AI is leaving the demo stage and entering the supply chain

On 30 July 2026, Amazon will stop accepting new customers for Mechanical Turk. Existing customers can keep sending work through the old pipes, but Amazon says it does not plan to add new features. That is a quiet ending for one of the stranger bargains of the internet: tiny units of human judgement, sliced thin enough to be mistaken for software.

The obvious read is that AI is replacing the people who labelled, checked and cleaned up the data behind automation. I think that misses the more interesting shift. AI is leaving the demo stage and entering the supply chain. The product is no longer always a chatbot, an avatar, or a magic text box. Increasingly, AI is the hidden production capacity behind shampoo, cookies, private schooling and weekend software archaeology.

TechCrunch reported that Mechanical Turk will close to new customers while staying available to existing ones. The symbolism is hard to ignore. Mechanical Turk was part labour market, part API, part confession. It made visible, if you knew where to look, the amount of human effort needed to make machines look competent.

Now the interface is flipping. Instead of humans hiding inside automation, automation is disappearing inside ordinary goods and services.

Reuters reports that L’Oréal, Nestlé, Haleon and Mondelez are using AI in product development, ingredient testing, recipe ideas and supply-chain planning. L’Oréal says AI helped identify skincare molecules that could be repurposed for shampoo. That is a much less theatrical use case than a chatbot writing sonnets, and probably a more commercially durable one.

This is the part many AI debates still underprice. The economic value of AI will not only come from products labelled “AI”. It will come from cycle-time compression: fewer dead ends in R&D, cheaper experiments, faster variations, tighter planning. Consumers may never ask whether a biscuit involved a model. They will just see a new flavour, a different price point, or a product that arrived on the shelf faster than it used to.

The hidden factory

There is a historical parallel with electrification. The interesting phase was not the first electric motor bolted onto old machinery. It was the later redesign of work around a new source of power. AI is approaching that kind of moment in some sectors. The question is less “where is the AI feature?” and more “what process got rebuilt because prediction, generation or classification became cheap?”

Education is the emotionally charged version of the same pattern. The Decoder reports on AI private schools selling wealthy US families on personalised learning, including Alpha School’s mix of AI tutoring and project-based workshops. The pitch is not novelty. It is throughput: learn faster, waste less time, adapt the curriculum to the child.

That sounds humane when framed as individual attention. It sounds less comforting when framed as a premium supply chain for cognitive development. If AI tutoring becomes a luxury layer first, the result may be a two-speed education system: children whose learning loops are constantly measured and tuned, and children stuck with the older queue. The same technology can be an equaliser or a moat, depending on who gets access and under what business model.

Then there is the oddest story in the batch: The Decoder reports that a Google AI Studio product lead used Claude Code and Fable 5 to port Command & Conquer: Generals Zero Hour to iPhone and iPad in “a few hours”. The port runs natively on ARM64 rather than through an emulator, and the source code was published openly, though users need their own game assets.

That is not enterprise transformation. It is something more culturally slippery: solo development with industrial leverage. Old code, niche taste and modern AI tooling combined into a result that would previously have required a small team, more time, or both. The weirdness is the point. AI production capacity will not only optimise corporate workflows; it will revive abandoned software, spawn hobby products and make tiny markets economically sane.

For builders, the lesson is blunt. Stop treating AI as a surface you add to a product and start treating it as a change in the cost structure of making the product. The next durable AI companies may be the ones users barely recognise as AI companies at all.

---