Proposal: keep a PROMPT_SOURCE.md to document human direction in AI-built projects
niosurfer · reddit · 2026-09-19
A Reddit post argues that when AI writes much of the code, a GitHub repo no longer reflects a programmer's real contribution — which now lives in defining architecture, setting constraints, catching mistakes, and steering iterations. The proposal: maintain a chronological PROMPTSOURCE.md in AI projects recording each interaction's prompts, follow-ups, corrections, attachments, and results. The code shows what was built; the prompt history shows why and how the human directed the process, serving as both an archive of contribution and a record others can audit. The post also invites debate on whether this is meaningful documentation or glorified logging.
More from coding & agent
- Grok Build tests Remote Control: link your PC via CLI and drive it from your phone — XFreeze · 2026-09-19
- Developer Says AI Models Now Handle Nearly All His Work — He Just Approves — tushaarmehtaa · 2026-09-19
- 8 components of harness engineering: why the system around the model, not the model, makes agents reliable — blaizedsouza · 2026-09-19
- Google's 12-page Agentic Engineering guide lays out a 5-stage pipeline for building agent teams — blaizedsouza · 2026-09-19
- AgentRun: a purpose-built harness for high-frequency repetitive knowledge work — blaizedsouza · 2026-09-19
- Qwen 3.8 27B on a single RTX 5090 builds a full animation using only code — Acceptable-Object390 · 2026-09-19