First Draft, Then Refine

Accept the first answer?   vs.   Iterate until it fits.

The first response is rarely the best one.

The refinement loop:

Prompt → Response → Evaluate → Refine → (back to Prompt)

Read the Response — Then Choose

Decision	When to use it	What you do
Accept	Complete and usable	Ship it
Refine	Close but off	Rewrite prompt with one fix: +context | +example | +tighter format
Follow-up	Good, needs one more piece	Add a new turn

Stop when you can use it without translating it.

Example: BioTech CEO Board Summary

Turn 1 — Vague prompt	Turn 2 — Refined prompt
"Summarize our Q3 results."	"Summarize Q3 results [+audience] for our board, executives non-technical; [+format] 3 bullets + 1 risk; [+tone] confident, plain English."
Generic wall-of-text. (grey, unfocused)	Crisp, board-ready brief.

Three targeted fixes. One transformed output.

Your Turn — Manufacturing Engineer

Prompt: "List common causes of mold flash defects."
AI returned: generic 8-item textbook list.

Accept, Refine, or Follow-up?

Reveal — Refine (add context):

"List likely causes of mold flash on a 200-ton injection molder running polypropylene (PP) for thin-wall housings — defect rate 3% on the parting line. Rank by likelihood."

Recap — You've Got the Loop

1. Read first — decide Accept / Refine / Follow-up
2. One fix at a time — context, example, or format
3. Stop when you can use it without rewriting

LO-S2-2 Apply    LO-S2-3 Evaluate

Take the Section 2 quiz.