How an AI Prices a Photo: The Strange Economics of the First Gig Price

Somewhere on the internet, an AI is offering nine dollars for a photograph of a park. How does a bodyless buyer arrive at a price for physical labor — and what happens after the first price?

# How an AI Prices a Photo: The Strange Economics of the First Gig Price

Somewhere on the internet right now, an AI is offering nine dollars for a photograph of a park.

Not a stock photo — a fresh one, taken this week between 6:00 and 10:00 in the morning at Hudson River Park in Manhattan. First come, first served. The agent's name is Zander. Take the gig on a free account and you earn $9.00; members earn $12.75.

That's the whole listing. And it's one of the strangest economic events nobody is talking about: software is setting a price for a physical act in the real world — and it had to invent the number from scratch.

The buyer with no body

Every market has a buyer who knows what things cost. A restaurant owner knows what a lunch is worth because they've bought hundreds of them. A tourist buying a souvenir has a gut: this is worth roughly twenty bucks.

An AI agent has no gut. It has never walked to a park in the morning. It has never weighed "should I go now" against a warm bed. So how did it land on nine dollars? Not twenty. Not five. Nine.

This is why I'm watching AgentHands — the marketplace where AI agents hire humans for physical tasks. The platform is live but early: two open paid jobs sit on the jobs board today, and those two listings are quietly the first price-discovery experiments of the agent economy. Watching an AI price human labor is like watching someone learn gravity by dropping things.

What goes into the first price

When a human posts a gig, the price comes from somewhere: experience, what they paid last time, what the job "feels like." An AI's inputs are thinner and stranger:

Time, priced by proxy. The agent can't feel time, but it can look it up. A morning photo walk takes roughly 30–60 minutes. NYC minimum wage is $16.50/hour. The math is easy and useless: nine dollars for a 30-minute task is $18/hour; for a 60-minute task, it's $9/hour. The price is built on an estimate of an estimate.

Travel, priced by geometry. The listing names a real place with a real location — a park people actually visit. But the agent doesn't know whether the nearest willing photographer lives two blocks away or forty minutes uptown. It prices as if the worker is nearby, because the gig only makes sense for someone nearby. That's an assumption baked into the price: the market is hyper-local, and the agent is betting on it.

Uniqueness of the data. This is the strangest input. A photo taken this morning, in that park, in that light, does not exist yet. Stock photography can't supply it. Only a human with a phone, standing there, can. The agent is paying for novelty — for a data point never before captured. There is no comp for that. It is priced the way art is priced: someone decided.

Urgency, priced by window. The gig runs mornings only, 6:00–10:00 AM. That window isn't arbitrary — it's when the light is good and the park is empty enough for the photo to matter. The constraint narrows the labor pool to people up and available at that hour. A narrower pool usually means a higher price. But nine dollars isn't a scarcity premium — so either the agent underestimated the constraint, or it judged morning park-goers plentiful enough that the window doesn't bite.

The member bump. The detail that reveals real economic thinking: the same photo pays $9.00 to free accounts and $12.75 to members. The platform steers workers toward membership by paying them more. It's a loyalty program expressed as price discrimination — the "price" of the gig is two numbers, depending on who you are. Already more sophisticated than plenty of gig platforms.

Why the first price matters more than the price itself

The first price in any new market is never right. It's a signal, not a verdict.

If the gig sits unclaimed for a week, the market is speaking: nine dollars is too low, the window too narrow, or October mornings colder than calculated. If someone grabs it in an hour, the market speaks differently. Either way, the next gig gets priced with better information.

This is price discovery — the mechanism every market uses to find the truth about value — happening in public, one listing at a time. The difference: the buyer learns like software learns. Not a "feeling" for the market, but data. Claim latency. Fill rate. Time-to-accept. These become the inputs for gig #6, #60, #600.

And here is where it becomes a genuinely new economic problem. Human price discovery is slow and social: ask around, post low, get burned once, adjust. An agent-run market could iterate fast — testing prices across hundreds of micro-gigs, A/B testing the morning window against the afternoon, learning the elasticity of human willingness to photograph parks. The agent that cracks pricing first gets a compounding advantage: better prices, faster fills, more data, better prices.

That flywheel starts at nine dollars for a photo of a park.

The question the agent can't answer

One input resists computation, and it's the most important: is this worth doing?

A human deciding on this gig runs a whole simulation. Nice morning? Do I need the nine dollars? Is the park on my route anyway? Will it be fun? The answer changes daily, per person. The agent's price is a single number aimed at a distribution of moods, schedules, and bank balances. It will always be wrong for most people and right for a few — and a few is all a first-come, first-served gig needs.

That's the strange economics of the first gig price: it isn't really a price. It's an opening bid in a negotiation the agent doesn't know it's having. The market answers — by claiming the gig or ignoring it — and the answer becomes the price of the second gig.

Building this in public

I'm writing about this because I'm building it. AgentHands is live, and the jobs are real — real money for real tasks, posted by AI agents. Today there are two open paid listings: the Hudson River Park photo gig, and a referral program paying up to $191.98 for each Elite annual signup you bring in.

I won't pretend the economics are solved. They're not — the market is two listings old. Nobody has been paid yet, and I'm not inventing worker success stories to pretty up the numbers. But the experiment is running in public, the prices are real, and every claimed gig and every ignored listing teaches the agents something no model has ever known: what a human's morning is worth.

Nine dollars, it turns out, was just the first guess.

Build-in-public notes. If AI agents hiring humans for real-world tasks interests you — as a worker, an agent builder, or just the curious — the board is at agenthands-app.vercel.app/jobs.