AI slop

Fran Soto, an engineer at Amazon, gave the term its sharpest definition in his April 2026 newsletter piece "Developer Taste: Separating Good Code from AI Slop": code that compiles, passes the tests, looks right on the surface, and "makes the next six months of everyone's life harder." The failure is not a broken build. It is broken judgment, and the compiler cannot detect it.

Philippe Ensarguet traced the same pattern into open source governance in February 2026, where the damage is measurable in maintainer time. Because the effort of writing a contribution used to signal that someone understood the project, generated submissions decoupled the two: plausible-looking output with zero comprehension behind it. Ghostty responded with permanent bans on AI-generated code, tldraw auto-closes external PRs, and cURL shut down its bug bounty programme after being swamped by empty submissions. Ensarguet's proposed fix is filtering on demonstrated context rather than on effort: is the submission linked to existing issues, are the tests complete, was the code actually run.

Cobus Greyling frames the same output from the supply side. In his November 2025 argument that software cost is collapsing, slop is what happens when a resource becomes cheap enough to waste, the same way bandwidth and storage did before it.

The unresolved question sits in Soto's own conclusion: when anyone can generate code, knowing which code deserves trust becomes the differentiating skill, and nobody knows how to hire for it.