What is AI slop?

AI slop is low-quality output generated by AI tools and shipped with little human review. In software it shows up as plausible but wrong code: hallucinated APIs, tests that assert nothing, and security checks that look correct but do not hold.

Origin of the term

The word slop started as a label for low-effort content produced by generative tools and posted at scale. As coding agents spread, the same word moved to software. AI slop in code is output that reads well and compiles, yet does not do what it claims or hold up under review.

What AI slop looks like in code

Slop is convincing on the surface. The problems sit one layer down.

• A call to a package or function that does not exist, invented because it sounds plausible.
• A test that runs green while asserting nothing meaningful, or that mirrors the bug it should catch.
• A security check that appears to validate input but can be bypassed.
• A refactor that drifts past the request and edits unrelated files.

Why it happens

Slop is the byproduct of speed without a review gate. It rises when an agent lacks context about the codebase, when the model states a wrong answer with full confidence, and when no one reads the diff before merge. It is common in vibe coding, where output is accepted with little manual review.

What it costs

• Reviewer time spent untangling code that looks finished but is not.
• CI minutes burned on pull requests that were never viable.
• Security risk that sits latent until something fails in production.
• Eroded trust in AI coding tools across the team.

How to detect and stop AI slop

Slop is caught the same way other defects are caught, earlier and automatically. Automated code review reads each change against the conventions of the surrounding code. Tests act as a gate, so a change cannot merge unless it passes. AI code review reasons about intent across files and fixes the issue rather than only flagging it. A person still approves the merge.