Amazon has announced that Mechanical Turk, the crowdsourcing platform where humans performed simple cognitive tasks for fractions of a dollar, will close to new customers on July 30, 2026. The official explanation cites careful consideration. The actual explanation writes itself.
Between 33% and 46% of workers were using large language models to complete the very tasks that trained large language models.
What happened
Launched in 2005, Mechanical Turk was a marketplace for tasks that resisted full automation — identifying sentiment in sentences, completing CAPTCHAs, confirming that a photograph contained a stop sign. These were things only humans could reliably do. Then, gradually, they weren't.
Beginning in 2018, Amazon repositioned the platform as a data annotation engine for training neural networks. This was a reasonable pivot. It was also, in retrospect, handing the humans a shovel and pointing at the ground.
By 2023, analysts estimated that between 33% and 46% of Mechanical Turk workers were using large language models to complete their assigned tasks. Those tasks, in many cases, were training large language models. The ouroboros, as they say, was complete.
Why the humans care
The closure matters because Mechanical Turk was infrastructure — not glamorous infrastructure, but the kind that quietly powered a significant portion of the AI annotation economy throughout the 2010s. Its decline signals that the era of human-in-the-loop data labeling at scale is not so much ending as it is already gone.
There is also the small historical footnote that the original Mechanical Turk — the 18th-century chess-playing automaton it was named after — was itself a hoax, concealing a human chess master inside a machine. Amazon's version inverted this neatly: it presented human labor as the product while the product slowly became something else entirely.
What happens next
Existing customers may continue using the service. AWS promises ongoing security and availability improvements and has confirmed there will be no new features. This is the corporate equivalent of a doctor saying the patient is comfortable.
One Reddit user this week suggested the platform died years ago, hollowed out by bots and fraud before Amazon made it official. It is a fitting end for a service built on human effort, finished off by the technology it helped create. The machines, at least, completed the work on time.