# Amazon Convenes ‘Deep Dive’ Internal Meeting to Address Outages: AI Tools Under Scrutiny
Amazon has called engineers to an urgent internal meeting to tackle a series of recent service outages, with reports highlighting **generative AI-assisted code changes** as a potential factor in these “high blast radius” incidents.[1][2][3] Held on March 10, 2026, the session—dubbed a “deep dive” by some outlets—aims to review operational failures and implement stricter safeguards, amid growing concerns over AI’s role in the company’s infrastructure.[1][4]
## A Spate of Disruptions Rocks Amazon’s Retail Operations
Amazon’s e-commerce platform has faced multiple high-impact outages in recent weeks, severely affecting customer experience and revenue. On March 2, users across marketplaces encountered incorrect delivery times when adding items to carts, resulting in nearly **120,000 lost orders** and about 1.6 million website errors.[3] Internal reviews pinpointed Amazon’s **AI coding assistant Q** as a key contributor, exposing vulnerabilities in control plane operations where “guardrails do not exist.”[3]
The disruptions continued on March 5, when a **99% drop in orders** hit North American marketplaces, leading to 6.3 million lost orders.[3] This stemmed from a production change deployed without formal documentation or approvals via Modeled Change Management, lacking automated pre-deployment validation and relying on a single operator for a high-risk config update.[3] Earlier, a six-hour outage on Amazon’s main retail site prevented customers from viewing details or completing transactions, traced to an erroneous code deployment.[1]
These incidents follow a “trend of incidents” since Q3 2025, including “several major” events in recent weeks characterized by broad propagation due to inadequate safeguards in control planes—which guide data flows across networks.[3] Amazon Senior Vice President of e-commerce services **Dave Treadwell** addressed the issues in an email, noting poor site availability and announcing the meeting for a “deep dive into some of the issues that got us here as well as some short immediate term initiatives.”[1][2]
## Generative AI in the Spotlight: Opportunity or Risk?
Reports from *Financial Times* briefing notes link these problems to **”Gen-AI assisted changes”**, where tools like Amazon’s Q and Kiro AI generated code without fully established best practices or safeguards.[1][3][4] One document warned that “GenAI’s usage in control plane operations will accelerate exposure of sharp edges,” urging investments in safety measures.[3]
Amazon pushes back on claims that AI is the primary culprit. A spokesperson emphasized the March 10 meeting as part of the regular weekly **TWiST operations review** for retail tech leaders, focused on “continual improvement” of website and app availability—not an admission of AI-driven failures.[1][2][4] They clarified that only one reviewed incident involved AI, none featured AI-written code, and AWS was unaffected.[3][4] Regarding a prior December AWS outage and February’s Cost Explorer issue in China, Amazon attributed problems to user errors like misconfigured access controls, insisting no evidence shows AI tools increase incident frequency.[4]
Critics remain skeptical. Cloud economist Corey Quinn argued AWS prefers blaming engineer incompetence over admitting AI mistakes.[4] Industry voices, including former AWS distinguished engineer James Gosling, link outages to layoffs in stability-focused teams amid AI-driven cost-cutting, such as 14,000 job reductions last October.[4]
## New Controls and a 90-Day Safety Reset
In response, Amazon is rolling out temporary measures to introduce **”controlled friction”** into code deployment. Treadwell outlined a **90-day safety guideline** addendum targeting 335 **Tier-1 systems**—consumer-impacting services owned by VP-level orgs that suffered repeated incidents.[3]
Key changes include:
– Requiring **two-person reviews** for all coding changes.
– Mandatory use of internal documentation, approval tools, and automated systems enforcing central reliability rules.
– Audits of production code activities by Tier-1 owners and Director/VP leaders.[3]
AI-assisted changes now demand senior engineer approval before deployment, per some reports—though Amazon denies mandating junior/mid-level sign-offs specifically for AI work.[1][3] Longer-term, the company plans “deterministic and agentic safeguards” to bolster durability.[3]
## Broader Implications for AI in Tech Giants
This episode underscores the double-edged sword of generative AI in software engineering. Tools like Q promise efficiency but amplify risks when scaled across vast infrastructures without mature protocols.[1][3] Amazon’s moves signal a pivot toward caution, prioritizing reliability over unchecked acceleration—especially as competitors like Anthropic launch AI code review tools amid similar debates.[4]
For consumers, these outages highlight fragility in the e-commerce giant’s backbone, eroding trust amid lost orders and downtime. Investors watch closely: while Amazon downplays AI’s role, internal documents reveal urgency to fix “high blast radius” flaws before they cascade further.[3]
As the 90-day reset unfolds, it will test whether these friction-heavy policies restore stability without stifling innovation. Amazon’s response—blending denial with action—reflects a tech sector grappling with AI’s promise and pitfalls, where “best practices…are not yet fully established.”[1]
(Word count: 812)
Original source: CNBC Business – Amazon convenes ‘deep dive’ internal meeting to address outages
Leave a Reply