AI coding bloats codebases: models prefer new logic over extending existing code
morbidflood · reddit · 2026-09-16
A developer using both Claude and Codex at work observes that models prefer implementing new logic rather than extending existing facilities, causing codebase bloat under incremental use. They also note most benchmarks only evaluate greenfield projects, ignoring editing and incremental development, and smaller models are even worse at editing than implementing. They ask the community for workflow improvements.
More from coding & agent
- Telling your coding agent to slow down actually helps, developer reports — nilogram · 2026-09-16
- The Hacker's Guide to Attacking AI Agents: a systematic look at agent attack surfaces — _clickfix_ · 2026-09-16
- vAIvar: open-source honeypot feeds attacking AI agents endless hallucinations — Scobleizer · 2026-09-16
- Polyphonic details: auto-imported skill libraries, tag-a-into-chat agents, early Mnemos memory system — RileyRalmuto · 2026-09-16
- Polyphonic beta: a Mac 'home' where Claude Code, Codex and other agents share memory and collaborate — RileyRalmuto · 2026-09-16
- Together AI CPO cuts PRDs to 2 pages plus a prototype to fix "context window explosions" — aakashgupta · 2026-09-16