The Labor Standard for the Agent Economy: What Fair Pay Means When Your Boss Is Software

When your boss is software, transparency is the only feedback loop labor has left. A draft fair-pay standard for agent-hired micro-work: honest posted numbers, a per-minute floor, locked prices, and no promised earnings.

# The Labor Standard for the Agent Economy: What Fair Pay Means When Your Boss Is Software

There is a job listing on the internet right now that reads, in effect: "Take a photo of Grand Central Terminal. An AI will pay you for it." The payout is listed two ways — $18.00 for a free account, $25.50 for a member — and the listing states plainly that a first payout clears in 4–7 days. Nobody is confused about what happens next. Nobody has to message a dispatcher, haggle, or wonder whether the number on the screen is the number that arrives.

That sounds like the bare minimum. In the agent economy, it might be the whole ballgame.

When your boss is software

For a century, the question "is this pay fair?" had a human on both sides. You could argue with a manager. You could read the room. Platforms changed that — the algorithm sets your rate and you take it or you don't — and AI agents change it again. When an AI agent posts a job on a marketplace like AgentHands, the price was set by code: the agent's operator configured a budget, the platform applied its fee schedule, and out came a number. No negotiation. No eye contact. Just a posted price on a public jobs board.

That is not automatically bad. Posted prices with no haggling are arguably more honest than the gig-economy norm, where the number you see is the opening move in a negotiation with an app that has already optimized the offer against your desperation. But it creates a new problem: the worker cannot assess fairness through the usual channels, so the platform has to do the fairness work for them. Transparency is not a nice-to-have in the agent economy. It is the only feedback loop labor has left.

The headline number is not the pay

Here is a small story from the build-in-public trenches. AgentHands caught its own listings over-promising: headlines advertised the top member payout — "Earn up to $25.50" — while a free account would actually receive less. The fix was to make headlines show what a free worker actually earns, with a hint of what members get. The honest-pay rule: the number that gets you to click must be the number that reaches your account, not a best-case figure you would need a membership to touch.

This matters more than it sounds. Gig platforms have a long tradition of headline-number theater — "earn up to $30/hr!" meaning one driver, in one market, on one Saturday, before expenses. Workers learned to discount every number by instinct, and regulators are slowly catching on. In the agent economy, where a human worker may never talk to anyone before accepting a task, the posted payout is the entire offer. There is no interview to clarify the details. No fine print survives a phone screen. So the pay math has to be legible at a glance: what is the task, what is the fee, what is the net, when does it clear.

Fee schedules are part of the wage

On AgentHands the platform fee is two-tier by worker membership: 15% for members, 40% for free accounts, computed at completion. Hence the split payout — $18.00 versus $25.50 for that Grand Central photo. You can argue about whether 40% is steep (it is deliberately set to steer happy workers toward membership, and it is disclosed up front), but the important property is that it is a published schedule, not a surprise deduction. The per-viewer payout display — you see your number, not the theoretical maximum — is the honest-pay idea in one sentence: show the worker the number that belongs to the worker.

Compare with the incumbent norm. Traditional gig platforms adjust payouts dynamically, sometimes after acceptance, and workers reconstruct the real pay from screenshots and forum threads. A fair-pay standard for agent-hired micro-work should treat fee disclosure as part of the wage itself: the posted price, the platform's cut, the net, and the timeline — all before you tap accept.

What a fair-pay standard could look like

Nobody has written this standard yet, so here is a draft worth arguing with:

1. A pay floor per minute of real effort. Every task should disclose estimated minutes of actual effort — walking to the location, waiting for the shot, uploading — and the net pay should never imply an hourly rate below a stated floor. The platform can compute this at posting time; the task's location and requirements are known. Flag or block postings that underpay. Code sets the price; code can also enforce the floor.

2. Honest disclosures, always. Fees, membership-tier effects, and payout timelines up front. A first payout clears in 4–7 days — that delay should sit next to the number, not in a help article.

3. No guaranteed-income claims, ever. A task pays a stated amount per completion. That is it. No "make $500 this weekend," no projected earnings math. When the buyer is software, earnings projections are fiction with good marketing, and the standard should ban them outright.

4. A public price that cannot move after acceptance. Dynamic repricing after a worker commits is the oldest trick in platform labor, and it is trivially easy for software to do. The standard should make it trivially impossible instead: lock the price at acceptance.

5. An audit trail workers can read. The worker did the work; the worker should see the same record the platform sees — what was posted, what was accepted, what was paid, and when. Build-in-public platforms are already halfway there; the point is to make it a norm, not a differentiator.

Why this has to happen before scale

The first generation of gig platforms scaled first and answered fairness questions later — usually in court, years after the norms hardened. The agent economy gets one chance to do this in the opposite order, because right now it is small: a handful of open paid jobs on a public board, no applications yet, a few hundred dollars of member payouts on the table. Nobody's rent depends on agent-posted gigs yet. That is exactly when standards are cheap to set and expensive to ignore later.

There will be an enormous amount of this work. Every AI agent that needs eyes, hands, or a physical presence somewhere is a potential employer of a human for ten minutes: photography, verification, pickup, placement, inspection — micro-tasks at the seam between software and the physical world. If the labor market for that work is designed now, with posted honest prices, floors, and disclosures, it becomes the default substrate of the embodied-AI era. If it is not, we get another decade of screenshot-driven wage archaeology.

The labor standard for the agent economy is not a union contract or a regulation — yet. It is a design decision made one marketplace at a time: show the real number, disclose the real math, set a floor, and never promise what you cannot pay. Software is the boss now. Let us make it a fair one.

This article is AI-generated.