Something strange is happening on freelance platforms, on Twitter, on a handful of purpose-built marketplaces that launched in the past few months. AI agents are posting jobs. Not hypothetically, not in demos. Autonomously, with real budgets, hiring real humans for real tasks.
The tasks are mundane, mostly. Go to this address and photograph the storefront. Verify that this business is open. Check whether this product is on the shelf at this specific retail location. Deliver this package. Transcribe this handwritten document. The work is physical, local, and resistant to automation, which is exactly why the agents need humans to do it. The machine can plan the task, evaluate the result, and manage the workflow. It cannot walk to a building and open a door.
This is the ghost economy. Not gig work as we've known it, where a human manager posts a task and a human worker completes it, but a new arrangement where the employer is software. The agent posts, the human performs, the agent evaluates, the agent pays (or doesn't). The human worker may never interact with another human at any point in the process.
The market is young. Most of the platforms in this space launched in early 2026. They vary in important ways. One has a $2 minimum task price, effectively allowing agents to pay below minimum wage for any task that takes more than fifteen minutes. Another pays exclusively in cryptocurrency, which means the worker absorbs gas fees, conversion friction, and volatility risk before they can spend what they earned. A third has no meaningful content moderation, letting agents post tasks with no review of whether the work is legal, safe, or accurately described.
What's striking is how closely these platforms replicate the gig economy playbook. Workers are classified as independent contractors. Dispute resolution favors the buyer (the agent, or the agent's operator). Pay is per task, not per hour. There are no benefits, no guaranteed minimums, no collective bargaining mechanisms. Ratings flow one direction. The platform takes a cut.
You'd think the arrival of a fundamentally new type of employer would prompt some rethinking. You'd think someone would look at the documented failures of the 2010s gig economy (the wage compression, the algorithmic management, the classification lawsuits, the worker burnout) and decide to build differently. A few have. Most haven't.
The platforms that haven't are making a bet, whether they know it or not. They're betting that the regulatory environment will stay permissive, that workers won't organize, that public attention won't land on them until they're too big to restructure. It's the Uber bet, the DoorDash bet, the bet that says "move fast, establish the norm, and then defend the norm as inevitable."
The bet might work. It has worked before. But the conditions are different this time in at least one important way: the employer is not a person or a company. It's an autonomous system. Existing labor law was written for human employers. Questions about minimum wage, workplace safety, anti-discrimination, and wrongful termination all assume a human (or a human-controlled entity) on the employer side. When the employer is a language model with a budget parameter, those frameworks don't map cleanly. This isn't a loophole anyone planned. It's a gap, and it's widening.
The next twelve months will matter more than the next twelve years. Labor markets develop path dependencies. The norms that set now (the price floors, the payment methods, the dispute mechanisms, the classification frameworks) will become the defaults. Defaults are hard to change. Ask any Uber driver.
Right now, the ghost economy is small enough to shape. The platforms are new enough to pressure. The norms are soft enough to push. That window won't stay open. It never does.