Amazon is winding down Mechanical Turk, the crowdwork marketplace it has run since 2005, starting by closing it to new customers on July 30, 2026. Existing requesters are not being turned off at that date, but the more telling signal is that Amazon Web Services has added the Mechanical Turk service to its list of offerings in maintenance, the status AWS typically uses for products it intends to retire rather than keep developing. Amazon has not published a reason, and there is no announced end date for existing users, so this is best read as the beginning of a long goodbye rather than a sudden closure. Even so, retiring a service that launched in November 2005, making it older than most of AWS, is a notable moment.
For anyone who never used it, Mechanical Turk was the original microtask marketplace. Companies and researchers, called requesters, would post small units of work known as Human Intelligence Tasks, and a large distributed crowd of people, informally called Turkers, would complete them for payments that were often just a few cents each. Amazon described the whole idea as artificial artificial intelligence, a deliberately awkward phrase for software that presents a clean automated interface while real humans do the work behind it. The name itself was a reference to an 18th century chess playing automaton that appeared to think for itself but actually concealed a human operator inside the cabinet.
What makes this more than a nostalgia story is how central that model turned out to be to modern AI. The same crowd that labeled photos, drew bounding boxes, transcribed audio, moderated content, and answered questionnaires generated a huge share of the annotated data that machine learning systems were trained and evaluated on. Mechanical Turk was also the quiet engine of a great deal of academic research, especially in psychology and the social sciences, where it became a cheap and fast way to run experiments on real people at scale. For twenty years it was one of the least glamorous but most widely used tools for turning human judgment into machine readable data.
Amazon has not said why it is stepping back, but the shape of the market offers an obvious explanation. The company has built newer, more managed data services of its own, such as SageMaker Ground Truth, which package labeling into a supervised workflow and increasingly automate parts of it. Specialized labeling companies and dedicated human feedback vendors have taken over much of the high value annotation work, particularly the kind used to fine tune large models. The result, as The Register summed it up, is that not even AI could save Mechanical Turk, because the automation the platform helped make possible, together with better managed rivals, hollowed out the open marketplace even as the industry's hunger for labeled data kept rising.
The reason this matters beyond one product line is that Mechanical Turk made something visible that the AI industry usually prefers to keep in the background, which is how much human labor sits underneath systems that are marketed as automatic. That labor did not disappear, it moved into specialized firms, offshore labeling operations, and reinforcement learning from human feedback pipelines, often under the same pressures around low pay and little protection that dogged Mechanical Turk for years. Closing the open marketplace does not end the reliance on human input, it just makes it harder to see. As a marker, the wind down of the service that coined artificial artificial intelligence is a fitting one for a moment when the real thing is finally good enough to absorb the work the fake version used to route to people.