Larridin was recently cited by WebProNews in a long, thoughtful, and useful piece, "The Codebase That Breeds AI Slop: Why Inconsistent Patterns Doom Software Rewrites". The article cites one finding we see as particularly interesting: managers give AI-generated code above-average marks in reviews, but the same scores is below average in maintainability, generating more rework than human-generated work. The code, to use an old phrase, flatters to deceive. And it features another key insight: The article features a valuable insight: the slop problem starts inside your existing codebase, not inside the model.
As the article states, AI coding tools were supposed to make teams faster. Instead, many teams watch their codebases swell with brittle, hard-to-maintain code. It compiles. Tests pass. Production incidents follow anyway.
The article directly quotes Larridin’s March 2026 analysis of AI slop:
"AI slop is AI-generated code that compiles, passes tests, and quietly rots your codebase from the inside."
That sentence explains why velocity metrics look great while delivery doesn't improve. The code works today. Tomorrow it resists change.
The article cites a paper posted arXiv in March 2026, titled “An Endless Stream of AI Slop”: The Growing Burden of AI-Assisted Software Development, and adds the author's own insights and findings. Key takeaways:
1. The model learns from your codebase, every single time. Popular stacks like React and Spring Boot win because models have seen those patterns millions of times. Feed an AI assistant an inconsistent system full of historical decisions and private abstractions, and it spends tokens learning instead of solving. Variance rises. Quality drops. Inconsistent patterns teach poor habits, and the model reproduces them at scale.
2. The numbers are worse than the hype suggests. A 2026 study found AI-assisted code carries 1.7 times more issues than human-written equivalents, with technical debt climbing 30 to 41 percent within six months of adoption. New Relic's data shows the gap starkly: 94 percent of leaders call AI code superior at first read, yet 78 percent report more production incidents tied to it. Reviewers can't see the rot. Production reveals it.
3. Rewrites should target AI compatibility, not just modern frameworks. The article's core prescription: a rewrite is an opportunity to rebuild around clear, consistent patterns that play to AI's strengths instead of fighting them. Teams that standardize gain speed and quality at once. Teams that don't pay twice: first in prompting overhead, then in maintenance load.
The article notes that some engineers point to tooling as the answer, to include our AI Slop Index, which scans for duplication, architectural drift and durability signals. This tool can be used to treat symptoms, but it can also be used to identify “sick” codebases that need treatment before they can be used with AI.
And the article fails to discuss a related point. Humans are just as vulnerable to writing worse code for worse codebases as machines are. In fact, these codebases may have started out small, but flawed; they were extended poorly in part because the original flaws showed flawed patterns to maintainers, and forced painful choices onto them when they were fixing problems or adding features.
Humans also suffer from the “bad” code base because AI will write sub-par code on top of it, and it’s humans who have to go in and clean up problems when they inevitably arise.
The piece closes with a question every engineering leader should sit with: how consistent are your patterns today? A clean, consistent codebase is becoming a competitive asset. One riddled with special cases and tribal knowledge becomes a liability that grows faster than humans can manage.
The implied solution: identify flawed codebases (the Larridin AI Slop Index helps), then rewrite them. Treating the rewrite as an investment in AI-native architecture helps to justify it. So does the point the article misses: that humans software engineers benefit from the rewrite too.
Read the full article at WebProNews →