A municipal AI platform wants every department chatbot to share the same role framing, citation rules, and safety constraints without each team hard-coding prompts in Lambda. What approach best establishes that control?
Select an answer to reveal the explanation.
Short Explanation
City hall does not let every desk invent its own letterhead—shared templates keep the voice consistent. Prompt management is that letterhead for GenAI: role, constraints, and citation rules live in one governed place. Apps pull the approved template instead of reinventing the system message in every Lambda.
Full Explanation
A municipal AI platform that wants every department chatbot to share role framing, citation rules, and safety constraints needs a governed distribution channel for prompts—not ad hoc string pasting. Centralizing versioned prompt templates in a prompt management service lets applications pull the same role, constraint, and citation instructions at runtime, supports change control, and reduces drift across permitting, parks, and 311 bots. Teams stop hard-coding divergent system prompts in Lambda and instead reference approved template versions with parameters for department-specific details. Asking departments to paste preferred wording into Slack before go-live is informal and non-auditable; Slack history is not a prompt registry. Storing one unmanaged free-text system prompt only inside each frontend JavaScript bundle exposes instructions to clients, makes updates inconsistent, and bypasses server-side governance. Disabling prompts and relying solely on the foundation model’s default pretraining behavior abandons municipal voice, citation policy, and safety framing entirely. Exam caveat: when stems emphasize shared role, citation, and safety without Lambda hard-coding, the intended control is centralized prompt management—not chat channels, client bundles, or “no prompt.” Operational check: publish templates with owners and versions, require apps to fetch by template ID, gate production promotions through review, and periodically diff live bot behavior against the approved template set for each department.