A prompt is not just a string. In production, it behaves more like code: it changes behavior, affects cost, introduces risk, and can quietly break systems when nobody is watching. This talk shows how to move from ad hoc prompt writing to an engineering approach built on template systems, version control, reviews, testing, and rollback.

Using familiar patterns like Jinja2 templates, reusable fragments, registries, eval suites, and safety rails, you’ll see how to separate prompt logic from data, reduce copy-paste drift, defend against prompt injection, and make prompts portable across models. We’ll connect these ideas to tools and workflows engineers already know, from PRs and CI/CD to observability and A/B testing. By the end, you’ll have a practical playbook for treating prompts like production assets instead of magic strings hidden in your codebase.