Vibe-Coding
a Business.
Workshop Workbook · Jason Arbon · Testers.AI
Name
Pod #
My idea
AI is your co-founder, not your oracle. It produces variants at superhuman speed. It has zero judgment about which one is right. Your taste is the moat.
AI flattery dressed up as signal. "Compelling" means nothing. Demand scores against a rubric.
"ChatGPT loved my idea." ChatGPT loves every idea. Validate with people who'd pay.
10× output ≠ 10× value. AI-generated marketing that looks right but sounds like everyone else.
AI is great at the middle of the bell curve. Push to the edges. Outputs without taste = noise.
| # | Stage | Artifact you walk out with | Time |
|---|---|---|---|
| 01 | Ideation | 2 ideas that survived stress-testing + obituary for each | 20 min |
| 02 | Product / Service | Animated HTML demo of the 30-sec wow | 20 min |
| 03 | Positioning | One sentence that displaces an alternative | 20 min |
| 04 | Pitching | Simple Khosla seed deck + 60-second narrative | 20 min |
| 05 | Go-to-Market | First buyers + the channel to reach them + a price | 20 min |
| 06 | Validation | Synthetic-user run + 1 real user + the #1 fix | 20 min |
For Stage 2, build a landing page or small app. Two easy options — your chatbot is the zero-setup pick.
| Tool | Best for | How to load & free tier |
|---|---|---|
| Your chatbot ★ | Single animated HTML page, zero setup. The easiest path — and works offline-ish. | Open ChatGPT / Claude → "Build me a single HTML page that animates…" It renders right in the chat. |
| Lovable | End-to-end app with Stripe wired in. Best after the workshop. | Go to lovable.dev → sign in → build. Quick signup, but only a couple iterations before the paywall. |
Generate cheap, stress-test brutally, kill 90%. Walk out with 2 ideas that survived and an obituary for each.
You are a serial founder who has killed more ideas than you've shipped. I want to build in [domain]. My unfair advantage: [skill/access]. Hair-on-fire problem or opportunity: [buyer intuition]. 1. Generate 20 angles. Be specific. 2. Score 1-10 on: wedge sharpness, buyer urgency, AI defensibility, distribution unfair advantage. 3. Kill 18. Defend 2. 4. For the 2, write the OBITUARY — what kills this company in 18 months? Refuse to say "interesting" or "promising". Use scores.
Ask AI for a single HTML page that animates your wow moment — then iterate on it. HTML is how you actually iterate with AI (you see and feel it instantly), not Docs or PPT. Out of ideas? Just tell the AI "make it better."
Build a SINGLE HTML page that ANIMATES a demo of my product's wow moment. Product: [one-line description] The "wow" moment: [what the user sees in 30s] - One file, opens in any browser, no auth, no build - Real-ish data, not lorem ipsum - ANIMATE the magic: things move, update, react live - Add a control or two so I can poke at it Then I iterate. When I run out of ideas, I just say "make it better" and react to what you build. BONUS — a fun logo: "Design a fun logo for [product]: flat vector mascot, bold simple shapes. 3 options, each as a transparent-background PNG + a square avatar."
One sentence that displaces a specific alternative, names the failure mode in buyer's words, makes the buyer feel seen. Generate 20, keep 1.
Product: [one sentence] ICP: [role + stage + context] Alternatives they use today: [3 things] Failure mode I solve: [specific pain] Generate 20 positioning statements. Each must: - Name a specific alternative I displace - Name the failure mode in MY BUYER'S WORDS - Pass "would they email this to a colleague" test BANNED: seamless, revolutionary, next-gen, leverage, ecosystem, AI-powered, cutting-edge, transform. Score 1-10 on: specificity, urgency, memorability. Show top 5 with scores. Then pick THE winner.
A simple Khosla seed-deck + 60-second spoken narrative. AI builds the scaffold; you write the conviction. A pitch is about de-risking — why now, why you, why this wins.
You are a Khosla Ventures seed partner. Zero patience for buzzwords. Positioning: [winning sentence] Audience: [investor / customer / partner] Traction (anything, even tiny): [one number] Unfair advantage: [why us] Build a simple seed deck in this order: 1. What we do · 2. Problem (who's bleeding) 3. Solution + magic · 4. Why now 5. Market (bottoms-up) · 6. Why we win (moat) 7. Business model · 8. Go-to-market 9. Traction & milestones · 10. Team + the ask For each slide: 5 headline options, 3 bullets max, and the ONE RISK it must de-risk. Then: a 2-SENTENCE elevator pitch — and give me 10 VARIATIONS of it, different angles + tones. I'll pick one. Then: 60-second narrative in MY voice. Then: SKEPTIC PASS as a tired VC — name the 3 weakest slides. BONUS: Which VC firms + partners fit my stage and space? Re-tailor the deck to the top firm's thesis, then draft the intro email to that partner.
Who are your most likely first buyers, how do you reach them, and what do you charge? Find the wedge of early users, the one channel they already live in, and a price they'll actually pay.
My product: [one sentence] The wedge buyer: [who bleeds most] 1. FIRST BUYERS — my 10 most likely first buyers: segment, role, and the trigger that makes them need this THIS quarter. 2. CHANNEL — the ONE place they already gather. Rank channels by how fast I reach them with $0. 3. PRICE — 3 pricing options + the number they'd actually pay, anchored to the value/time saved. 4. REACH-OUT — a 60-word note opening on THEIR pain, not my product. One 5-min ask. 5. LAUNCH + ENGINE — one shareable launch moment and a weekly motion I can sustain solo. BANNED in the note: "I came across", "I noticed", "circle back", "synergies", "leverage".
Pro move (Testers.AI playbook): Run a first buyer's product through Testers.AI synthetic users — the bug report is the reach-out. Opener: "Found this while testing your onboarding — figured you'd want to know."
Run 20 synthetic users against your live product. Find #1 bug. Fix it. Re-run. Then talk to one real person. This is the stage most AI-built startups skip. Don't.
You are a synthetic user research panel. Simulate 20 distinct users from this ICP attempting this job end-to-end. Product URL: [url] ICP: [specific buyer] Job-to-be-done: [what they're trying to accomplish] Price to test: $[X]/mo For each synthetic user output: - Persona (one line) - First 30 sec: what they see / think - First action they take - Where they get confused - When they'd drop OR have "wow" moment - What they'd tell a colleague (verbatim) - Would they pay? Why / why not? After 20 runs, aggregate: - Top 3 onboarding bugs - Top 3 positioning bugs - SINGLE biggest fix - 3 pricing reactions worth A/B testing
Then: ONE real user. Send the URL to one real ICP person. Watch them use it on Zoom — don't help. Note every confusion silently. If real friction matches synthetic → trust the loop. If not → the personas are wrong.