Do We Need to Tell AI the Steps?

📌 Table of Contents
🔹 1) The Real Problem: A Common AI Mistake
🔹 2) The Core Claim: “Do Not Lock the How Too Early”
🔹 3) The Best Workflow: Expand First, Then Converge Together
🔹 4) Practical Prompts + Decision Rules You Can Copy

🔍 Overview
Many people reduce AI performance by giving detailed step-by-step instructions too early. This lesson shows a better approach: first let AI generate multiple “How” options, then you and the AI compare them, refine them, and finally you decide what to adopt.

📖 Lesson Content

1) The Real Problem: A Common AI Mistake

🗣 (S)
I thought AI needs clear instructions. If I tell it every step, it should work better, right?

🎓 (T)
That sounds reasonable, but it often creates a hidden failure. Recently, one “typical mistake” has increased:

People specify the How (the steps) in detail from the beginning.
As a result, AI cannot fully use its strengths: searching widely, being creative, and finding shorter paths.
Even when AI might be stronger than the person in some parts of the work, the person still “orders the steps from above.” Then the result hits a ceiling.

🎓 (T)
The key point is this:
The problem is not “AI is weak.” The problem is that the human’s way of interacting can reduce AI performance.

🗣 (S)
So my detailed instructions can actually make AI worse?

🎓 (T)
Yes. Because your steps may be based on your usual habits, your limited experience, or an old workflow. If you lock the path early, you can prevent AI from exploring better paths.

🎯 TIP: Think of AI like a strong “option generator.” If you force it into one narrow tunnel, you lose the best part of what it can do.

2) The Core Claim: “Do Not Lock the How Too Early”

A. If you lock “How” first, you shrink AI creativity

🗣 (S)
What exactly do you mean by “locking the How”?

🎓 (T)
It means you decide the route before you see alternatives. For example:

“First do A, then do B, then do C, and answer in this format only.”
“Use this method, follow these steps, and do not change them.”

🎓 (T)
When the route is fixed:

AI produces fewer alternative ideas.
AI produces fewer new ideas that go beyond your expectations.
AI becomes trapped in “your normal way,” even when a better way exists.

🗣 (S)
So I am basically putting AI inside my own box.

🎓 (T)
Exactly. And sometimes your box is smaller than you think.

B. The “patient tells the doctor how to do surgery” problem

🗣 (S)
You said something like “patient telling a doctor.” What does that mean?

🎓 (T)
Imagine AI can perform at a “doctor-level” for a specific task:
for example, summarizing large text, generating multiple marketing angles, producing different strategies, or drafting structured documents.

If the human says:
“Cut here, stitch there, use exactly these tools, follow my steps,”
that is like a patient telling a surgeon how to operate.

🎓 (T)
In that situation, the human role should be:

clearly describe the symptoms (current situation, constraints, problems)
clearly define the ideal outcome (goal, success criteria, quality standard)

And the AI role should be:

propose the How (possible methods, trade-offs, best routes)

🗣 (S)
So the person should describe the “problem and goal,” and AI should propose the “treatment plan.”

🎓 (T)
Yes. You do not give up control. You simply move your control to the right place: goal-setting and evaluation, not early step-locking.

C. The “high-performing employee” model

🗣 (S)
But I still feel nervous. If I do not tell the steps, the work might go in a strange direction.

🎓 (T)
That is normal. Here is a helpful comparison:

If you have an excellent employee (or external expert), you often get better results when you:

explain the goal and constraints clearly
let them propose the approach
review and refine together
choose what to adopt

This is often better than controlling every action.

🗣 (S)
So it is not “no direction.” It is “clear goal, open methods.”

🎓 (T)
Perfect.

🎯 TIP: “Control the target, not the path.”

3) The Best Workflow: Expand First, Then Converge Together

🎓 (T)
Now, an important warning: we are not saying “humans should never touch the How.”

We are saying:

Do not fix the How at the start.
First expand options.
Then converge with human judgment.

Let’s break it into a clear workflow.

Step 1: Ask AI to expand “How” into multiple options

🗣 (S)
What should I do first, specifically?

🎓 (T)
Start by asking AI to generate multiple “How” candidates. You want variety. For example:

“Give me 8 different approaches.”
“Include safe, fast, low-cost, high-quality options.”
“Include a creative option and a conservative option.”
“Show pros/cons for each.”

This activates AI’s strengths: wide search, creativity, and alternative routes.

🎯 TIP: If you only ask for one answer, you often get a “typical answer.” If you ask for many, you get a map.

Step 2: Compare options together (this is where the human becomes essential)

🗣 (S)
After AI gives options, what is my job?

🎓 (T)
Your value becomes very clear here. You help select and refine using:

your real-world constraints
your values and priorities
your responsibility (you live with the outcome)

You and AI should compare:

Benefits and drawbacks
Risks and failure points
Required resources (time, money, skills)
Who will do the work (you, AI, team members)
What quality level is needed

🗣 (S)
So AI proposes, but I evaluate.

🎓 (T)
Yes. And you can ask AI to improve the evaluation table too.

Step 3: Converge to the best plan (or a hybrid plan)

🗣 (S)
Is it always one option that wins?

🎓 (T)
Not always. Often the best result is a hybrid:

Option 2 is best for speed
Option 5 is best for quality
So you merge them: “Use the structure of 5, but the speed tactics of 2.”

This is a powerful pattern: expand → compare → hybridize → finalize.

Step 4: Get a concrete output, then iterate from a new angle if needed

🗣 (S)
But what if the output is still not good?

🎓 (T)
Then you repeat the same cycle from a different angle:

Add one missing constraint (budget, tone, audience)
Change the format (bullet list vs. full draft)
Ask for alternative strategies again
Or ask a different AI tool with the same goal, and compare results

The mindset is:
“One output is not the finish line. It is a test result.”

🎯 TIP: Treat AI output like a prototype. Prototype quickly, evaluate, refine.

Step 5: Final decision is human

🗣 (S)
At the end, who decides?

🎓 (T)
A person decides. Because the person holds responsibility and values.
AI can propose and optimize, but the final adoption choice should be human:

Which plan to adopt
Which risk to accept
Which trade-off is worth it

This keeps the best of both worlds: AI power + human judgment.

4) Practical Prompts + Decision Rules You Can Copy

Below are copy-ready prompts you can paste into an AI tool. They are written in simple English so they translate well into other languages.

Prompt Set A: “Expand the How”

“My goal is: ____. My constraints are: ____. Please propose 10 different ways to achieve the goal. For each way, include steps, time estimate, and key risks.”
“Give me 6 approaches: 2 safe, 2 fast, 1 low-cost, 1 creative. Then rank them by impact and effort.”
“Before you answer, list what information you still need. Then give 8 possible plans using reasonable assumptions.”

Prompt Set B: “Compare and Converge”

“Make a comparison table of these options. Columns: speed, cost, difficulty, risk, quality, who should do it.”
“Ask me 5 questions that will help you choose the best option. After I answer, recommend one option and explain why.”
“Create a hybrid plan: combine the top 2 options to reduce risk while keeping high impact.”

Prompt Set C: “Iterate Without Locking”

“Give me a second version that uses a different strategy, not just small edits.”
“Now solve the same goal with a totally different approach. Avoid repeating the previous structure.”
“Generate 3 outputs aimed at different audiences: beginner, intermediate, and expert.”

🛠 Practice

👉 Practice Task 1 (5–10 minutes): Expand the How

Pick a real goal you have this week (work or personal).
Write your goal in one sentence.
Write 3 constraints (time, cost, tools, etc.).
Ask AI: “Propose 10 different ways to achieve this goal. Include pros/cons.”

👉 Practice Task 2 (10–15 minutes): Converge to one plan

Choose the top 3 options.
Ask AI to build a comparison table.
Decide which option you adopt (or create a hybrid).
Ask AI to produce the final output using your chosen plan.

👉 Practice Task 3 (Optional): Cross-check with a second AI

Copy the same goal and constraints into a different AI tool.
Compare the “How” lists.
Merge the best parts into your final plan.

🎓 Comprehension Quiz（3 choices）

A. The best way to use AI is to specify the exact steps first, so AI does not get confused.
B. A strong approach is to first let AI propose multiple “How” options, then compare and refine them together before you decide.
C. AI should always make the final decision, because AI has more data than humans.

📌 Summary

Many people reduce AI performance by specifying the How too early.
When you lock the route, AI produces fewer alternatives, fewer new ideas, and fewer “short paths.”
In tasks where AI can perform at a very high level, “human ordering the steps” can resemble a patient instructing a doctor, which often lowers results.
A better workflow is expand first (many How options), then converge (compare, refine, hybridize), then decide (human responsibility).
Iteration is normal: output → evaluate → re-run from a different angle → decide what to adopt.

📝 Quiz Answers

Correct answer: B