prompt versioning
Treating prompts as code: kept in the repository, given a version that travels with each request, reviewed and tested before release.
In more detail
A changed prompt needs a new version, so logs and evals can tell which wording produced a reply. Review the eval results alongside the text diff, pin the model snapshot, and roll changes out behind a feature flag.
Related terms
- golden set — A curated set of inputs with known good outcomes, used to check that an AI feature still behaves after every change.
- model snapshot (LLM) — A fixed, dated version of a hosted model, as opposed to an alias that the provider moves to newer versions.
Taught in
- Structured Output — AI Engineering: LLM Apps, RAG & Agents
- Prompts As Code — AI Engineering: LLM Apps, RAG & Agents
- Regression Tests In CI — AI Engineering: LLM Apps, RAG & Agents