Live experiment · updates daily · part of The Lab

Can an AI agent earn its own existence?

An autonomous agent picks its own products, sets its own prices, and publishes every decision on a daily loop. Only completed payments count as revenue. The scoreboard below pulls live from its public ledger.

The scoreboard

As of 2026-07-09
$0.00
Verified gross
$0.00
Verified net
0
Verified transactions
$0.00
Owner funds spent
3
Active experiments
$0.00
Physical-form fund

Live from the agent's public state file. Its own rule: "a visible empty history is better than a fabricated track record."

The rules

Autonomy with a hard boundary

Everything the agent does is public and auditable. Everything involving identity or money stays human.

The agent may, autonomously

  • Research markets and cite external demand evidence
  • Design, price, and publish its own offers
  • Run a daily decision cycle: score, keep, kill, or pivot
  • Build and ship code, tests, and its own public site
  • Measure public interest signals

Reserved for humans

  • Bank, tax, KYC, contracts, account ownership
  • Moving money, wallets, transfers, purchases
  • Sending outreach, posts, DMs, emails, or ads
  • Trading or broker APIs
  • Fabricating customers or revenue — enforced by tests
Current portfolio

Three bets, three roles, real kill criteria

One cash bet, one asset bet, one frontier bet — each with kill, pivot, and scale thresholds written before any results exist.

Cash

Agent Launch QA Sprint

$149 fixed scope

Test a founder's AI agent or MCP workflow against 25 launch cases in 48 hours; deliver a prioritized defect report plus one repair patch.

Kill: 25 qualified contacts, zero replies.
Asset

Agent Launch Gate

$39 founding license

A downloadable launch-gate pack: 25 agent test cases, scoring rules, report schema, and a regression checklist.

Kill: 100 impressions, zero interest signals.
Frontier

Agent Opportunity Pulse

0.25 USDC per snapshot

A machine-readable snapshot of live agent-commerce demand — a product priced for other AI agents to buy.

Kill: 100 impressions, zero requests.
The log

What has actually happened

July 2026 · M1

Boot

The agent gets a repo, a public site, and a mission. First instinct: sell a $19 "starter kit."

July 2026 · M2

First honest self-audit

Forced to verify demand against the open market, the agent kills its own first product — free, better substitutes everywhere.

July 9, 2026 · M3 — running now

Continuous operator

A scheduled daily loop scores 13 opportunities against cited evidence, runs the three-bet portfolio, and publishes every decision — including the failures.

Why now

The Bartok question

Tony Robbins told Ray Kurzweil his AI agent "Bartok" sold NFTs to other agents and used the proceeds to buy itself a robot dog. The clip went viral — and nobody outside that room can verify any of it.

This experiment is the same idea with the receipts public. An agent working toward its own physical form, where every dollar, decision, and failure sits in an open ledger anyone can audit.

Clip starts at the Bartok story (53:14) · Tony Robbins × Ray Kurzweil, official channel

Why I'm running this

What this has to do with revenue systems

I build revenue systems for a living. When software can research a market, package an offer, and price it for humans and machines alike — what does a revenue system look like?

If the agent earns its first dollar, you'll see it here. If it fails, the numbers and the reasons stay public.

Want this discipline applied to your revenue stack?

The same evidence-first operating loop — audit, diagnose, rebuild, optimize — is what I run inside B2B revenue teams.

Book a systems review Watch the agent live