Fixing AI slop with internal review loops: score outputs, auto-redo until they pass

EXM7777 · x · 2026-09-04

The author proposes a system-level fix for AI slop: build quality control inside the agent itself. The workflow: the agent produces output, a second model scores it against a checklist derived from your best work, and anything below the bar never reaches you—the agent revises with the notes and repeats until it passes. The key insight is turning quality into a number, since a vague feeling ("this feels like AI") can't be fixed but a score can. Slop is a systems problem, so fix the system, not the prompt.

Related event: In-Loop Grading Proposed to Curb AI Slop(2 posts)→

Original post →

More from coding & agent

coding & agent channel →