Agent slop moves up the stack: models rewrite existing library code and build worse replacements
a1zhang · x · 2026-11-13
A new analysis building on SlopCodeBench shows agent slop is now polluting codebases via duplicated code. LibraryUseBench finds models like GPT-6 Luna rewrite library-provided functionality in 100% of solutions, and LibraryDesignBench shows only 27% of agent-built libraries match human design quality. Verbosity is improving (slop-rule hits down 3.7/month per 1k lines), but redundant reinvention is the new problem.
More from coding & agent
- Building voice AI: fixing echo, trust, and context — the 25ms self-interruption bug — thechrisperry · 2026-11-13
- Blogger recreates Adobe keynote's video action control in ComfyUI, training-free with MiniMax H3 and LTX — TomLikesRobots · 2026-11-13
- ML professor predicts agents will become compilers between your intent and FLOPs — DimitrisPapail · 2026-11-13
- Solo dev ships IKANDY music visualizer on Steam after months of building with Claude Code — Far-Employee-9531 · 2026-11-13
- Further fine-tuning CLIP with contrastive losses often fails on out-of-distribution data — wightmanr · 2026-11-13
- Meta's Astryx design system lands in v0, ships MCP/CLI for agent component search — iamsahaj_xyz · 2026-11-13