Iteration as a Method

L2 Lesson 3 of 5 — Intentional Prompting

1

2

3

4

5

The First Answer Is the Beginning, Not the End

In L1, you learned that AI responses are drafts, not finished products. Now we turn that insight into a method. Instead of hoping for the one perfect prompt, you work in a deliberate cycle: write a prompt, evaluate the output, diagnose the gap, adjust the prompt, repeat.

This works because the AI retains the entire conversation history within its Context Window The span of text an AI model can 'see' at once — your entire conversation history plus the current input. Everything within this window influences the response. . You don’t have to start over each time. Instead, you make targeted adjustments — like a conversation with a colleague who refines your draft step by step.

The Refinement Cycle in 5 Steps

1. Write the prompt — First version with role, context, task, and format
2. Evaluate the output — Is the tone right? The length? Is something missing?
3. Diagnose the gap — What exactly is the problem? (Not just “I don’t like it”)
4. Adjust the prompt — One targeted change that addresses the problem
5. Repeat — Until the result fits

The critical step is number 3: the diagnosis. “It’s bad” doesn’t help you or the AI. “The tone is too formal for our audience” is an instruction the AI can work with.

A Concrete Example: The Project Status Update

Round 1: First Draft

You:

Write an email to my team with a project update.

AI: (writes a generic, formal email with placeholders like “[Project Name]” and “[Milestone]”)

Diagnosis: Too generic, no concrete content, wrong tone. The AI didn’t have enough context.

---

Round 2: Adding Context

You:

The project is the migration of our customer database to a new CRM. We’re 2 weeks behind schedule, but the data migration kicked off yesterday. The team is 8 people from Sales and Engineering. Tone: direct and honest, don’t sugarcoat it. 150 words max.

AI: (writes a more specific email, mentions the delay and current status — but phrases things too cautiously and formally)

Diagnosis: The content is right now. But the tone is too cautious — “we are currently experiencing a delay in our timeline” instead of a straight answer.

---

Round 3: Fixing the Tone

You:

The tone is too corporate. Write it the way I’d say it in a standup. Instead of “we are currently experiencing a delay in our timeline,” just say: “We’re 2 weeks behind.” No buzzwords, no filler.

AI: (writes a noticeably more direct version — clear status, no sugarcoating, concrete next step)

Diagnosis: Tone is right. But it’s missing a clear call-to-action: what should the team do next?

---

Round 4: Final Polish

You:

Good. Add a specific call-to-action at the end: everyone should update their open migration tickets in Jira by Friday.

AI: (delivers the final version — direct, honest, with a clear next step)

Result: After 4 rounds, you have an email you can actually send. Not because the first prompt was perfect, but because you refined systematically.

What Happened in Each Round

Notice: in no round did you rewrite the entire prompt from scratch. Instead, in each round you adjusted one thing:

• Round 2: Added context (narrowed the scope)
• Round 3: Fixed the tone (with a concrete example)
• Round 4: Added an element (call-to-action)

This is the strength of the iteration cycle: each adjustment is small and specific. That makes the results predictable — and with each round, you learn what the AI needs.

Iteration: Do’s and Don’ts

Do

• Name the gap: 'The tone is too formal' instead of 'Make it better'
• Change one thing per round — tone, length, content, or structure
• Give concrete examples: 'Write it like a standup update, not a board report'
• Use the conversation history: you don't need to repeat everything

Don't

• Rewrite the entire prompt after a bad response
• Request multiple changes at once and hope everything lands
• Give vague feedback: 'I don't like it' without saying why
• Start a new chat just because the first answer wasn't perfect

Try It

Exercise 1: The 4-Round Test

Write a piece of text you actually need — an email, a meeting summary, a presentation intro. Start deliberately with a simple prompt. Refine in exactly 4 rounds. Observe how the output changes with each iteration.

Exercise 2: Practice Diagnosing

Ask the AI for a product description of something from your daily work. Evaluate the output against this checklist: Tone right? Length appropriate? Audience matched? Content complete? For each shortcoming, write a targeted correction.

Exercise 3: Change Only One Thing

Take an AI response that’s “okay but not good.” Identify the one thing that would make the biggest difference. Change only that. Repeat three times.

Think Further

Iteration is not a workaround for bad prompts — it is the method. Even experienced users work in cycles, because the best formulation often only emerges through dialogue.

In the next lesson, you’ll see how different task types call for different prompt strategies — from writing and analysis to brainstorming and formulas.