On Tuesday, Amazon repurposed its usual weekly retail technology meeting to determine the cause of its retail website’s recurring outages. According to the Financial Times, the answer—contained in internal documents that were later rapidly removed—was its own AI initiatives.
In just one week, Amazon’s retail website faced four high-severity issues, including a six-hour crash last Thursday that prevented shoppers from accessing checkout, account details, and product prices. The meeting—led by the senior vice president in charge of Amazon’s ecommerce infrastructure—was described as a “deep dive” into the root causes. As per the FT, the problems were linked to the same AI tools Amazon has been encouraging its engineers to use.
An internal document drafted for the meeting initially cited “GenAI-assisted modifications” as a contributing factor to a series of incidents dating back to the third quarter. The Financial Times, which reviewed both versions of the document, reported that this reference was removed prior to the meeting.
Amazon has disputed the reports. In a blog post, the company stated that only one incident involved AI tools, that “none of the incidents included AI-generated code,” and that the issue stemmed from “an engineer acting on incorrect guidance an AI agent derived from an outdated internal wiki.” Amazon also informed that the meeting was a standard weekly operations review, not an emergency session. Additionally, the company denied introducing new approval rules for engineers using AI tools and noted that AWS was not part of any of the incidents.
“As part of our regular operations, the meeting will cover a review of our website and app availability as we prioritize ongoing improvement,” an Amazon spokesperson told .
Internal documents obtained and reported by CNBC paint a different picture. Dave Treadwell, Senior Vice President of eCommerce Foundation, communicated to staff that site availability had been subpar lately, and the series of Sev 1 incidents— the most critical category for outages that disable key systems—required urgent attention.
However, CNBC reports that the original version of the internal documents reveals a more complex scenario. Treadwell admitted in his memo that “best practices and safeguards” for generative AI use had not been fully developed, and stated the company would implement “controlled friction” in deployments affecting the most vital aspects of the retail experience. Regardless of how Amazon frames it, engineers received the message that AI-assisted changes will now undergo greater scrutiny.
This type of acknowledgment comes at a challenging time for Amazon. The company, which recently overtook Walmart to lead the 500, is investing more in AI infrastructure than any other company globally—with projected capital expenditures of $200 billion this year.
Amazon is also actively reducing its workforce. The company let go of approximately 14,000 corporate employees in October—mostly middle managers—followed by another 16,000 in January. This adds to the over 27,000 staff cuts between 2022 and 2023. In June, Jassy wrote in an internal memo that Amazon would require fewer employees due to AI-driven “efficiency improvements,” reiterating his focus on an AI future with fewer workers at the retail giant. When the October layoffs occurred, Jassy shifted the reasoning during an earnings call to focus on “culture,” stating the company had expanded too quickly during the pandemic and needed to be “lean” and “agile.”
However, a separate Amazon memo about the same layoffs mentioned the need to adapt to “transformative technology”—language that aligns more clearly with AI-driven workforce cuts than a routine restructuring. Yet, regardless of the reasoning, Amazon appears to have realized it needs more human involvement in its processes.
This is a striking contrast to the prevailing narrative around AI-related layoffs. Jack Dorsey’s Block laid off nearly half its workforce last month—4,000 employees—and directly linked the decision to AI-driven productivity gains. Dorsey predicted most companies would reach the same conclusion within a year. Salesforce’s Marc Benioff stated he needed fewer employees after cutting 4,000 support positions. The executive consensus is that increased AI investment will offset costs by reducing workforce sizes.
Yet the promise that AI would ease workloads isn’t materializing—at least not for the remaining employees or the systems they oversee. A new analysis by ActivTrak of 164,000 workers, reported by the Wall Street Journal, found that AI is increasing the pace, volume, and complexity of work instead of reducing it. After adopting AI tools, employees spent more than twice as much time on email, messaging, and chat apps. Time dedicated to focused, uninterrupted work—the type needed to solve complex issues—dropped by 9%. Meanwhile, new research from Anthropic indicates a huge gap between what AI can theoretically automate and what it’s actually doing. Even in software and math, where 94% of tasks could theoretically be handled by AI, only around 33% are automated today. Anthropic noted that legal restrictions and organizational challenges are slowing deployment. Amazon’s outages might serve as a real-world example of why this is the case.