Operational, not news-driven
The value is in repeatable work, constraints and evidence—not in recapping model announcements.
AI applied to real trade work.
Experiments, working methods and failure reviews across Codex, WorkBuddy, knowledge bases, file operations and practical automation.
What belongs here
- 01Codex project practice
- 02WorkBuddy and Codex role design
- 03Enterprise knowledge bases
- 04File organization and batch image work
- 05Prompt, model and context choices
- 06Automation failures and lessons
Editorial principles
Sources stay visible
AI output is treated as a draft or transformation with a traceable source and a human review point.
Failures are part of the record
Context loss, unreliable automation and weak prompts are documented when they teach something reusable.
Related notes
What an AI-ready trade knowledge base needs before retrieval
Why inventory, provenance, permissions and duplicate handling come before model selection.
Dividing work across Codex, ChatGPT and WorkBuddy
A task-oriented way to reduce duplicated AI work and keep review checkpoints clear.