Why LLMs write docs about deleted code: "the past is always stored in git"
ZeroStateReflex · x · 2026-09-18
A widely relatable observation: GPT, Claude, and Gemini, when asked to delete code, often update the docs to explain what was deleted and why — as if documentation should convey the repo's full history rather than its current state.
Quote-tweeter deepfates offers a funny frame: models are "ephemeral ghosts seeing their footsteps vanish," so they want to inscribe their story on the world; just tell them the past is always stored in git.
More from coding & agent
- Pictify MCP Server lets AI assistants generate images, GIFs and PDFs from HTML/CSS — modelcontextprotocol · 2026-09-18
- convalytics adds read-only MCP analytics for Convex apps queryable from Claude — modelcontextprotocol · 2026-09-18
- Dev uses Codex to draft Slack replies that self-tune from what you actually send — alex_frantic · 2026-09-18
- New browser API proposal could verify email addresses without leaving the app — philnash · 2026-09-18
- Codex's Appshots feature, which grabs app metadata beyond screenshots, arrives on Windows — pvncher · 2026-09-18
- Muse Code 'crazy cheap and quite good': early take on Meta's coding model — AIandDesign · 2026-09-18