Dev cleaning up AI slop: no model, local or frontier, escapes mixed-quality output
zeeg · x · 2026-09-12
Developer zeeg says he's unwinding a pile of AI-generated "slop" code produced with tools including Sol, Fable, Astra and Opus — and argues none of them, nor local models, avoid the problem.
- There is no "magical software factory": the tech can be useful and frustrating at the same time
- Anyone preaching otherwise, he says, will show tirefire production software: poor performance, bugs, bad design
- He admits his own code isn't exempt despite hard effort
A sober front-line reality check against AI coding hype.
More from coding & agent
- Anthropic engineer: Claude-written production code should meet a higher bar — cto_junior · 2026-09-12
- Steve Klabnik's ERSC argues Git isn't the future of source control in the agent era — rseroter · 2026-09-12
- Yutori's Navigator n2 computer-use model drives your Mac locally via MCP — DhruvBatra_ · 2026-09-12
- Asking agents to draw portraits exposes huge gaps in fine-grained mouse control — xwang_lk · 2026-09-12
- Agent Memory Doesn't Auto-Survive Model Upgrades: Free-Form Notes Drop 13 Points — rohanpaul_ai · 2026-09-12
- YC-backed Fabraix hits $1.04M ARR in 3 months with autonomous red-teaming agents — ycombinator · 2026-09-12