@rafa
The NPC Guide to Enterprise System Prompt Design
“Prompting by individuals fails at scale because it cannot reliably encode organizational intent, authority boundaries, and risk posture across thousands of users, teams, and contexts. Individual prompts also do not compose. They fragment behavior, create inconsistent commitments, and make outputs difficult to audit. In practice, a large enterprise needs their global LLM system to exhibit stable behaviors under recurring conditions: how to speak about products, what language is permitted for commitments, when to escalate, how to recognize and react to sensitive data, what sources are authoritative, and how to communicate uncertainty. Those behaviors must be centrally governed and versioned, even if local teams can add overlays for their context.”
https://open.substack.com/pub/npcmemo/p/the-npc-enterprise-playbook-for-master