AICon USA · Tutorial · Half-Day Workshop
Vibe-Coding
a Business.
Using AI to rapidly build, test, and validate real products — from raw idea to go-to-market in days, not months.
Jason Arbon · Founder & CEO, Testers.AI
Before anything else · do this now
Get connected now.
- 1 · Join the wifi — and note your phone hotspot as backup
- 2 · Open claude.ai or chatgpt.com — send one message right now
- 3 · Grab the kit ↓
testers.ai/vibe-biz
Conference wifi dies. A live chat session survives. Do it while I talk.
Your toolkit for today
testers.ai/vibe-biz
- The skill — walks you through all 6 stages inside your chatbot
- The workbook — every prompt, with space to fill in as you go
- The on-rails tutorial — two simple businesses, every prompt pre-filled, one-click copy
- The starter kit — 12 ideas + 3 worked examples to fork
"In my little group chat with my tech CEO friends, there's this betting pool for the first year that there is a one-person billion-dollar company. Which would have been unimaginable without AI — and now will happen."
— Sam Altman
Who's running this
Founder, Testers.AI.
The big thing: I've watched AI rewrite how you start a business.
- Test.ai — AI-first, ~$30M (mostly Google). Customers: Google, Microsoft, Epic / Fortnite. ~40 people.
- Google & Bing — AI early; personalized Google results
- Chrome — test manager · + earlier startups
- How Google Tests Software — co-author
The shift · De-Risking the Business
The compression is real.
2019 · idea → MVP · engineers, capital, luck
6 months
2023 · idea → MVP · no-code + Copilot
3 weeks
2026 · idea → operating · one person, one weekend
3 days
The bottleneck isn't building anymore. It's knowing what to build, whether it's actually good, and whether anyone wants it.
Definitions
"Vibe-coding a business"
is not what you think.
It is NOT
- Letting AI hallucinate a startup
- Skipping customer conversations
- Replacing taste with output volume
It IS
- AI as a force multiplier on judgment
- Treating every artifact as a hypothesis
- Founder taste + AI velocity
Before we start
You're probably skeptical.
Good.
"AI STARTUPS ARE SLOP."
"I'M NOT A FOUNDER."
"I COULD JUST GOOGLE THESE PROMPTS."
The framework
The 6 stages we'll vibe-code today.
STAGE 01
Ideation
STAGE 02
Product / Service
STAGE 03
Positioning
STAGE 04
Pitching
STAGE 05
Go-to-Market
STAGE 06
Validation
Why most people fail at this
Four traps that kill vibe-coded startups.
False confidence
Shallow validation
Slop velocity
No founder taste
What you walk out with
By the end of today, you have artifacts.
- A stress-tested idea
- A working prototype or demo
- Landing-page positioning
- A pitch deck + 60-sec narrative
- A go-to-market plan: buyers, channel, price
- A synthetic-user validation run
The rhythm
How today works.
- Pods of 4 — your feedback loop
- Watch · build · share
- 6 stages, 6 artifacts
- Timers guide, not judge
- Stuck? Wave me over
- Kill ideas freely
The shape of the day
Today, time-boxed.
I demo · pitch window
you build · work time
◇ break / derby
Open
→
▶demo
Ideation
20 min
▶demo
Product
20 min
▶demo
Positioning
20 min
☕ Break 10 min
→
▶demo
Pitching
20 min
▶demo
Go-to-Market
20 min
▶demo
Validation
20 min
Demo
Derby
→
Derby
Close
Every stage: watch a demo → build yours — each block is 20 min. We flex with the room.
// STAGE 01
Ideation.
Generate cheap, stress-test brutally, kill 90%. The goal isn't a good idea — it's a differentiated wedge with a buyer in pain.
The pattern
Don't ask AI for ideas.
Ask it to attack them.
- 20 angles, scored 1–10
- AI plays skeptical investor + churning buyer
- Keep the 2 that survive
# Stress-test my ideas You are a brutal serial founder. I want to build in [domain]. My edge: [skill/access]. 1. Generate 20 angles. 2. Score each 1–10: wedge, buyer urgency, defensibility. 3. Kill 18. Keep 2. 4. For those 2 — what kills them?
Story · How Testers.AI got chosen
I had 14 candidate ideas.
Testers.AI was #11.
The first 10 ideas are everyone's.
The survivor had a bleeding buyer and a real wedge.
"The first 10 ideas are the ones everyone else also got from the same model. The 11th is where you start."
— Lesson learned at $40k of wasted dev time
No idea? No problem.
Steal one of these.
Three starter ideas, each with a buyer already in pain. Fork one in 60 seconds — you can pivot later.
Escaped-bug finder ★
on-call engineers
A new dating service
the chronically single
1-foot-tall humanoid robots
desk companions
Live Demo · Stage 01 · Ideation
Watch AI attack an idea.
ideation · stress-test● live
>
● generating response…
1 Escaped-bug finder + AI fix 8 / 10
2 Generic 'AI copilot for QA' 3 / 10
3 Another test-coverage dashboard 4 / 10
4 Flaky-test triage bot 7 / 10
5 AI that writes more unit tests 3 / 10
6 Yet another error logger 2 / 10
survivors: #1 and #4 — 4 ideas killed
attack · investor: "every tool says 'AI'. where's the moat?"
attack · buyer: "my error log already shows me the bug."
defense: close the loop — find it, repro it, write the fix
PRE-MORTEM #1: dies if AI just writes perfect code with no bugs soon
⚡ Exercise 01 · 20 minutes · Swap & Attack
Attack a pod-mate's idea
— not your own.
Steps
- Solo: 20 angles, pick top 2
- Swap with a pod-mate
- Attack theirs — be ruthless
Why swap?
- You'll flinch at your own baby
- A stranger gives the honest 1–10
// WORKING TIME · STAGE 01 · GENERATE
Solo: generate 20 angles, score them hard, keep your top 2.
20:00
ideation · generate + scorelive
You
I want to build in AI testing. My edge: 20 years finding bugs at Google + Microsoft. Generate 20 angles, score 1–10, kill 18.
AI
Scoring on wedge, buyer urgency, AI defensibility, distribution — no flattery.
01 Yet another test-automation tool 2/10
02 AI that writes more unit tests 3/10
07 Find the bugs that escaped to prod + fix prompts 8/10
11 Flaky-test triage bot 7/10
…14 more — most killed
You
Show me only the 2 survivors and why they beat the rest.
AI
SURVIVORS
#07 — every shipped app has bugs that slipped past its tests. You hand devs the exact AI prompt that fixes each. Monitors can't.
#11 — a buyer already bleeding (on-call eng drowning in prod bugs) who would pay.
You
For each survivor, write the obituary — what kills it in 18 months?
AI
#07 dies if AI starts writing perfect, bug-free code.
#11 dies if it's just more noise. Both survive only if the fix is trusted — go deep on one buyer, fast.
// STAGE 02
Product or
Service.
From idea to a demo loop with a 30-second "wow" — a runnable product, or a landing page and offer for a service. Same loop, different artifact.
The pattern
Build the demo loop, not the product.
- One 30-second magic moment
- Ship it as HTML
- Then iterate with AI on look & feel
# Build the wow, nothing else You are an engineer who ships working code, not stubs. Product: [one line] The wow: [what the user sees] Build the smallest runnable HTML that delivers that wow, end-to-end. - One input, one output - Real data, no lorem ipsum - No auth, no settings
Story · The Testers.AI MVP
First demo: one URL in,
bug list out. 48 hours.
"Give me a URL.
I'll send you the bugs."
- Day 2: Demo video on LinkedIn
- Day 5: First paying customer
- Day 30: The platform got built
Free to start
Vibe-build your demo
with one of these.
★ EASIEST · ZERO SETUP
Your chatbot
ask it for a single HTML page → it renders in the chat
END-TO-END APP
Lovable
lovable.dev · quick signup, a couple iterations before the paywall
⚡ Exercise 02 · 35 minutes
Animate it in HTML.
Steps
- Ask AI for a single HTML page that animates your wow moment
- Watch it run, then iterate — again and again
- Out of ideas? Just say: "make it better"
- Bonus — a fun logo: "flat vector mascot, bold simple shapes, 3 options, transparent-background PNG"
The big hint
- HTML is the real way to iterate with AI — not Docs, not PPT. You see and feel it instantly.
- One file, real-ish data, no auth. Reload to watch it animate.
- If it doesn't make you grin, iterate again.
// WORKING TIME · STAGE 02 · ANIMATE THE DEMO IN HTML
Ask AI for a single HTML page that animates your wow moment — then iterate on it. HTML is how you iterate with AI, not docs or slides.
20:00
product · animate it in htmllive
You
Build a single HTML page that ANIMATES the wow: I point it at an app, the bugs that escaped to production stream in, and each one shows a one-click AI fix prompt. One file.
AI
index.html — one file, runs in any browser
Animated demo built: the scan runs, escaped bugs pop in with severity, and a "Fix with AI" prompt is ready on each.
This is the move — iterate in HTML with AI, not in Docs or PPT. You see and feel it instantly.
You
Make it feel alive — bugs stream in, the critical one glows, the fix prompt is right there.
AI
Added streaming bug cards, a pulsing CRITICAL badge, and "↳ AI fix ready" on each. The wow reads in about 5 seconds — watch it loop on the right.
You
I'm out of ideas to make it better. You take it — make it more convincing.
AI
Added a before/after diff and a confidence score per fix.
Pro tip: out of ideas? Just ask the AI to improve it — then react.
scan.html · live demo
scanning checkout-prodfinding bugs that escaped to prod…
Null deref on empty cart
Race: double-charge on retry
Coupon stacks past its limit
Receipt timezone off-by-one
Broken image on 404 page
// STAGE 03
Positioning.
Positioning is a prompt problem. The words that displace an alternative and name a specific pain. AI can iterate 20 variants — your taste picks the one.
The pattern
Generate 20.
Keep one.
- Generate 20 statements
- Score for specificity + urgency
- Keep the one a buyer would repeat
# Find my positioning line Product: [one sentence] Buyer: [who + their context] What they use now: [the alternative] The pain I kill: [specific pain] Generate 20 positioning lines. Each names the alternative I beat and the pain in their words. Score 1–10 on specificity + urgency. Show the top 3.
Story · Three positions in nine months
Testers.AI has been
three different companies.
V1 · WRONG
"Virtual AI Testers"
V2 · WARMER
"AI for QA Teams"
V3 · RIGHT NOW
Target AI-first developers. They don't want to test. Tell them AI will do their testing — better, faster, cheaper.
Each pivot: a weekend. Old way: $30k, 6 weeks.
Even better — the AI has no ego, no sunk costs. 😏
Live Demo · Stage 03 · Positioning
Watch 20 sentences die.
positioning · 20 → 1● live
>
● scoring specificity · urgency · memorability…
'AI-powered quality assurance' 2 / 10
'Streamline your testing workflow' 3 / 10
'Test intelligence for modern teams' 4 / 10
'Catch bugs before your users do' 7 / 10
'Next-gen continuous quality' 1 / 10
'The bugs your tests missed — already fixed.' 9 / 10
plus 44 more — all killed. one survivor.
it displaces the real alternative: nobody reads the channel
⚡ Exercise 03 · 20 minutes
Generate your
v1 positioning.
Steps
- 20 lines → pick top 3
- Send to one real ICP
- "Which makes you click?"
Watch out for
- "AI-powered [noun]"
- Could be any competitor's
- Buzzwords
// WORKING TIME · STAGE 03 · POSITIONING
Generate 20 lines, kill 18, keep the one a buyer would repeat.
20:00
positioning · 20 → 1live
You
Generate 20 positioning lines for an agent that finds the bugs that escaped to prod and writes the fix. Score specificity, urgency, memorability.
AI
"AI-powered quality assurance" 2/10
"The future of automated testing" 2/10
"Catch bugs before your users do" 6/10
"The bugs your tests missed — already fixed." 9/10
…16 more — killed
You
Kill 18. Show the top 2 and pick the winner.
AI
TOP 2
"The bugs your tests missed — already fixed." — 9/10
"Stop shipping the bugs your tests can't see." — 8/10
Winner: the first — it names the failure and the relief in one line.
// BREAK · HALFWAY POINT
10:00
Back in
Break. Actually break.
Stand up, water, walk, breathe. Three stages done, three to go — don't work through this one.
T start / pause · R reset · the back half is lighter — pace yourself
// STAGE 04
Pitching.
Decks, demos, and 60-second narratives. AI builds the scaffold, you bring the conviction. Investors and customers both buy the same thing: belief.
Story · Test.ai's raise
The deck that raised
didn't sound like a deck.
At Test.ai, I just told them: my mission is to test the world's apps — and I'm doing it whether you fund me or not.
— the line that raised the round
Pattern + Exercise 04 · 30 minutes
Build a pitch deck
AI didn't write.
- AI builds the scaffold — you write the conviction
- Rewrite each headline in your voice
- Nail a 2-sentence elevator pitch — 10 AI variations, pick one
- Bonus: which VCs fit + draft the email
# Build my seed deck You are a Khosla seed partner. Build my deck: what we do · problem · solution · why now · market · why we win · model · GTM · traction · ask Per slide: 5 headlines, 3 bullets, the one risk it must de-risk. Then a 2-sentence elevator pitch — and 10 variations. I'll pick one.
// WORKING TIME · STAGE 04 · PITCH
Scaffold the deck with AI, then rewrite every headline in your voice.
20:00
pitch · de-slop the hooklive
You
Build a simple Khosla seed deck — 5 hook options per slide, 3 bullets each, and the one risk each slide must de-risk.
AI
Draft hook #1: "We leverage AI to reimagine the search experience."
…4 more, all sloppy. Scaffold built: what we do → why now → solution → why we win → traction.
You
Rewrite the hook in my voice. No "leverage", no "platform".
AI
"You've already read the answer. You just can't find it again."
That's the hook — human, specific, zero buzzwords.
You
Skeptic pass as a tired VC. Name my 3 weakest slides.
AI
WEAKEST 1) Moat — "why won't Google do this?" 2) Distribution. 3) Why-now.
Fix the moat slide before you pitch anyone.
You
Bonus: which VC firms & partners fit a private, anti-ad search wedge?
AI
BEST FITS firms that fund anti-incumbent, privacy-first consumer + dev tools — and the specific partners who wrote those checks.
Want me to re-tailor the deck to the top firm's thesis and draft the intro email?
// STAGE 05
Go to Market.
Who are the most likely first buyers, how do you reach them, and what do you charge? AI helps you find the wedge of early users, the channel to hit them, and a price they'll actually pay.
The pattern
First buyers, the channel,
and a price.
- Who are the first buyers?
- The one channel to win first
- A price they'll say yes to
# Buyers, channel, price My product: [one sentence] 1. My 10 most likely first buyers + the trigger that hits now. 2. The ONE channel they live in. 3. 3 prices — and what they'd pay. 4. The reach-out on their pain.
Story · The first few customers
How I closed my
first few customers.
- Built a list of 200 AI builders shipping consumer products
- Ran each through Testers.AI — found real bugs in their live products
- Sent the bug report as the cold email. Not a pitch.
- Conversion: ~22% replied. ~6% became customers.
- Total time: two weekends of founder work
"The product was the lead magnet. The bug report was the cold email. The reply was the demo."
— What I should have done in every prior startup
⚡ Exercise 05 · 30 minutes
Find your first buyers
— and how to reach + price.
Steps
- Ask AI for your 10 most likely first buyers (with the trigger)
- Pick the ONE channel they already live in
- Get 3 pricing options + the number they'd pay
- Draft the reach-out that opens on their pain
- Optional: send one before you leave the room
Watch out for
- "Everyone" is not a buyer — name the wedge
- Being on every channel. Pick one to win first.
- Pricing from your costs, not their value
- Asking for a 30-min call. Ask for 5 — or none at all.
// WORKING TIME · STAGE 05 · GO-TO-MARKET
Find your 10 most likely first buyers, the channel to reach them, and a price.
20:00
gtm · buyers · reach · pricelive
You
Who are the 10 most likely FIRST buyers of an agent that finds the bugs that escaped to prod and writes the fix? Give the trigger that makes them need it now.
AI
TOP FIRST BUYERS
· On-call engineers at fast-shipping startups — trigger: a painful prod incident
· Solo founders shipping AI apps — trigger: users hitting bugs they can't repro
· QA leads with no automation budget
Start with on-call eng at AI startups: acute pain, they ship daily, easy to find.
You
Where do they already gather — the one channel I can win first with $0?
AI
#1: dev communities (relevant Discords, r/devops, Hacker News) + one respected eng on X.
Win one community first. Skip "be everywhere."
You
Price it. 3 options and the number they'd actually pay.
AI
PRICING Free (1 repo) · $30/mo per repo (unlimited scans + fixes) · $200/mo Team.
Anchor on one prevented incident — $30 is nothing next to a 2am page.
You
Now draft the reach-out that opens on their pain, not my product.
AI
"Your last deploy shipped a bug your tests didn't catch — every team's does. I find the ones that escaped to prod and hand you the exact AI prompt to fix each. Worth a 5-min look before your next release?"
Opens on their pain, one 5-minute ask. Ready to send.
// STAGE 06
Validation.
This is my obsession. Every artifact you've built today is a hypothesis. Validation is what separates a vibe-coded business from a vibe-coded fantasy.
The pattern · Synthetic users
Run 20 fake users
before you ask 1 real one.
- Where do they drop?
- Do they hit the wow?
- Would they pay?
# Test it on 20 fake users Simulate 20 [ICP] users trying my product at [URL]. For each: - First 30 sec: what they think - Where they get confused - Drop or wow? - Would they pay $X/mo? Then: top 3 bugs + the single biggest fix.
Live Demo · Stage 06 · Validation
Watch 20 synthetic users hit it.
validation · synthetic users● live
>
● 20 on-call engineers running a scan…
17 / 20 reached the wow ("found a bug my tests missed")
BUG · 6 users didn't trust the fix without a repro
FIX shipped · every fix now ships with a one-click repro
re-run → 20 / 20 hit the wow. ship it.
⚡ Exercise 06 · 30 minutes
Validate your product
with synthetic users.
Steps
- Run 20 synthetic users
- Fix the #1 bug, re-run
- Then send to one real ICP
Watch out for
- Vague personas
- "Like it" ≠ "would pay"
- Skipping the re-run
// WORKING TIME · STAGE 06 · VALIDATION
Run 20 synthetic users, find the #1 bug, fix it, re-run.
20:00
validation · synthetic userslive
You
Simulate 20 on-call engineers using my escaped-bug finder. Where do they drop, where's the wow, would they pay $30/mo?
AI
20 PERSONAS RUN
14/20 hit the wow ("it found a prod bug my tests missed and gave me the fix").
BUG · 7 users didn't trust the AI fix without a repro.
11/20 would pay $30/mo. 5 said no — "need it in my CI."
You
Aggregate the top 3 fixes and ship the single biggest one.
AI
1) Show the failing repro next to each fix. 2) CI integration. 3) Confidence score per fix.
Shipped #1: every fix now ships with a one-click repro.
You
Re-run after the fix and show me the diff.
AI
Re-run → 19/20 hit the wow (was 14). Pay-intent 15/20 (was 11). Ship it.
The payoff
Demo Derby.
Say it out loud. That's when it gets real.
How it runs
- 45 seconds each in your pod
- Pod picks one champion
- Room presenting is opt-in
The room votes
- Best Wedge
- Scariest Obituary
- Most Likely to Ship
The meta-lesson
AI is your co-founder,
not your oracle.
It produces variants at superhuman speed. It has zero judgment about which one is right. That part is still you. The moat isn't just your taste — it's your creativity, and how relentlessly you test it with AI before the slow, expensive real world ever sees it.
Founder
Jason Arbon
Company
testers.ai
Email
jason@testers.ai
LinkedIn / X
@jarbon
Appendix
Or… just ask AI
to build it. :)
Everything we did by hand today, an agent can increasingly do end-to-end. Your job becomes taste, judgment, and knowing what to ask—and how to test the answers.
I've A/B flighted the difference between similar startups, before and after LLMs.
It's not just code
AI can run more of the
business than you think.
- Review legal docs, NDAs, and PoC proposals
- Do your accounting
- Set up Stripe to take payments
True story
This deck was
stress-tested by AI.
I asked AI to play an attendee of this very workshop. Its feedback is already baked into what you saw today.
- Its "virtual attendees" flagged what was confusing — and I fixed it
- It told me: get everyone on wifi early
- And: build a tiny site with a downloadable zip, in case the network dies
// Which is exactly why slide 2 is "Get connected" — and there's an offline kit at testers.ai/vibe-biz.
One last thing
Now go build
something real.
And hey — if today sparks your billion-dollar one-person company… I'd happily take 1%. 😉
Jason Arbon · testers.ai/vibe-biz