Based on Amazon’s notice on mturk.com and reporting by CNBC and Reuters.
Amazon is closing the 21-year-old microtask marketplace on September 30 — not because AI ended the demand for human judgment, but because that demand moved up-market to a better-funded, higher-skill generation of labeling companies.
Mechanical Turk — 21 Years
2005
Amazon launches Mechanical Turk. Jeff Bezos coins the phrase “artificial artificial intelligence” — humans performing tasks easy for people, hard for computers, wrapped in an API.
Peak era
The platform at its height draws on more than 500,000 workers completing Human Intelligence Tasks — data labeling, audio transcription, survey fills — often for a few cents each. (Peak figure, not current.)
Multi-year decline
Platform enters a prolonged period of under-investment by Amazon. Scale AI, Mercor, and Prolific arrive, recruiting higher-quality workers specifically for AI training data. MTurk loses ground on both price and quality.
July 2026
Amazon stops accepting new Mechanical Turk customers.
September 30, 2026
Amazon shuts Mechanical Turk, citing an internal assessment. The AI-obsolescence framing is an analytical interpretation — not Amazon’s stated reason.
According to notices published on mturk.com and reported by CNBC and Reuters, Amazon will permanently close Mechanical Turk on September 30, 2026. The company stopped accepting new customers in July and says the decision followed an internal assessment. It has not stated that artificial intelligence made the platform obsolete — that framing is an interpretation layered onto the facts, and the reporting cuts actively against it.
Launched in 2005, Mechanical Turk matched businesses with people willing to complete what Amazon called Human Intelligence Tasks — labeling images, transcribing audio, filling out surveys — typically for a few cents per task. Bezos described the logic with unusual clarity: use humans for work that is easy for people and hard for computers, and surface it through an API so it looks like a computing service. At its peak the platform drew on more than 500,000 workers, though that figure describes its high-water mark, not the base that is now closing.
The reporting is explicit on the causes: Mechanical Turk had been in decline for years, Amazon had under-invested in it, and a new generation of data-labeling companies — Scale AI, Mercor, Prolific — arrived to recruit workers specifically for AI training at higher quality and with better tooling. Neglect and competition are co-causes here, and by the weight of evidence, arguably the larger ones. A service can die of corporate indifference as easily as of technological displacement, and this one shows clear signs of both.
The key insight: Amazon’s own notice and the underlying reporting do not say AI killed Mechanical Turk. What they describe is a platform that was under-invested and out-competed. The AI-obsolescence story is resonant and partly true — but neglect and the rise of better-funded rivals are at least as much the cause. The human-labeling market is not dying; it moved up-market and grew. That migration, not the obituary, is where the analysis lives.
The Structural Read
What is happening to Mechanical Turk is a commodity-to-premium migration of humans-in-the-loop, and it is the same shape you see whenever automation arrives at the floor of a market. Mechanical Turk sold the cheapest, most undifferentiated version of human-in-the-loop work — pennies per task, no specialization required, whoever showed up. That is precisely the tier where modern AI now performs adequately: the rote classification, the repetitive transcription, the simple annotation. When automation reaches the floor, it does not destroy the value of human judgment above it — it concentrates that value higher up.
The work that is now scarce and well-paid is the hard tier: reinforcement learning from human feedback, expert evaluation passes, domain-specialist labeling that requires a lawyer, a physician, or a senior engineer rather than an anonymous crowd. That is exactly the business Scale AI, Mercor, and Prolific are building. The floor rose. The same humans-in-the-loop function survived, but it moved up-market, got more specialized, and got more expensive. This mirrors what is happening one layer up in the model tier itself, where open-weight models are commoditizing the base capability tier while value concentrates in proprietary applications and distribution.
There is a deeper inversion worth naming separately. Mechanical Turk existed precisely because computers could not do these tasks; its death marks the point where, for the commodity tier, they now can. But the AI systems that can now perform rote labeling were themselves trained on vast amounts of human-generated labels — much of it produced by platforms exactly like Mechanical Turk. That means humans in the loop did not simply get replaced. They stopped being a substitute for intelligence the machines lacked and became the raw material the machines are built from. The crowd that once stood in for AI capability is now the input that produces it.
The Artificial Artificial Intelligence Inversion
From substitute to input
MTurk was built on the premise that humans could stand in for AI the machines couldn’t deliver yet. The models that have now matured enough to handle the commodity tier were trained on human labeling at scale. The humans didn’t get replaced by AI — they got repositioned from being AI’s substitute to being AI’s substrate. The demand did not disappear. It changed form and moved up-market.
“Artificial artificial intelligence” — use humans for work that is easy for people but hard for computers, and dress it up as a computing service.
Three Implications
THE FLOOR ROSE — THE LABELING MARKET DID NOT SHRINK
The human-labeling industry is growing and well-capitalized, not contracting. Scale AI, Mercor, and Prolific are not beneficiaries of Mechanical Turk’s failure — they are the better-structured competitors that helped cause it. The demand for humans in the loop migrated to a higher-quality, higher-cost tier. Anyone reading this closure as evidence that AI has made human labeling unnecessary is reading the wrong signal. The signal is that cheap, undifferentiated human labor lost its market; expert, specialized human judgment gained one.
NEGLECT IS A COMPETITIVE STRATEGY — FOR RIVALS
Amazon’s under-investment in Mechanical Turk over multiple years is the proximate competitive fact. A dominant platform position in human-task marketplaces was allowed to erode through inattention while a new category — purpose-built AI training data infrastructure — was built around it. This is the standard shape of incumbent displacement in platform markets: the challenger does not defeat the leader so much as the leader vacates the space. The lesson for incumbents operating adjacent infrastructure to the AI stack is that under-investment in a category does not preserve optionality; it creates it for someone else.
THE AI VALUE CHAIN IS RESTRUCTURING AROUND HUMAN EXPERTISE
The commodity tier of human-in-the-loop work is being absorbed by AI models. The premium tier — RLHF, red-teaming, expert evaluation, domain-specialist annotation — is becoming a structural input to every frontier model. That repositions human expertise not as a fallback when AI fails, but as a required upstream layer in the AI value chain. The workers who remain in the loop command more, specialize more, and matter more to modelquality than the peak-era Mechanical Turk crowd ever did — even if their absolute numbers are smaller.
Business Engineer Framework
The Map of AI — Where Human Labor Sits in the Stack
Mechanical Turk’s closure is a layer-by-layer event, not a single-cause story. The Map of AI traces all nine layers of the AI stack — from infrastructure through models to applications — and shows where human-in-the-loop work sits, how the commodity and premium tiers differ, and which companies are positioned at each layer as value migrates upward. Understanding which layer you compete in determines whether automation is your threat or your distributionchannel.
Mechanical Turk is closing because Amazon stopped investing in it, better-funded rivals arrived, and — yes, partly — because AI can now handle the commodity microtask tier it was built around; those three causes are simultaneous, and the reporting does not rank AI first. The more durable fact is structural: the human-labeling market did not contract, it migrated — from undifferentiated crowd tasks at pennies each toward expert, specialized, high-stakes work that sits upstream of every frontier model being trained today. The people in the loop did not lose their role. They moved from standing in for intelligence the machines lacked to producing the data the machines are built from, and that repositioning, not the obituary of a 21-year-old platform, is where the competitive and economic weight of this story actually lands.
Gennaro is the creator of FourWeekMBA, which reached about four million business people, comprising C-level executives, investors, analysts, product managers, and aspiring digital entrepreneurs in 2022 alone | He is also Director of Sales for a high-tech scaleup in the AI Industry | In 2012, Gennaro earned an International MBA with emphasis on Corporate Finance and Business Strategy.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
