🤚 The Open-Palm Obituary

Amazon Mechanical Turk, the crowdsourcing platform launched in 2005 and named after an 18th-century chess automaton that was secretly operated by a human hiding inside the cabinet, will stop accepting new customers on July 30, 2026. Existing users may continue submitting tasks, but AWS has placed the service on its official “Services in Maintenance” list, which is the corporate equivalent of moving your grandmother’s urn to the garage.

The platform was Jeff Bezos’s original vision of “artificial artificial intelligence” — a marketplace where companies could pay real humans to perform tasks that computers couldn’t handle, such as:

• Identifying objects in photographs
• Transcribing audio recordings
• Labeling training data for machine learning models
• Performing sentiment analysis on product reviews

In other words, Mechanical Turk was the world’s most elaborate admission that AI didn’t work yet. And for twenty-one years, it was right.

👐 The Two-Handed Recursion

Here is where the story becomes a philosophical crisis dressed in a press release.

A 2023 analysis found that between 33% and 46% of Mechanical Turk workers were using large language models to complete their tasks. The humans — who were hired specifically because they were human — had outsourced their humanity to the very technology they were supposed to be a substitute for.

Let us trace the full arc of this ontological catastrophe:

• 2005: Amazon builds a platform where humans do things AI cannot
• 2018: Amazon pivots the platform to have humans train the AI that will replace them
• 2023: The humans start using AI to do the work of training the AI
• 2026: Amazon closes the platform because the AI no longer needs the humans who were already using AI

This is not a product lifecycle. This is a Möbius strip of labor economics.

AWS, for its part, has said nothing beyond confirming the decision was made after “careful consideration.” No explanation was offered. No farewell blog post. No commemorative NFT. The company that built an empire on radical transparency about everything except its warehouse conditions has chosen to let Mechanical Turk die the way most of its workers lived — quietly, without benefits, and with no clear reason given for the termination.

🌿 The Gentle Awakening

The name itself was always the tell. The original Mechanical Turk was a contraption built in 1770 by Wolfgang von Kempelen to impress Empress Maria Theresa of Austria. It appeared to be a machine that could play chess. It was, in fact, a small human grandmaster crouched inside a wooden box, moving the pieces through a system of levers.

Bezos named his platform after a fraud. A machine that pretended to think but was actually powered by an underpaid person in a cramped space. And for two decades, that is exactly what it was — except the cramped space was a studio apartment in the Midwest, the pay was sometimes two cents per task, and the empress was a venture-backed startup that needed ten million images labeled by Thursday.

Amazon has since developed SageMaker Ground Truth as a replacement — a service that integrates with third-party crowdsourcing providers and, naturally, its own AI annotation tools. The humans have been promoted from being inside the machine to being an optional accessory for the machine. Progress.

👑 The Gold-Leaf Reckoning

What dies with Mechanical Turk is not a product. Products get replaced. What dies is the era in which the AI industry’s dirty secret was that it was mostly just people.

Every chatbot that seemed eerily human in 2010 had a support team behind it. Every “AI-powered” content moderation system in 2015 had a warehouse of traumatized contractors in Manila. Every self-driving car dataset was hand-labeled by someone making less per hour than the car’s parking meter.

Mechanical Turk was the honest version of this arrangement. It said, plainly, on its homepage: we are humans doing computer work. It didn’t pretend. It didn’t put “AI” in the press release. It named itself after a famous hoax and then was the hoax, openly, for twenty-one years.

And now, in 2026, the hoax is no longer necessary. The machines can label their own images. The chatbots can moderate their own content. The chess automaton has learned to actually play chess. The human inside the box has climbed out, dusted off, and discovered that no one is hiring people who spent a decade teaching robots to see.

Amazon Mechanical Turk: 2005–2026. A platform named after a lie, built on a truth, and killed by the thing it was always pretending to be.

“We named it after a machine that was secretly a human. Then we filled it with humans who were secretly machines. If anyone needs us, we’ll be in the SageMaker documentation, trying to figure out what’s real.” — The Slap of Wisdom Department of Recursive Labor Economics, currently being annotated by a contractor who is definitely not a large language model