Amazon repurposed its regular weekly retail technology meeting Tuesday to figure out why its retail website keeps breaking. The answer, buried in internal documents and then quickly deleted, according to the Financial Times: its own AI initiatives.
Four high-severity incidents hit its retail website in a single week, including a six-hour meltdown last Thursday that locked shoppers out of checkout, account information and product pricing. The meeting, run by the senior vice president who oversees Amazon’s ecommerce infrastructure, was framed as a “deep dive” into what went wrong. What went wrong, it turns out, involves the very AI tools Amazon has been pushing its own engineers to adopt, according to the FT.
An internal document prepared for the meeting initially identified “GenAI-assisted changes” as a factor in a pattern of incidents stretching back to Q3. That reference was deleted before the meeting took place, according to the Financial Times, which viewed both versions of the document.
Amazon has pushed back on the reporting. In a blog post, the company said only one incident involved AI tools, that “none of the incidents involved AI-written code,” and that the cause was “an engineer following inaccurate advice that an agent inferred from an outdated internal wiki.” Amazon also told Fortune the meeting was a routine weekly operations review, not an emergency gathering. The company also said it is not accurate that it introduced new approval requirements for engineers working with AI tools, and that AWS was not involved in any of the incidents.
“As part of normal business, the meeting will include a review of the availability of our website and app as we focus on continual improvement,” an Amazon spokesperson told Fortune.
The internal documents, obtained and reported by CNBC, tell another story. Dave Treadwell, SVP of eCommerce Foundation, laid it out for staff:. Site availability had not been good recently, he wrote, and the string of Sev 1s—the most severe classification for incidents that take down important systems—demanded immediate attention.
But the internal documents, as initially written, according to CNBC, tell a more complicated story. Treadwell acknowledged in his note that “best practices and safeguards” around generative AI usage haven’t been fully established, and wrote that the company would introduce “controlled friction” into deployments involving the most critical parts of the retail experience, according to CNBC. Either way Amazon calls it, the message to engineers was that AI-assisted changes now get more scrutiny.
The timing for that kind of admission is brutal for Amazon. The company, which just surpassed Walmart to top the Fortune 500, is spending more on AI infrastructure than any company on Earth—$200 billion in projected capital expenditures this year.
Amazon is also aggressively thinning out its workforce. The company laid off roughly 14,000 corporate workers in October — mostly middle managers — followed by another 16,000 in January. That’s on top of more than 27,000 employees cut between 2022 and 2023. In June, Jassy wrote in an internal memo that Amazon would need fewer employees thanks to AI-driven “efficiency gains,” repeating his drumbeat emphasizing the AI future of less workers needed at the giant retail platform. When the October cuts came, Jassy reframed the rationale on an earnings call to be about “culture,” saying that the company had grown too fast during the pandemic, and Amazon needed to be “lean” and “move fast.”
But a separate Amazon memo announcing the same layoffs cited the need to adapt to “transformative technology,” the kind of language that maps a lot more cleanly onto an AI-driven workforce reduction than a spring cleaning. But it seems that either way, Amazon has found itself in need of more humans in the process.
It’s an interesting narrative violation in a world of AI-related layoffs. Jack Dorsey’s Block cut nearly half its workforce last month — 4,000 employees — and tied the decision explicitly to AI-driven productivity gains. Dorsey said most companies would reach the same conclusion within a year. Salesforce’s Marc Benioff said he needed fewer heads after cutting 4,000 support roles. The C-suite consensus is that increasing AI investment will pay for itself with smaller workforces.
But the promise that AI would lighten the load isn’t playing out— at least, not for the workers who remain, and not for the systems they manage. A new analysis reported by the Wall Street Journal of 164,000 workers by ActivTrak found that AI is increasing the speed, density, and complexity of work rather than reducing it. Time spent on email, messaging, and chat apps more than doubled after workers adopted AI tools. Time devoted to focused, uninterrupted work—the kind required for solving complex problems—fell 9%. Meanwhile, new research from Anthropic suggests the gap between what AI can theoretically automate and what it’s actually automating is enormous. Even in software and math — where 94% of tasks could theoretically be handled by AI, only about 33% are being automated today. Legal constraints and institutional troubles are all slowing deployment, Anthropic said. Amazon’s outages could be a live demonstration of why.
