Amazon is reassessing how its engineering teams use AI-assisted development tools following several outages that affected key parts of its retail website.

Reports indicate the company experienced four major service disruptions within a single week. One of the incidents reportedly lasted around six hours and prevented customers from accessing product pricing, account information and checkout functions.

The outages prompted internal discussions about whether tools that assist developers using generative artificial intelligence may have contributed to the problems.

The incidents were reportedly examined during a routine operations meeting led by Dave Treadwell, senior vice president of ecommerce foundation at Amazon. Internal documents referenced in reports suggested that AI-assisted development changes had initially been flagged as a potential factor in a broader pattern of outages that had been occurring since the third quarter.

Amazon has since challenged that interpretation of events. In a public statement responding to the reports, the company said only one incident involved the use of AI tools and that none of the outages were caused by code generated by artificial intelligence.

According to the company, the issue occurred when an engineer acted on inaccurate advice produced by an AI system. The AI assistant had drawn information from an outdated internal wiki, which resulted in incorrect guidance.

A company spokesperson said the disruption was not related to AI-written software but instead to an engineer following guidance generated from outdated internal documentation.

Amazon also noted that the meeting where the outages were reviewed was part of its normal operational review process rather than a special investigation triggered by the incidents.

Even though the company disputes that AI-generated code caused the outages, Amazon acknowledged that its internal controls for the use of generative AI tools are still evolving.

Reports citing internal communications suggest the company may introduce additional verification steps when developers deploy changes created with the assistance of AI. These safeguards, reportedly described internally as controlled friction, would add extra review layers before updates are applied to critical parts of the platform.

Such systems include some of the most sensitive elements of Amazon’s online retail operations, including the infrastructure responsible for product pricing, customer accounts and checkout services.

The discussion around development practices comes at a time when Amazon is increasing its investment in artificial intelligence technology. The company is expected to spend about A$307 billion on capital investments this year as it expands its AI infrastructure and data centre capabilities.

At the same time, Amazon has been restructuring parts of its workforce. The company reduced its corporate staff by roughly 14,000 positions in October and announced an additional 16,000 layoffs in January. These reductions follow earlier job cuts of more than 27,000 employees across 2022 and 2023.

Chief executive Andy Jassy has previously indicated that improvements in productivity driven by artificial intelligence could eventually allow the company to operate with a smaller workforce.

The recent outages highlight the balancing act facing large technology companies as they adopt new AI-assisted development tools while maintaining the reliability of systems that support millions of daily users.