
Why Does AI-Generated Code Become Worse Over Time? This Business Is Designed to Clean Up the Mess
Many companies initially find using AI for coding very appealing because it’s fast and gets the work done quickly. However, they soon realize problems arise:
- When AI generates code, it’s like a blind man trying to describe an elephant; it doesn’t have a complete understanding of the project as a whole. Instead of reusing existing code, it simply copies and pastes the same logic whenever similar functionality is required.
- The code becomes increasingly messy: Redundant parts accumulate, leading to chaotic architectures. Adding new features becomes extremely difficult, and making small changes can cause multiple functions to fail.
Slopfix aims to address the consequences of this “empirical coding” approach by helping companies organize the messy AI-generated code, removing redundancy, and reengineering the architecture to make future maintenance easier.
Why Can the Slopfix Team Solve These Problems?
These three engineers are no ordinary programmers; their backgrounds specifically address the pain points of AI-generated code:
- They have long worked on “high-demand” projects: They previously collaborated on developing the Rust smart contract framework Odra, which requires extremely secure and maintainable code (e.g., free from errors and compatible with different blockchains). The issues with AI-generated code—redundancy, uncontrolled architecture, and lack of abstraction—are precisely what they are adept at resolving.
- They understand AI but don’t rely solely on it: While they use tools like Claude Code, they strictly limit the AI’s decision-making power. It’s the humans who guide the architectural design, with AI serving as a supplementary tool.
In short, they are experts who understand both AI and high-quality code, enabling them to address the root causes of the chaos in AI-generated code.
Controversy Rages: Some See It as a Genuine Need, Others Call It a “Pseudo-Need”
The launch of this business has divided developers into two camps:
Supporters:
- There is indeed a demand for such services. For example, some developers are already helping startups clean up AI-generated code. A CEO used Claude to create code that no one could maintain, so they had to hire professional engineers to sort it out.
- There is a clear value in this service, especially for three types of AI projects: Those created by people with little or no programming experience, as the resulting code is of poor quality and requires review by skilled engineers.
Opponents:
- They question the feasibility of the business model:
- Customers often struggle to articulate their business logic clearly, which leads to messy code in the first place. Even if the code is cleaned up, it’s difficult to maintain later on.
- Reusing AI to clean up AI-generated code can be counterproductive, similar to compressing an image multiple times, which only makes the issues worse.
- There are quality assurance and testing challenges: A two-week guarantee is too short; some problems may not be identified until months later. Without comprehensive testing, it’s hard to ensure that the reengineered code is free from errors.
Data Highlights the Problems with AI-Generated Code
A research team analyzed 6,299 GitHub repositories containing AI-generated code and found:
- A high proportion of issues: More than 15% of commits in each type of AI tool (Copilot, Claude, Gemini, etc.) introduced problems, with Gemini having the highest rate (29.1%) and Copilot the lowest (17.4%).
- Many issues persist over time: 22.7% of the problems remain unresolved even after several months.
- The types of issues include code duplication, unnecessary variables, logical errors, and security vulnerabilities (e.g., overly broad exception handling).
These data points demonstrate that the “technical debt” associated with AI-generated code is significant and does not resolve on its own, providing a solid basis for the existence of services like Slopfix.
Can This Business Succeed in the Long Term? Several Challenges Need to Be Overcome
For this business to become sustainable, several key issues must be addressed:
1. Understanding business logic: The core of code cleaning is understanding the business rules hidden within it, but customers may not always be able to articulate these clearly.
2. Maintaining the cleaned-up state: If companies continue to use AI for coding after the cleanup, the mess may return. A mechanism for continuous maintenance (e.g., establishing clear architectural standards) is needed.
3. Testing and quality assurance: Without thorough testing, reengineering can easily lead to new issues. The guarantee period should be extended, or companies should be helped to establish their own testing systems.
However, this business reflects a new need that has emerged with the widespread use of AI. Just like code outsourcing in the past, there will always be debates when new professions emerge. But as long as it addresses real problems, there is potential for growth.
In essence, AI has lowered the barrier to writing code, but it has also raised the bar for writing high-quality code. In the future, companies will either need to train their own engineers who can effectively manage AI or outsource to professional teams like Slopfix. This could be a new trend in software development in the AI era.