This is the student playbook for the S6 required homework.
The real point of this homework: in S5 you forged one cut. S6 is not more polishing — S6 is publishing your agent, sharing it, letting a real person use it once, and then interviewing them.
An agent is not a document, not a script — it is a product meant to be used.
Deadline: Sat 9/2 · 9:00 AM PT · AI-graded
Submit at: https://www.siliconroshi.ai/homework (after login, auto-redirects to /homework/mine)
Course ID:
ENS6-SHIP
PUBLISH your agent to WukongGPT, SHARE the link with at least 1 real person, interview that person, and write a short user-feedback report.
┌─────────────────────┐ ┌─────────────────────┐
│ Step 1 │ → │ Step 2 │
│ │ │ │
│ Publish + share + │ │ Interview + │
│ let a real person │ │ write a short │
│ use it │ │ report │
│ │ │ │
│ Publish · Share │ │ hear a real user · │
│ (30 min) │ │ write it down (30m) │
└─────────────────────┘ └─────────────────────┘
Total time: about 1 hour (not counting the wait for your user to reply).
Open WukongDojo Builder → your agent → Publish.
Choose visibility:
After publishing you get a share link.
Send the link to at least 1 real person. It can be:
⚔️ Hard constraint: that person cannot be you. Self-testing ≠ a real user.
Don't intervene, don't coach. Send the link and let them run it themselves.
💡 Watching over their shoulder is fine — if it's in person, notice where they get stuck, where they smile, where they frown. That's more honest than what they say.
After they've used it, interview them. Here's the interview template (use it as-is):
1. What did you use this agent for? (let them describe in their own words)
2. Was there a moment you thought "oh, it gets me"? Which line?
3. Was there a moment you thought "it's rambling / dodging the question"? Which line?
4. If you'd use it again, for what? If not, why not?
5. Describe this agent in one word.
💡 Record exact quotes. Don't summarize for them, don't polish it. If they say "it was okay I guess," write "it was okay I guess."
Then write a short report in the following three sections:
Who they are: __________ (no real name needed — relationship + background, e.g. "a friend in accounting")
How I sent it: ________ (WeChat / email / in person / other)
How long they used it: __________ (roughly)
What they used the agent for: __________________________________
Q1 · What did you use it for?
→ "______________________________________"
Q2 · Any "it gets me" moment?
→ "______________________________________"
Q3 · Any "it's rambling" moment?
→ "______________________________________"
Q4 · Would you use it again? For what / why not?
→ "______________________________________"
Q5 · One word for this agent?
→ "______________________________________"
The one thing that surprised me most: ________________________________
(where the user reacted differently than you expected)
What I learned: ________________________________________
(about your agent · about your domain · about "being used")
What I'll change next: ____________________________________
(specific — which place · not "keep polishing")
[ ] Agent is Published to WukongGPT
[ ] Link sent to at least 1 real person
[ ] That person actually used it (not me self-testing)
[ ] Interviewed that person (asked all 5 questions)
[ ] Section 1 · wrote user background + how they used it
[ ] Section 2 · pasted the interview quotes
[ ] Section 3 · wrote surprising finding + what I learned + next step
Submit at: https://www.siliconroshi.ai/homework
Paste the three sections above:
Course ID: ENS6-SHIP
This S6 homework doesn't survive AI ghost-writing — because what it asks for is the exact words a real person said after using your agent. If you never sent the link and never interviewed anyone, AI can't fake that "it was okay I guess" / "how did it know what I was thinking" realness.
AI can polish your report · but it can't invent a real user's reaction.
—— An agent is not a document — it is a product meant to be used.
Section 1 · Who + how
Who: my cousin, does finance, 40+, doesn't know AI, never used ChatGPT
How I sent it: WeChat link
How long: about 12 minutes
What they did: asked three questions about what to do with $300K in savings
Section 2 · What she said
Q1 · What did you use it for?
→ "I just asked whether it's worth leaving that money in the bank"
Q2 · Any "it gets me" moment?
→ "The part where it asked me to guess the 30-year multiple was fun — not like
those that dump a pile of formulas on you"
Q3 · Any "it's rambling" moment?
→ "The risk-tolerance section had too many terms, I didn't quite follow"
Q4 · Would you use it again?
→ "If it could remember what we talked about last time, maybe. Re-explaining
everything each time is a hassle"
Q5 · One word?
→ "Patient"
Section 3 · What I learned + next step
Most surprising: she didn't care about 7% vs 5% — she cared about "can it remember
me." I spent a week on pushback-bank, but her top need was Memory.
What I learned: a real user's first reaction was completely different from my design
focus. I thought the key was handling pushback; she thought the key was whether it
could remember her like a person.
Next step: replace the jargon in the risk-tolerance section with plain language. Add
Memory support the moment WukongDojo ships it.
One last reminder: S6 is not writing a report — it's letting your agent be used by a real person once.
Publish, share, interview, write it down — your agent is no longer yours alone.