Every team we work with has a Notion called 'AI Prompts.' Almost none of them get used. The reason isn't quality, most of the prompts are good. The reason is friction: finding the right prompt, formatting the input, and remembering which model to run it on takes longer than just doing the task.
The teams shipping the most output have replaced the library with a small set of orchestrated agents, typically five, each with a single clear job, a typed input form, and a known output format. The prompts still exist; they're just hidden inside the agent.
The five jobs that show up almost universally: research synthesis, competitive teardown, brief-to-creative, copy QA, and meeting-to-action-items. Anyone on the team can call any agent without knowing the prompt.
The retirement list is uncomfortable. Most prompt libraries should be archived, not migrated. The 80/20 of any library is usually 4–6 prompts. Promote those to agents and let the rest rest.
"A library of 200 prompts is a museum. Five agents anyone on the team can call is a workforce."