11 Managing prompts in production

The chapter introduces prompt management as the practical discipline that keeps prompts reliable after they are written. It argues that prompts should be treated like production software artifacts, not disposable text, because they can sprawl across tools, lose version history, and become hard to update, debug, or trust. Through examples of duplicated prompts, missing ownership, and slow incident response, it shows how unmanaged prompts create confusion and failures that are costly to trace.

To address this, the chapter presents a maturity journey and a structured lifecycle for prompts. Teams may begin with inline prompts, then move to version-controlled files, a shared prompt store, and eventually a dedicated prompt management system as needs grow. Across these stages, the same core practices matter: clear metadata, versioning, testing, approvals, deployment through aliases, observation of usage and change patterns, deliberate improvement, and explicit retirement. The chapter emphasizes that stronger management makes good process easier and makes bad process visible.

It also explains how lightweight governance and collaboration prevent prompt drift without creating unnecessary friction. Roles, approval paths, audit trails, and rollback procedures help teams change prompts safely while still allowing domain experts and other nondevelopers to contribute. Finally, the chapter shows how to read management signals such as edit frequency, reviewer activity, test coverage, and usage patterns to decide what needs attention next, and it closes by cataloging common failure modes like prompt sprawl, blind updates, metadata gaps, shadow edits, and overcomplicated all-in-one prompts, along with concrete fixes for each.