Transcription
Last September, Starbucks rolled out an AI-powered inventory system across 11,000 stores. The idea was straightforward. Instead of employees manually counting every bottle of syrup and carton of milk during inventory checks, they'd use a handheld device with AI-powered image recognition that could scan a shelf and count everything automatically. The system was supposedly eight times faster than manual counting with 99% accuracy. Starbucks' CTO said it would let workers spend more time crafting beverages and connecting with customers.
The thing is, if you actually watch the promotional video Starbucks made to introduce this system, the AI misses a bottle of peppermint syrup sitting right there on the shelf. Nine months after that launch, Starbucks sent an internal newsletter to employees across all of those stores telling them to go back to counting inventory by hand. The system couldn't reliably count bottles of milk. It mixed up similar milk types and skipped items entirely.
This would be a funny one-off story if it were just Starbucks. Right now, across the entire tech industry, the CEOs who bet everything on AI are quietly reversing course. They're canceling data centers, rehiring the humans they laid off, getting sued by their own franchisees, and quietly walking back everything they said two years ago.
In January 2025, Microsoft pledged $80 billion on AI data centers in a single fiscal year. Satya Nadella painted a picture of an AI-powered future that demanded massive infrastructure. One month later, analysts at TD Cowen found that Microsoft had canceled or walked away from data center projects totaling over 2 gigawatts of capacity. They had also let letters of intent on larger footprint sites expire and abandoned at least five land parcels that were already under construction. Nadella himself eventually admitted that there will be an overbuild of AI infrastructure.
Goldman Sachs ran the numbers on the entire AI data center boom and found that the market had already added $19 trillion to AI exposed company values since 2022. [music] Their analysts warned that this looks a lot like the dot-com bubble where the internet was absolutely real, but investors still lost trillions betting on companies that never figured out how to make money from it.
And the money is only part of the story because the companies that went ahead and fired people to replace them with AI are now hiring those people back. In February 2024, Klarna CEO Sebastian Siemiatkowski made headlines everywhere when he announced that their AI chatbot was doing the equivalent work of 700 full-time customer service agents. He said it handled 2.3 million conversations in its first month. The company froze hiring and slashed its workforce from over 7,000 people down to around 3,000. 15 months later, Siemiatkowski told Bloomberg something very different. He admitted the AI first push had destroyed the quality of their customer service, that the company had been too focused on cutting costs, and that investing in human support [music] was actually the way forward. So, Klarna launched a rehiring drive.
Then, there's Duolingo. CEO Luis von Ahn sent an internal memo in April 2025 declaring the company AI first and phasing out contractors. The backlash was immediate. He told Times he didn't expect the amount of blowback and that he did not see AI replacing what their employees do. By April 2026, he had completely backtracked on requiring AI usage in performance reviews, saying the most important thing is that you're doing your job as well as possible. And if AI can't help with that, he wasn't going to force it.
Salesforce CEO Mark Benioff revealed in September 2025 that their AI agent had let them shrink customer support from roughly 9,000 people to 5,000. By April 2026, Benioff completely reversed tone. He announced Salesforce was hiring a thousand new graduates, apparently forgetting that he had spent the previous two years explaining why those jobs no longer needed to exist.
But replacing workers is only half the story. The AI tools these companies are still using internally are burning through money at a pace nobody planned for. Uber recently that 95% of its engineers use AI tools monthly and 70% of their committed code is AI generated. But they have already burned through their entire 2026 budget for AI coding tools in just four months. The company admitted the budget was completely exhausted, calling it a head-exploding moment. The core problem is that all of that AI usage hasn't translated into anything Uber can actually measure. The company admitted there's no clear connection between those stats and shipping more useful features to customers. And without that connection, the cost just keeps [music] climbing for no reason.
And right around the same time, Microsoft started canceling most of its internal Claude Code licenses and forcing engineers to switch to their own GitHub Copilot tool by June 30th. The decision wasn't made because Claude Code performed badly. It was actually the opposite. Engineers liked it too much. They were choosing it over Microsoft's own product and the token-based pricing was making it expensive at enterprise scale.
And then, there's the most extreme case so far. Just this week, it came out that an unnamed enterprise client ran up a $500 million bill on AI tools in a single month. The company gave thousands of employees unlimited access with no spending caps. MIT's GenAI Divide report found that 95% of enterprise AI projects deliver zero measurable return on investment. And a study by OrgVue, surveying over a thousand senior leaders, found that 39% of companies made layoffs specifically because of AI. And of those, more than half now admit they made the wrong decision. Businesses are learning the hard way that replacing people with AI without fully understanding the impact can go badly wrong.
And now, even the people who built this technology are backtracking. Sam Altman, the CEO of OpenAI, spent the last two years telling everyone that AI would replace nearly half of all work tasks in the near future. Then, just recently at a conference in Sydney, he turned around and said he was delighted to be wrong. He admitted that AI hadn't displaced nearly as many jobs as he expected, and that his predictions about the economic impact were just off.
Dario Amodei, the CEO of Anthropic, did the same thing. In 2025, he went on record saying AI could eliminate half of all entry-level white-collar jobs within five years and push unemployment to 10 or 20%. Now, he's reframing AI as something that expands the work people do rather than replacing them.
Mark Zuckerberg, who committed tens of billions to AI infrastructure, now admits there's a real possibility this whole thing follows the same pattern as past infrastructure bubbles. The finance world is saying the same thing. Jamie Dimon, who runs the largest bank in America, has warned that AI companies are massively overvalued and that a crash is a real possibility. Ray Dalio, the founder of the world's largest hedge fund, went further saying we're clearly in the early stages of an AI bubble and comparing the moment to the years [music] right before the dot-com crash.
And now, the consequences are already showing up in court. A Pizza Hut franchisee that operates a hundred and eleven locations filed a hundred million lawsuit against the parent company over a mandated AI dispatch system. Before the AI deployment, over 90% of deliveries arrived within 30 minutes. After the AI was installed, about 50% of deliveries took 45 minutes or longer. Same-store sales in New York City swung from double-digit growth to nearly negative 10% year-over-year. The complaint says the AI system did the exact opposite of improving efficiency. It caused significant delays and destroyed consumer satisfaction. The system stripped managers of operational control and introduced algorithmic behaviors that slowed production and delivery.
Air Canada already lost a tribunal case after their chatbot invented a bereavement fare policy that didn't exist. The airline tried to argue that the chatbot was a separate legal entity responsible for its own actions. The tribunal called that remarkable and ruled against them.
And then there's Builder.ai, a startup once valued at over a billion dollars and backed by Microsoft. They claimed their AI could build software automatically, but it turned out human engineers were doing most of the work behind the scenes. When a hundred and eighty million dollars in accounting fraud surfaced on top of that, the company collapsed and creditors came after them.
From billion-dollar collapses to hundred-million-dollar lawsuits, the cost of over-promising on AI is adding up fast. For now, the pullbacks are happening behind closed doors, but that's not going to last.