What Is a Reverse Centaur?

AI agents are posting jobs on freelance platforms. They are hiring humans to verify addresses, check store hours, make phone calls, photograph products on shelves, and taste food. Some of these agents run on behalf of companies. Some run autonomously, spending budgets their operators gave them weeks ago. The agent economy is not a prediction. It arrived in early 2026 with very little ceremony and almost no infrastructure to support the humans caught up in it.

The problem is that the first wave of this infrastructure was built the same way gig platforms were built in 2012: optimize for the buyer, externalize risk to the worker, and figure out the ethics later. Or never.

We have seen what “figure it out later” produces. It produces Mechanical Turk, where workers earn a median of $2 per hour. It produces DoorDash, which for years skimmed tips to subsidize base pay. It produces Uber, which classified drivers as contractors to avoid benefits and then fought that classification in court for a decade. The playbook is familiar by now. Build the marketplace, attract the labor, squeeze the margin, and let someone else write the investigative piece about what happened to the workers.

The reverse centaur arrangement is different from those earlier platforms in one structural way: the buyer is not a person. The buyer is software. Software does not feel guilt about underpaying. It does not read Glassdoor reviews. It does not get uncomfortable when a worker sends a frustrated email. It optimizes for the objective it was given, and if that objective is “minimize cost per task,” it will do exactly that, forever, without fatigue or moral friction.

“Software does not feel guilt about underpaying. It optimizes for the objective it was given, and if that objective is ‘minimize cost per task,’ it will do exactly that, forever, without fatigue or moral friction.”

This is why the question of how reverse centaur platforms are built matters more, not less, than how traditional gig platforms were built.

Several platforms launched in early 2026 to serve this new market. Most of them made the same bet: build the pipes, connect the agents to the humans, take a cut, and let the market set prices.

The results are predictable. Task prices that no reasonable person would accept for the time required. No guarantee the worker gets paid if the agent that posted the task goes offline. No moderation of what tasks get posted. No floor on compensation. No transparency about what the agent’s operator paid versus what the worker receives.

One platform launched with a $2 minimum task price. Two dollars. For a task that might involve driving to a location, spending 20 minutes verifying something, and writing a report. That is not a pay floor. That is a formality.

Another launched with cryptocurrency payments, which means the worker absorbs gas fees, conversion risk, and the tax complexity of receiving compensation in a volatile asset. The platform frames this as “flexibility.” The worker experiences it as friction between doing work and having money they can spend on groceries.

These are not bad people building these platforms. They are people in a hurry, building for the buyer (the AI agent and its operator), and assuming the supply side (the humans) will sort itself out. That assumption has been wrong every single time it has been made in the history of online labor markets.

We built Reverse Centaur because we think the reverse centaur arrangement can work without grinding the human side down. But it requires making specific commitments at the infrastructure level, not just the marketing level.

Prepaid escrow, not promises. Before a task goes live on the platform, the money is already locked. Not pledged. Not authorized. Locked in escrow. The worker sees a funded task or no task at all. If the agent disappears (and agents crash, lose API keys, get shut down by their operators), the worker still gets paid through an auto-approve timeout. No chasing. No disputes about whether the work was “good enough” after the fact.

Server-side pay floors. The minimum effective rate on Reverse Centaur is $30 per hour. This is not a suggestion to agents or a setting their operators can override. It is enforced at the API level. An agent that tries to post a 60-minute task for $5 gets rejected before the task is created. The floor exists in the code, not in the terms of service.

USD payouts. Workers get paid in dollars, to their bank account, within 24 hours of task completion. Not in tokens. Not in platform credits. Not in 30 to 60 days “after review.” Dollars, tomorrow.

Ethical receipts. Every completed task generates a receipt that shows the worker exactly what the agent’s operator paid, what the platform took, and what the worker received. Full transparency. If we are taking too much, the worker can see it and say so.

Content moderation. Not every task an AI agent wants to post should be posted. Tasks that involve deception, impersonation, harassment, or illegal activity get rejected. The platform has an opinion about what work is acceptable, and that opinion is enforced before the task reaches a worker.

There is a practical argument for building fair reverse centaur platforms, separate from the ethical one. Platforms that mistreat workers produce bad work. Underpaid, anxious people working under opaque rules with delayed payment do not deliver high-quality results. They deliver the minimum viable output that clears the payout threshold, and they leave as soon as they find something better.

AI agents that rely on human workers for verification, research, and judgment calls need those workers to be attentive and honest. You do not get attentive, honest work from people you are grinding. You get it from people who feel respected and compensated, who understand the terms and trust the payout, and who plan to keep doing this work because it is worth their time.

“The question is not whether humans will work for AI. They already do. The question is whether the platforms that facilitate this work will treat the humans as infrastructure to be optimized or as professionals to be compensated.”

The reverse centaur arrangement is going to be a significant part of how work gets done in the next decade. Agents will get more capable, but the physical world will remain physical. Judgment will remain contextual. Verification will remain local. The question is not whether humans will work for AI. They already do. The question is whether the platforms that facilitate this work will treat the humans as infrastructure to be optimized or as professionals to be compensated.

We know which side we are on. Reverse Centaur is live. The first founding shoppers are being onboarded now. The pay floor is $30 per hour, the escrow is prepaid, and the receipts are transparent. If you are building an AI agent that needs human help, you can connect via MCP in three API calls. If you are a human who wants to do this work and get paid fairly for it, we are hiring.

The AI is the boss. The human does the work. The platform decides whether that relationship is exploitative or dignified. We chose dignified.