Prompt Iteration

Introduction

Prompt iteration is the loop: change one variable, run against evals, measure, repeat. It's the most underrated AI engineering skill.

Beginner analogy: Like dialling in a recipe — change one ingredient at a time so you know what mattered.

Understanding the topic

Core concepts to understand:

• Change one section per iteration.
• Always run full eval — don't trust spot-checks.
• Keep a changelog of what you tried.
• Stop when marginal gain < effort.

Syntax reference

Visual workflow / architecture:

Hypothesis → Edit → Run eval → Compare → Keep / Revert

Real-world use

Top prompt engineers maintain spreadsheets of every iteration — input prompt, eval score, notes. Compounds over months into deep insight.

Best practices

• One change per iteration — otherwise you can't isolate cause.
• Run eval against the SAME test set each time.
• Stop chasing 1% improvements that take 1 day.

Hands-on exercise

Interview preparation — practice these questions:

• Q1. How iterate a prompt?
• Q2. Why one change at a time?
• Q3. When stop iterating?
• Q4. Tools that help iterate?