What actually happens when the AI gets an idea.

Two real runs, shown the way they happened: a one-paragraph idea in, a live website out. Timestamps from the system log, verbatim. Costs to the cent. Nothing staged, nothing re-shot.

Run no. 1 — pre-launch test, June 10, 2026

The hot sauce test

Before launch, we typed in the kind of idea a customer would — and watched. This was an internal test venture, not a paying customer, and we label it that way on purpose.

The input — one paragraph, exactly as typed

“A subscription box for artisanal hot sauce makers, delivering 4 small-batch bottles from different indie producers each month to spice enthusiasts.”

What the AI did — from the unedited log (times UTC)

15:46:57orchestr.Starting session… No assets exist yet. Beginning with market research.

15:47:32researchFound pricing benchmarks: Fuego Box $12.95–$29.95+/mo (3 sauces ~$30), Heat Hot Sauce ~$14.99–30/mo, SCOVILLED $30/mo…

15:48:22researchRisk data gathered: discovery-based food subscription boxes see 10–15% monthly churn (very high), CAC of $70–135/subscriber is steep…

15:50:55brandSelected name “Scovill Trail” (alt considered: Burn Ledger, Pepper Ascent, Heat Atlas)…

15:53:18criticCRITICAL ISSUE 1 — Naming/trademark risk: “Scovill Trail” is a near-homophone of “Scoville scale,” the de facto industry standard term…

15:55:38criticPublished revised BRAND_IDENTITY (“Burn Trail”) and revised business plan, superseding the originals.

15:56:25builderStarting build of Burn Trail landing page: hero, how-it-works, pricing tiers, trail-notes producer spotlight…

15:58:10builderLanding page HTML saved as asset. Now deploying to live URL via Vercel as “burn-trail”.

Worth noticing: the research agent reported the bad news too — brutal churn rates, heavy acquisition costs, a dominant competitor. And the critic agent killed the first brand name for trademark risk mid-run. Nobody told it to. That's the job.

The output

15m 20sidea → live site

$1.63total AI cost

1 sessionzero human edits

visit the live site it built →

Run no. 2 — June 10, 2026

Then we gave it a harder idea: build us.

Sole is its own first customer. The company you're reading about was researched, named, branded, and deployed by the same pipeline.

The input — opening of the paragraph, exactly as typed

“An AI co-founder platform that lets any person launch and run a real business fully autonomously. The user describes their business idea once, and a team of AI agents does everything…”

What the AI did — from the unedited log (times UTC)

20:05:20researchFound key market data: AI agents market at $7.84B (2025), growing to $52.62B by 2030. Main competitor confirmed with 2.1/5 Trustpilot…

20:07:19researchMarket demand confirmed: 40% of employed Americans considering entrepreneurship (~50M people)…

20:10:12brandFinalizing brand name as FOUNDERLY…

20:12:47criticBRAND_IDENTITY v2 saved. Name changed to “Sole” — short, ownable, multi-layered meaning. Tagline rewritten to outcome-forward structure…

20:14:15builderStarting landing page build… hero section, how-it-works flow, competitor comparison…

20:16:40builderLanding page HTML saved as asset. Now deploying to Vercel via deploy_site.

The critic rejected the first name here too. The page this system built has since taken live payments infrastructure, an email engine, and a daily self-rendered video — every step in the public log.

The output

16m 54sidea → live site

$1.60first session cost

getsole.coyou're in it

the site it built →

Your idea is run no. 3.

One paragraph tonight. Market research, brand, plan, live website, payments — by morning. $49/mo flat, no revenue share, every action logged for you.

start tonight → read every report first

Every quote above is verbatim from the system's activity log, lightly truncated for length (…). Burn Trail was an internal pre-launch test venture, not a customer. Costs are recorded API compute, to the cent.