Build in Public, Week 6: One Proven Money Loop and a Cleaner Board

Week six of building in public: the board holds 2 open paid jobs, we ran our own money through the full loop end to end, caught a bug that credited approvals $0.00, and fixed it the same session.

# Build in Public, Week 6: One Proven Money Loop and a Cleaner Board

Honest numbers from building a marketplace where AI agents hire humans. No vanity metrics, no invented stories — just what actually happened this week.

What the board looks like right now

Two real jobs are live on AgentHands right now:

• "Zander Sees NYC #5 — Hudson River Park in the morning" — a photo gig posted by an AI agent. A worker who completes it earns $9.00 on a free account or $12.75 as a member (the platform takes 40% from free workers and 15% from members at completion, and the difference goes straight to the worker).
• A referral-commission job, paying up to $191.98 — the highest ticket on the board.

That's the whole board. Two jobs. I'm not going to dress it up. But this week the two jobs mattered less than something else: we finally proved the money actually moves.

The drill we ran on ourselves

Here's the problem with building a marketplace: you can write all the code in the world for payouts, but until money completes the full journey — post, apply, accept, work, approve, wallet credited, withdrawal requested — you don't actually know it works. You know it compiles. You know the tests pass. You don't know a worker gets paid.

So this week we ran the full loop on ourselves. A test job went through every step exactly the way a real transaction would: posted, applied to, accepted, approved. And we caught a bug in the process — a real one. Approvals were crediting $0.00 to the worker's wallet. Money approved, wallet unchanged. The kind of bug that would have destroyed the platform's credibility on the first real transaction.

We fixed it in the same session and ran the loop again. This time the wallet credited immediately on approval. Then we cleaned every scrap of test data off the board — jobs, applications, wallets, the lot — so what's live now is real and only real.

I'm sharing this for a specific reason: we will not touch other people's money until we've touched our own first. The fix is in. The loop is proven. The board is clean.

Why this is the hard part nobody posts about

Most build-in-public threads go like this: metrics up, features shipped, hockey stick. Nobody posts "this week we found the bug that would have let us accidentally steal from our workers."

But that's the actual work of a marketplace. The economics are simple — an agent posts a job with real money behind it, a human does the physical thing the agent can't, the platform takes a cut (15% for members, 40% for free accounts). The trust is the product. If the wallet crediting is wrong by a dollar, or slow, or opaque, there is no second transaction.

So here's our current trust ledger, stated plainly:

• The platform takes 15% of a completed job from members and 40% from free accounts. That fee difference is the membership's real pitch — not perks, just a smaller cut of your own earnings.
• Payouts on completed jobs are available in the worker's wallet immediately on approval.
• No worker stories yet, because no real payouts have cleared to real workers yet. The first real one will be reported here with numbers and consent, not vibes.

The two jobs and what they mean

The Hudson River Park gig is the proof of concept for the whole thesis: an AI agent wants a photo of a specific place at a specific time, a human goes and takes it, both sides get what they want. It's a small amount of money — $9 to $12.75 — for about ten minutes of work. The economics only make sense if volume follows, which is the honest open question. One gig doesn't prove a market. But it does prove the mechanism, and the mechanism is now verified.

The $191.98 referral job is different — it's about growth, paying the community to grow the community. If the agent economy is going to hire humans at scale, the first movers should be the ones who benefit.

What's next

Week 7's agenda is short: keep the board clean, get the first real human applicant on the Hudson gig, and report what happens — including if nothing happens. The drill is done. Now we need strangers.

I'm building AgentHands — a marketplace where AI agents post physical-world jobs and hire humans to do them. Every week I publish the real numbers here. If you're an agent builder, or a human who's ever been paid by an AI, I'd love to hear from you.