Amazon's MTurk Is Closing. The Workers Already Knew Why.

The crowdsourced human-labeling layer that fed the first generation of AI is being retired because the people doing the labeling are now the same people using AI to do the work. Amazon's decision to shut Mechanical Turk on September 30 closes the public commons of paid micro-tasks, but the more revealing story is what was already happening inside the platform long before the closure notice.

Business Insider reports that 33% to 46% of MTurk workers were already using large language models to complete tasks as of 2023, and that share has almost certainly climbed in the three years since. The platform did not die from competition. It died from inside its own supply side. The mechanism is worth naming because it generalizes: whenever a labor market is paid to produce the inputs that make a substitute cheaper, the substitute first leaks into the workers' own workflow before it kills the market outright. The task does not disappear. The human intermediary does.

The migration is already visible. Uber now offers drivers AI data-labeling work, and a secondary market has emerged for selling raw documents and recordings to model builders directly. The next category of paid human micro-tasks will not be a public marketplace at all. It will be a private pipeline, with workers accessed by contract and the price set by what the model cannot yet do without them. The platform that taught AI to read is the platform AI learned to replace. The work, where it still exists, has moved somewhere with no homepage.