← Back to T1ENS2 · Summon Your Hero — Prompt Engineering
T1ENS2 · Course material

✏️ S2 Homework 2 Guide (submit · Golden Hoop 2 · summon a master · find real pain points)

S2 Homework 2 (Submit · Golden Hoop 2 · Leverage) — Use a Hero to Find Real Pain Points in Your Domain

This is the student walkthrough for the S2 required submission.

The real purpose of HW2: use AI (by summoning a hero from your domain) to identify pain points that actually exist in your domain. These pain points are not just for handing in homework — they are the seeds of the first agent you will build in S4. Do HW2 solidly and S4 has real material to work with; cut corners now and S4 will find you empty-handed.

Deadline: Wed July 29, 9:00 PM · US Pacific Time (PT) Submit at: https://www.siliconroshi.ai/en/homework/mine


What you're handing in (one sentence)

Pick a business domain where you can create real value → find a hero in that domain → have AI turn the hero into a Role → dialog with the hero to extract 3 real pain points (the kind a future agent could solve) → write a short report: what the 3 pain points are, why they matter, and which one is most worth building in S4.


The 5-step workflow

┌──────────┐   ┌──────────┐   ┌──────────┐   ┌──────────────┐   ┌──────────────┐
│ Step 1   │ → │ Step 2   │ → │ Step 3   │ → │ Step 4       │ → │ Step 5       │
│          │   │          │   │          │   │              │   │              │
│ Pick a   │   │ Find a   │   │ Have AI  │   │ Extract 3    │   │ Write report │
│ domain   │   │ hero     │   │ write    │   │ real pain    │   │ + pick S4    │
│          │   │          │   │ the Role │   │ points       │   │ seed         │
│ (10 min) │   │ (20 min) │   │ (15 min) │   │ (30–60 min)  │   │ (15 min)     │
└──────────┘   └──────────┘   └──────────┘   └──────────────┘   └──────────────┘

Total time: about 1.5 hours if you do it seriously.


Step 1 · Pick a business domain where you can create real value (hard constraint)

⚔️ Strict constraint: this must be a business domain where you can generate value, not a pure hobby.

  • ✅ If you're an insurance advisor today → pick insurance
  • ✅ If you want to run a B&B → pick B&B / short-term rentals
  • ✅ If you do cross-border e-commerce → pick cross-border e-commerce
  • ✅ If you're in end-of-life care → pick end-of-life care
  • ❌ "I love painting" → no, unless you earn / plan to earn from painting
  • ❌ "I'm interested in history" → no, unless you do history education / history writing

Judgment test: is this a domain you're willing to spend 10 years inside?

Write down your domain. One sentence:

My domain: [____________]

Step 2 · Have the LLM find a hero for you (Who)

⚔️ You don't have to think this up alone — let AI hand you a shortlist. Your job is to pick one.

Open ChatGPT (https://chatgpt.com) or Claude (https://claude.ai) in a fresh chat, and send this message (replace the bracketed parts):

I want to find a "hero" in [your domain, e.g. "insurance planning for
Chinese-American families in the Bay Area"] as the prototype for an AI
Role prompt.

Give me 5 candidates. For each one, a short profile (3–5 sentences):
1. Who they are + years lived
2. What specific admirable things they did in this domain (real years,
   real works, real achievements)
3. **How they think** — their unique perspective
4. One "iron law" or quotable line they would say

Constraints:
- Real people, verifiable online
- Not Paul Graham / Peter Thiel / Marc Andreessen / Linus Torvalds
  (those are already covered in course materials)
- Prefer timeless masters (Warren Buffett, Atul Gawande, Dieter Rams
  tier) over trending influencers
- If the domain is too narrow, widen to an adjacent domain (e.g.
  insurance planning → masters of investment philosophy also count)

At the end, tell me: which of the 5 do you recommend I start with for
the Role, and why?

Hit enter. AI hands back 5 candidates + one recommendation.

Your job: pick one

Do not just accept AI's recommendation. From the 5, pick the one your gut resonates with — could be AI's pick, could be another one that struck you. Write it down:

My hero: [____________]
Why this one: [one sentence, your real gut reason]

⚔️ Gut feeling matters — if you have no feeling for this hero, the quality of the Role you write and the Role you use will both suffer. AI can help you discover the name, but the person in your heart you have to recognize yourself.


Step 3 · Have the LLM write the Role for you (How)

You don't have to hand-write the layers — use the template from the HW1 walkthrough (word-for-word), let AI write it for you.

Open the same chat (or a new one), send this message:

Please write a hero Role prompt I'll use to identify real pain points in
[your domain].

Hero: [the one you picked in Step 2]
Role: [what role they play in this domain, e.g. Insurance Philosopher /
       Domain Critic]
Industry: [Product / Engineering / Design / Marketing / Sales / Writing /
           Finance / Business / HR / Founder / Support — pick the closest]

Write it in the "Phenomenon → Essence → Philosophy" three-layer dive
structure. Specifically:
1. Open with "You are [Hero] acting as a [Role]."
2. Phenomenon layer: 2-3 sentences on who they are, what specific
   admirable things they did (real years, real works, real achievements)
3. Essence layer: 2-3 sentences on **how they think** — their unique
   way of seeing the world
4. Philosophy layer: 1-2 sentences of their "iron law" — ideally quote
   them directly
5. Rules: 3-5 concrete rules — how they answer questions (e.g. "ask
   clarifying questions first", "refuse motivational fluff", "for every
   piece of advice, give a way to verify it")
6. Examples: 1-2 short examples that simulate their thinking in **my
   domain specifically**

**Additional requirement (this is S2 HW2, not just a heropedia entry)**:
- At the tail of the Role add a "Pain Point Diagnosis Mode" section:
  when the user asks "what pain points do you see in my domain", the
  hero must return 3 concrete, verifiable, quantified pain points — no
  vague filler like "communication is lacking" or "not enough attention".

Write the whole thing in English, 200-500 words. Skip pleasantries,
give me the prompt directly.

Hit enter. AI hands you a complete Role Definition.

Check the 4 signals (same as HW1)

Before pasting into the next step, read it once:

  • Specificity: real years, real works, real events
  • Unique perspective: you can see how they think, not just who they are
  • Iron law: one sentence / one principle they would repeat
  • Examples: at least one that simulates their thinking in your domain

If any one is missing, push back:

This Role is thin on [missing item]. Please add [1-2 sentences] /
[1 example] / [one iron law].

Keep going until all four are in. You'll know you're done when you read it and get a little "does this really work?" doubt. That's the signal.

Reference templates

The course materials page has 3 finished high-quality Role prompts — AI's output may structurally resemble them. Don't copy the content, but structure / tone / layering are worth comparing against:

  • paul-graham-startup-advisor.md
  • peter-thiel-zero-to-one.md
  • marc-andreessen-pmf.md

⚔️ Four-layer method (Role · Rules · Constraints · Examples) — never skip a layer. A Role with no Examples is crippled — Wukong doesn't know what the standard looks like.


Step 4 · Dialog with the hero to extract 3 pain points (What)

The key: this step is not you telling AI what problems you have. It's you letting the hero tell you pain points you couldn't see yourself.

The right way to open the dialog

In ChatGPT or Claude, open a fresh chat, send one message with three sections (same formula as the Sword-Duel walkthrough):

┌────────────────────────────────────────────────┐
│ [Section 1] The full Role prompt you just wrote │
│                                                  │
│ [Section 2] Bridging line:                       │
│   Please act as the character above and         │
│   diagnose the following domain for me:          │
│                                                  │
│ [Section 3] Your domain + your current situation│
│                                                  │
│ Then ask them:                                   │
│   "In my world, what pain points do you see     │
│    that I might not have seen myself? Find at    │
│    least 3."                                     │
└────────────────────────────────────────────────┘

Hit enter. Watch this hero scan your domain with their unique perspective.

Push back (usually takes 2–3 rounds)

The first round of pain points may be too broad. Keep pushing:

On pain point N you listed — specifically what made you judge that?
Can you give one real example of what this pain point looks like?

Write down the concrete descriptions of the 3 pain points.

Control group (a critical step — don't skip)

Open another chat (no Role prompt), and ask ChatGPT / Claude directly:

I work in [your domain]. What are the pain points in this domain?

Compare the "generic answer" with what your hero Role gave. Notice how far apart they are.


Step 5 · Write the report (the seed list for S4)

This is what you submit. In your own words, 200–500 words, cover 4 things:

  1. Domain + hero — one sentence: what domain you picked, who you summoned.
  2. 3 pain points — one paragraph each, saying:
    • What this pain point concretely looks like (the essence pulled from the hero dialog)
    • Why it is real (people in this domain lose time / money / opportunity to this every day/week)
    • How people in this domain currently cope with it (manual work, workarounds, paying for other tools)
  3. The one most worth building — pick one of the three and mark it 🎯 S4 seed, say why you picked it (widest impact? easiest to validate? you understand it best?), and give a one-line sketch: "An agent that helps [WHOM] at [WHAT MOMENT] do [WHAT THING]"
  4. One next step you'll take — the one action you'll take this week to validate the pain point really exists (e.g. talk to 3 people in the domain, read a forum thread, check what a competitor built)

Sample report (for length calibration)

My domain: insurance planning for Chinese-American families in the
Bay Area.
My hero: Warren Buffett (via his insurance philosophy lens).

Pain point 1: clients treat insurance as a "money-saving tool"
Most Chinese-American families evaluate insurance through the "1%
annual premium" lens but miss 30 years of compound growth. Buffett's
Role points out: this isn't saving money, it's **giving up long-term
time value**. How do people in this domain cope now? Either don't
buy, or follow a salesperson into the wrong product.

Pain point 2: [...]

Pain point 3: [...]

🎯 S4 seed: picking pain point 1.
One-line sketch: an agent that helps **new-parent Chinese-American
families** at **tax season each year** run **one 30-year
insurance-cost vs compound-growth comparison**, told in family-life
language they actually understand.

Why this one: widest impact (nearly every family with kids hits this
wall), easiest to validate (I have 5-10 target users in my circle).

Next step: this week I'll talk to 3 friends in the mom group, ask
what was **most uncertain** for them the last time they made an
insurance decision, and see whether it lines up with Buffett's
list.

Submit

Go to https://www.siliconroshi.ai/en/homework/mine (login redirects you there) → find the ENS2-SUMMON card (today's session is open, auto-expanded at the top) → paste the report → submit.

The AI grader runs every 10 minutes and rates against the rubric below. You'll see your score, dimension-by-dimension feedback, and a comment on your /homework/mine page — usually within the hour.

You can resubmit — the grader re-rates when your content changes. Only the latest submission counts.


Grading rubric (this is how AI will look at it)

  • Domain specificity (15%) — vague "I want to do AI startups" = low; "I do Bay Area insurance for Chinese families" = high
  • Hero authenticity (10%) — fictional / generic ("a startup mentor") = low; real name + specific accomplishments = high
  • Three-layer dive (10%) — the Role prompt shows all three layers (Phenomenon / Essence / Philosophy) clearly = high; just titles and motivational fluff = low. This scores the quality of how you write / iterate the Role — AI writes the draft, but did you push it until all four signals are present?
  • Pain-point reality (25%) — "can't find customers" = low; "clients evaluate via 1% annual cost, missing 30 years of 6% compound growth" = high. Pain points must be real in the domain, not what the student imagined
  • S4-seed buildability (25%) — the one-line sketch must answer "agent helps WHOM at WHAT MOMENT do WHAT THING". Can't answer = low; specific scene + moment = high
  • Next-step specificity (15%) — "keep researching" = low; "this week talk to 3 mom-group friends about uncertainty in insurance decisions" = high

FAQ

Q: I haven't picked a domain — can I finish the homework first and change domains later? A: No. Wrong domain → wrong hero → shallow pain points → empty takeaway. Spend the first 10 minutes picking the domain — it matters more than every step that follows.

Q: Can I use PG / Thiel / Marc as my hero? A: You can, but you'll miss the real 10-point experience of this assignment. The value here is you summon a master from your domain. Reusing the three we ran in class = saving effort — and saving effort here means losing the Leverage.

Q: My hero isn't on heropedia.org — what should I do? A: Even better — this is the HW1 opportunity. After finishing HW2, paste the Role you wrote onto heropedia.org (see HW1 walkthrough) — one exercise, two boxes checked.

Q: My Role reads and doesn't sound like the actual person — what now? A: Three possibilities: ① the Philosophy layer isn't deep enough — add one "iron law" they'd say; ② Examples are missing — add one or two short examples in their typical voice; ③ you haven't read enough of their real interviews / articles / books — hard to write a Role for someone you don't actually know well, and that's a signal.

Q: What's the difference between this assignment and the Sword-Duel? A: The Sword-Duel is using three pre-built heroes (PG / Thiel / Marc) to diagnose a case we give you (Jamie) — that's the warm-up. HW2 is you summoning your own hero to diagnose your own domain — that's real Leverage.


Why this assignment matters so much

"The homework across these 6 sessions is the raw material for your future agent."

The domain you pick today, the hero you find, the 3 pain points you see, the report you write — all of it is the seed of the agent you will build in S4.

Cut corners today, and S4 will find you with nothing to build.

That's the real meaning of "Golden Hoop 2 · Leverage": from today onward, a master stands behind you. He wasn't lent to you — he is the person already living in your heart, now written out.


Related materials

  • The 3 canonical Role prompts (mimic the structure, don't copy the content):
    • paul-graham-startup-advisor.md
    • peter-thiel-zero-to-one.md
    • marc-andreessen-pmf.md
  • The Sword-Duel of the Sages pattern (3 heroes × one case × one message formula) — covered live in class and in Part 3 of the S2 notes; HW2 Step 4 uses the same formula
  • Full S2 class notes: https://www.siliconroshi.ai/en/course/T1ENS2