Amazon is reportedly endeavoring to bolster its internal safeguards for coding practices following a series of outages that beleaguered its e-commerce operations. Notably, one such incident is allegedly linked to its AI coding assistant, Q.
“We are instituting temporary safety protocols that will introduce deliberate friction to modifications in pivotal segments of the Retail experience […]
Concurrently, we will invest in more sustainable solutions that encompass both deterministic and agentic safeguards,” stated Dave Treadwell, Amazon’s Senior Vice President of e-commerce services, as reported by Business Insider.
Treadwell addressed staff during an internal gathering on March 10, reflecting on the “trend of incidents” that have surfaced since the latter half of 2025, including “several major” disruptions in recent weeks, according to a briefing document referenced in the report.
In response to these challenges, Amazon will demand that engineers document code alterations with heightened rigor and secure additional approvals. Furthermore, new safeguards are expected to be integrated into the code-review process.
The deep-dive meeting on Tuesday was convened subsequent to significant disruptions predominantly affecting the United States earlier in the week.
One of the contributory factors identified was “novel GenAI usage for which optimal practices and safeguards have yet to be fully established,” according to a report by the Financial Times.
Additionally, a 13-hour Amazon Web Services (AWS) outage in December 2025 has been linked to the company’s AI coding tool, Kiro, as per internal documentation reviewed by the business daily.
This conundrum at Amazon reveals a mounting dichotomy in contemporary software development, wherein engineers are under immense pressure to deploy AI-assisted code while appropriate systems to assess and validate the burgeoning volumes of AI-generated outputs remain inadequately instituted.
This scenario unfolds against the backdrop of widespread layoffs in the tech sector, with companies eager to harness AI tools to streamline operations and curtail expenses.
In January 2026, Amazon disclosed plans to eliminate over 16,000 corporate positions. Although the company has refuted claims that these job cuts have exacerbated service disruptions, Amazon engineers have indicated an uptick in ‘Sev2’ incidents, necessitating rapid intervention to avert product outages.
An Amazon spokesperson clarified that Tuesday’s meeting formed part of a routine weekly assessment. “In the usual course of business, the meeting encompasses a review of our website and app’s functionality as we continuously strive for improvement,” the spokesperson stated.
They further contested that AWS was involved in any of the recent outages, indicating only one incident under review related to AI.
When did the outages occur?
On March 2, 2026, numerous users reported discrepancies in delivery timings when adding items to their shopping carts on Amazon’s platform. Consumers also voiced frustration over difficulties in completing transactions and accessing essential functionalities such as account details and product prices.
This nearly six-hour disruption culminated in a staggering 120,000 lost orders and approximately 1.6 million website errors. An internal review attributed the bulk of the disturbances to its AI coding tool, Q.
A few days later, on March 5, 2026, Amazon’s marketplace endured another outage that triggered a 99 percent decline in orders across the U.S., resulting in the loss of 6.3 million orders. A deployment modification executed without approval was pinpointed as a primary factor in this occurrence.
Separately, Amazon’s AWS segment reportedly experienced at least two outage incidents tied to the utilization of AI coding assistants.
One such incident involved a 13-hour disruption that incapacitated a customer cost calculator, traced back to alterations made by engineers leveraging Amazon’s proprietary Kiro AI coding tool, which reportedly opted to “delete and recreate the environment.”
Under the newly instituted policy, Amazon engineers must obtain two peer reviews of their work prior to any coding alterations.
They are also mandated to employ an internal documentation and approval system alongside an automated coding framework that strictly adheres to Amazon’s central reliability engineering protocols.
Additionally, Amazon has instructed the custodians of 335 Tier-1 systems, along with their director- and VP-level leaders, to conduct audits of all production code change activities within their respective domains. Tier-1 systems provide services that can exert direct influence on consumers.
Source link: Indianexpress.com.
