The Diagnose-Change-Recheck Iteration Loop
CoreIterate deliberately to improve Claude's output · Difficulty 1/5
Explanation
Prompting Is Iterative
When a first response misses, the skill is diagnosing why and changing one thing deliberately -- not simply asking again and hoping for a better roll.
The Loop
- Read the output against your intent. What exactly is wrong -- wrong format, missing detail, wrong tone, a factual error, too long?
- Make a targeted change. Add the missing constraint, supply an example, tighten the audience, or split the task into steps. Change one variable so you can tell what helped.
- Re-run and compare. Keep the better version; repeat the loop if there's still a gap.
Why This Is the Same as the Diligence Competency
This loop mirrors the *Diligence* competency in AI Fluency: you stay responsible for judging and refining the result rather than treating the first output as final or outsourcing that judgment back to the model. Iteration is not "regenerate until it looks right" -- it is a controlled edit-and-recheck cycle where you drive the change and you judge the outcome.
Common exam traps
- Repeatedly hitting "regenerate" without changing the prompt. Randomly re-rolling the same prompt is not iteration -- it doesn't diagnose or fix the actual gap, it just resamples.
- Changing five things at once. If you rewrite the tone, format, length, and constraints all in the same edit, you can't tell which change fixed (or broke) the output. Isolate one variable per iteration.
Key Takeaways
- Iteration means diagnosing a specific gap and making a targeted change, then re-checking
- Change one variable at a time so you can attribute the result to that change
- This loop mirrors the Diligence competency in AI Fluency -- you stay responsible for judging the result
- Regenerating without changing the prompt is not iteration
- Changing multiple things at once makes it impossible to tell what helped or hurt
Glossary Terms
Anthropic's framework for effective, responsible human-AI collaboration, built around competencies including Delegation (deciding what to hand to Claude versus handle yourself), Description (framing work clearly enough to act on), and Diligence (using AI effectively, ethically, and safely, with the human owning the outcome).
A Claude Code workflow pattern that builds solutions incrementally through small, verifiable steps rather than attempting complete implementation in one pass. Each step produces testable output; failures are caught early. Pair with test-driven iteration for maximum reliability.
Related Concepts