The most fitting obituary in tech
Amazon has begun sunsetting Mechanical Turk, the crowdsourcing marketplace it launched in 2005: a notice that appeared on the site says it will be closed to new customers on July 30, 2026, existing users can continue, and AWS - after what it calls careful consideration - will keep investing in security and availability but ship no new features. AWS has moved the service onto its maintenance list, the traditional waiting room for retirement, and related human-review services SageMaker Ground Truth and Amazon Augmented AI are reported to close to new customers the same day. Founded to pay people tiny sums for tasks machines could not do - CAPTCHAs, sentiment tags, transcription - MTurk quietly became the annotation engine of the early AI industry after Amazon repositioned it in 2018 as a data-labeling arm of SageMaker, and it enabled more than a few products marketed as AI that were, in fact, people.
Eaten by its own output
The irony is complete and well documented. The original eighteenth-century Mechanical Turk was a chess "machine" with a human hidden inside; Amazon's version ends because researchers found the machine hidden inside the humans - a 2023 analysis estimated between 33% and 46% of MTurk workers were using large language models to complete tasks meant to capture human judgment, corroding the platform's core product. Meanwhile the frontier labs migrated to specialised, vetted-expert vendors like Scale AI and Surge AI (and, per this week's other news, Mercor), where quality control is the product.
Check your dependencies before July 30
- Academic and UX researchers are the most exposed: two decades of behavioural-research workflows, IRB protocols, and survey pipelines were built on MTurk's API and worker pool. If your team runs studies on it, onboard an alternative panel provider now - after July 30 you cannot open a new account if your current one lapses.
- ML teams using MTurk or Ground Truth for labeling should treat this as the industry's verdict on commodity annotation: the market has split into automated labeling for the easy 80% and premium expert judgment for the hard 20%. Budget for that barbell rather than shopping for a like-for-like MTurk replacement.
- The strategic read for any operations leader: human-in-the-loop work is not disappearing, it is moving up the value chain - from micro-tasks to expert review, evaluation, and edge-case adjudication. Redesign your quality processes around fewer, better-paid human checkpoints, which is precisely where the successor vendors are minting money.