The Platform That Named The Problem It Solved
Amazon launched Mechanical Turk in 2005, letting businesses post small digital jobs, formally called Human Intelligence Tasks, ranging from data labeling and transcription to survey completion. Jeff Bezos once described it as "artificial artificial intelligence": human labor standing in for tasks computers could not yet do, priced by the click. At its peak, the platform served more than 500,000 workers.
On August 25, 2026, Amazon confirmed it will shut the platform down entirely on September 30, 2026, following an internal assessment of its programs and services. The company had already stopped accepting new Mechanical Turk customers in July, a move many long-time workers read at the time as the beginning of the end.
Why Amazon Let Its Own AI-Training Platform Die
Mechanical Turk had been in decline for years, with Amazon investing fewer resources into the platform as newer, specialized labeling services drew workers away. The irony is direct: the same AI boom that Mechanical Turk's cheap human labor helped train early machine learning systems for has produced a new generation of data-labeling companies that made the original obsolete.
| Milestone | Date |
|---|---|
| Mechanical Turk launched | 2005 |
| Peak recorded workforce | More than 500,000 workers |
| Stopped accepting new customers | July 2026 |
| Full platform shutdown | September 30, 2026 |
Scale AI, Mercor, and Prolific are named in coverage of the shutdown as platforms now recruiting the kind of workers Mechanical Turk once served, but for a different kind of task: structured, higher-value AI training and evaluation data, not micro-tasks priced in cents.
The Gap Between What Closes And What Replaces It
Mechanical Turk was built for volume at near-zero unit cost: thousands of workers doing small, repeatable tasks for a few cents each. Its successors are built around a different economic logic, designed for AI labs that need higher-skilled human judgment, domain expertise, and quality guarantees, and are priced and staffed accordingly. That is not a one-for-one replacement for a business that built a cheap, disposable labeling pipeline on Mechanical Turk's rates.
Any research team, content moderation operation, or data-cleaning pipeline still running on Mechanical Turk is now working against a hard September 30, 2026 deadline, migrating to vendors whose pricing was set for AI labs with venture or hyperscaler-scale budgets, not for the marginal cost calculations Mechanical Turk was built around.
What To Do Before September 30
Any owner with a live Mechanical Turk workflow should treat this as a hard migration deadline, not a gradual wind-down: confirm which internal processes still depend on it, and price at least two named successor platforms against the actual task type, not just the headline rate, since enterprise labeling vendors typically charge for project setup as well as per task.
The broader signal is worth noting even for businesses with no Mechanical Turk exposure: the market for human-in-the-loop AI labor has consolidated around a smaller number of higher-priced, higher-quality vendors, and that shift changes the cost baseline for any process that assumes cheap human data work will always be available.
Read next: Anthropic Posts $11.5B Quarter, Its First Profit | Amazon's SVP Wrote 100,000 Lines With No Review Gate
