AI-assisted programming causes problems? Amazon experiences system failures four times in a week, executives urgently hold review meetings

E-commerce giant Amazon has recently experienced a series of website and app outages, prompting senior management to hold an emergency internal review meeting. CNBC reported that some of these incidents are related to AI-assisted coding. Against the backdrop of large-scale layoffs and accelerated adoption of AI technology, this has also sparked discussions about the risks and oversight of “AI coding” in large-scale systems.

Four major outages in one week, Amazon urgently convenes review meeting

CNBC reported that Amazon has recently experienced multiple website and application failures, including one incident that lasted about six hours, during which some users were unable to check out, view product prices, or log into their accounts.

Dave Treadwell, senior vice president responsible for Amazon’s e-commerce technical infrastructure, later sent an email to employees announcing that a deep review would be conducted during the company’s routine “This Week in Stores Tech (TWiST)” meeting to discuss the causes of the recent series of major system failures.

In the memo, Treadwell pointed out that four high-severity “Sev-1” level incidents occurred within the past week. Such incidents typically cause core service outages or significant performance degradation, making it necessary to reevaluate engineering processes and related mechanisms.

“AI-assisted coding” named as a main cause of the incidents, Amazon downplays the impact

The Financial Times disclosed that an internal document before the meeting mentioned that since Q3, some “Generative AI-assisted production changes” have been linked to the frequent recent incidents. However, this content was removed before the meeting.

A Amazon spokesperson later stated, “Only one recent incident was related to AI tools, and none involved AI-generated code. The TWiST meeting is just a routine operational review.”

Nevertheless, the company internally admits that current best practices and safety mechanisms for generative AI in software development are not yet fully established. They are working to strengthen governance measures, including adding review processes for AI-involved code changes and implementing more automation and proxy-based safety mechanisms to reduce future risks.

(Spanish engineer unexpectedly “took over” 7,000 DJI floor-cleaning robots, exposing security vulnerabilities in smart home devices)

Layoffs and investments go hand in hand as tech giants accelerate into AI-assisted era

In recent years, Amazon has significantly increased investments in AI and cloud infrastructure. In last month’s earnings report, the company projected capital expenditures reaching an unprecedented $200 billion this year to meet the rapidly growing computational demands of AI services.

While ramping up AI investments, Amazon continues to streamline its organization. Since 2022, the company has cut over 27,000 jobs, and in the past six months, about 30,000 roles related to business and technology have been eliminated, reflecting a trend of accelerating shift toward AI-assisted enterprise operations.

AI coding efficiency improves, but quality and stability become new challenges

As large language models (LLMs) are widely used in software development, more companies are employing AI tools to write code, modify systems, or assist in application deployment. However, AI creator @TukiFromKL worries that while AI-assisted coding can speed up development, lacking proper review and testing mechanisms may increase errors, system failures, and potential security risks:

Amazon has proven that senior executives who don’t understand AI are more dangerous than AI itself. And every company rushing to imitate this approach will likely learn no lessons from these incidents.

Amazon’s recent system failures highlight the contradictions faced by large tech companies in adopting AI to optimize workflows. While pursuing efficiency and automation, companies must also redesign engineering governance and quality control processes to ensure the stability and reliability of critical infrastructure.

(CZ mocks AI proxy tool “Lobster”: claiming to free your hands, but ends up busy fixing it every day)

Did AI-assisted coding cause the mishaps? Amazon’s four outages in one week prompt senior management to hold an emergency review. Originally published on Chain News ABMedia.