Stage 2 · Lesson 5 of 9
Prompts in version control
A prompt is code: changing it changes how the product behaves. So keep , in with a review, not pasted into a dashboard or hidden in a string nobody can find. Give each prompt its own file or named constant.
Record which model and settings it was written for. Log the prompt version with every request, so you can any output back to the exact text that produced it. Before merging a change, rerun the same fixed set of inputs (ten is enough to start) and compare old and new outputs side by side. This habit grows into the suite two stages later.
A question to check yourself, an AI practice session and 1 cheat sheet entry come with the full roadmap. Unlock
Resources for this stage
Start with· follow one of these
- CoursePaidGrokking Prompt Engineering for Software Engineers
Design Gurus · Beginner · Subscription
Prompting patterns for engineers building features, not for chatting
- GitHub repoFreeAnthropic Courses
Anthropic · Beginner
Free notebooks: API fundamentals, then the interactive prompt engineering tutorial
Practise
Keep handy
Also worth your time
Few-shot, chain of thought, JSON enforcement and parsing failures in one place