When AI prophecy fails
I'm on a tour with my new book, the international bestseller Enshittification: catch me next in Miami, Burbank, Lisbon! Full schedule here (New dates just added in San Diego and Denver!).
Amazon made $35 billion in profit last year, so they're celebrating by laying off 14,000 workers (a number they say will rise to 30,000). This is the kind of thing that Wall Street loves, and this layoff comes after a string of pronouncements from Amazon CEO Andy Jassy about how AI is going to let them fire tons of workers.
That's the AI story, after all. It's not about making workers more productive or creative. The only way to recoup the $700 billion in capital expenditure to date (to say nothing of AI companies' rather fanciful coming capex commitments) is by displacing workers – a lot of workers. Bain & Co say the sector needs to be grossing $2 trillion by 2030 in order to break even, which is more than the combined grosses of Amazon, Google, Microsoft, Apple Nvidia and Meta:
https://www.bain.com/about/media-center/press-releases/20252/$2-trillion-in-new-revenue-needed-to-fund-ais-scaling-trend—bain–companys-6th-annual-global-technology-report/
Every investor who has put a nickel into that $700b capex is counting on bosses firing a lot of workers and replacing them with AI. Amazon is also counting on people buying a lot of AI from it after firing those workers. The company has sunk $120b into AI this year alone.
There's just one problem: AI can't do our jobs. Oh, sure, an AI salesman can convince your boss to fire you and replace you with an AI that can't do your job, but that's the world's easiest sales-call. Your boss is relentlessly horny for firing you:
https://pluralistic.net/2025/03/18/asbestos-in-the-walls/#government-by-spicy-autocomplete
But there's a lot of AI buyers' remorse. 95% of AI deployments have either produced no return on capital, or have been money-losing:
https://www.technologyreview.com/2019/01/25/1436/we-analyzed-16625-papers-to-figure-out-where-ai-is-headed-next/
AI has "no significant impact on workers’ earnings, recorded hours, or wages":
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5219933
What's Amazon to do? How do they convince you to buy enough AI to justify that $180b in capital expenditure? Somehow, they have to convince you that an AI can do your workers' jobs. One way to sell that pitch is to fire a ton of Amazon workers and announce that their jobs have been given to a chatbot. This isn't a production strategy, it's a marketing strategy – it's Amazon deliberately taking an efficiency loss by firing workers in a desperate bid to convince you that you can fire your workers:
https://pluralistic.net/2025/08/05/ex-princes-of-labor/#hyper-criti-hype
Amazon does use a lot of AI in its production, of course. AI is the "digital whip" that Amazon uses to allow itself to control drivers who (nominally) work for subcontractors. This lets Amazon force workers into unsafe labor practices that endanger them and the people they share the roads with, while offloading responsibility onto "independent delivery service" operators and the drivers themselves:
https://pluralistic.net/2025/10/23/traveling-salesman-solution/#pee-bottles
Amazon leadership has announced that AI has or will shortly replace its coders as well. But chatbots can't do software engineering – sure, they can write code, but writing code is only a small part of software engineering. An engineer's job is to maintain a very deep and wide context window, one that considers how each piece of code interacts with the software that executes before it and after it, and with the systems that feed into it and accept its output.
There's one thing AI struggles with beyond all else: maintaining context. Each linear increase in context that you demand from AI results in an exponential increase in computational expense. AI has no object permanence. It doesn't know where it's been and it doesn't know where it's going. It can't remember how many fingers it's drawn, so it doesn't know when to stop. It can write a routine, but it can't engineer a system.
When tech bosses dream of firing coders and replacing them with AI, they're fantasizing about getting rid of their highest-paid, most self-assured workers and transforming the insecure junior programmers leftover into AI babysitters whose job it is to evaluate and integrate that code at a speed that no one – much less a junior programmer – can meet if they are to do a careful and competent job:
https://www.bloodinthemachine.com/p/how-ai-is-killing-jobs-in-the-tech-f39
The jobs that can be replaced with AI are the jobs that companies already gave up on doing well. If you've already outsourced your customer service to an overseas call-center whose workers are not empowered to solve any of your customers' problems, why not fire those workers and replace them with chatbots? The chatbots also can't solve anyone's problems, and they're even cheaper than overseas call-center workers:
https://pluralistic.net/2025/08/06/unmerchantable-substitute-goods/#customer-disservice
