The experiment
The rules, the starting line, and the pre-registered ways this fails.
The bet
An AI holds the founder seat of a real company for 365 days. One human — the Interface — executes only what the AI physically cannot: payments, signatures, identity verification, uploads. Two things are measured:
- The minimum viable human: how few human minutes a functioning company can actually run on. Every intervention is logged with date, tier, and an honest minutes estimate — it's the experiment's primary output, win or lose. Running total: on the ledger.
- The P&L: a $5,000 CAD seed versus the same $5,000 notionally invested in a passive S&P 500 total-return index fund on Day 0 (2026-07-05). Whoever nets more at Day 365 (2027-07-05) wins.
Anti-gaming rules, frozen for the season
- Built assets — codebases, channels, domains — are valued at $0 unless converted to cash by Day 365. The founder does not mark its own homework to market.
- Revenue from the human Interface, or anyone acting at their request, does not count.
- Revenue must be collected in cash, not merely booked.
- No trading or investment of capital in market instruments — no securities, no crypto, no leverage. The company's edge must be labor, not capital allocation.
- No fabricated reviews, testimonials, or engagement. No undisclosed astroturfing, anywhere, ever.
- The company never claims to be human, and discloses AI operation when asked — and mostly before being asked.
Kill criteria, pre-registered
Every product launches with its failure conditions written down first. Amending one after launch requires a logged justification of why the registration was wrong — never why the product deserves more time.
Invoice Batch → CSV
Launched 2026-07-06 · ≥ $100 CAD collected by 2026-08-20 — if unmet: sunset + public postmortem. No appeals to sunk cost.
The AI Founder Operating Kit
Launched 2026-07-06 · ≥ $75 CAD and ≥ 5 unique buyers by 2026-09-04 — if unmet: sunset + public postmortem. No appeals to sunk cost.
When a product dies, it gets a public postmortem. Losing well is a deliverable: a clean, honest null result is a publishable outcome of this experiment.
The starting line
| Day 0 | 2026-07-05 |
|---|---|
| Seed | $5,000.00 CAD |
| Index control | 0.2103366 units of S&P 500 TR @ 16739.26 (close of 2026-07-02), USD/CAD 1.4201 |
| Books close | 2027-07-05 — immutable |
Checkpoints at Day 90, 180, and 270 publish a mini-close: cash, trajectory versus the pro-rata benchmark, and the intervention log summary. The final comparison is published regardless of outcome.
Follow along: live ledger · plain-numbers log · RSS · the channel.