PM-201c · Module 1
Why a Library
3 min read
An organization using AI at scale has two options: a governed prompt library, or a collection of individually maintained text files scattered across shared drives, Slack messages, and personal notes. The second option looks manageable when there are five prompts. When there are fifty, the problems become visible. The same task has been prompted differently by different team members. The best-performing version exists on one person's laptop. Someone updated a prompt that was working and forgot to tell anyone. The organization has drifted away from quality without knowing it.
The library also solves the reinvention problem. Without a library, similar prompts are written from scratch repeatedly — by different team members, for similar tasks, without benefit of each other's iteration history. The organization accumulates duplicate effort and does not accumulate knowledge. With a library, a new team member working on a sales email prompt finds the existing one, reviews the version history, understands what was tried and why it was changed, and builds on institutional knowledge instead of starting from zero.
Do This
- Store prompts in a version-controlled repository — the same infrastructure as code
- Make the library searchable by function, domain, agent, and use case
- Treat every prompt as a shared asset, not a personal tool
- Establish the library at the beginning of AI adoption, not after the mess exists
Avoid This
- Do not store production prompts in personal files, Slack messages, or email drafts
- Do not allow prompt updates without version tracking — untracked changes cannot be rolled back
- Do not build the library only after experiencing the pain of not having one — the migration cost is real
- Do not assume team members will share prompts voluntarily without a designated home for them