Memo: Amazon SVP Dave Treadwell says junior and mid-level engineers now require more senior engineers to sign off any AI-assisted code changes, after outages
Ecommerce giant says there has been a ‘trend of incidents’ linked to ‘Gen-AI assisted changes’ — Amazon's ecommerce business …
Financial TimesRafe Rosner-Uddin
Context & Ripple Effects
Amazon’s engineering organization had already been pushing wider AI use and higher output expectations, according to reports of increased pressure on engineers to use AI. It later directed engineers toward its in-house Kiro coding assistant rather than third-party alternatives, making AI-assisted development a more managed internal workflow.
The new approval rule is a corrective control: Amazon is preserving AI use while inserting senior review where Gen-AI-assisted changes have been associated with incidents. That matters because AWS developers were also being asked to take on new roles with AI assistance and added writing requirements, as described in the changing AWS developer workflow.
First-order effects
Junior and mid-level Amazon engineers must now route AI-assisted code changes through senior engineers, adding a formal review gate after the reported outages.
Senior engineers take on more accountability for production changes, while teams using AI coding tools face a slower path from generated code to deployment.
Second-order effects
Amazon’s internal AI coding rollout will be judged less by adoption than by whether it can operate within reliable review and incident-control processes; the earlier push toward Kiro gives the company a direct channel for that feedback.
The additional review workload can concentrate bottlenecks in senior engineering ranks, encouraging teams to distinguish low-risk AI assistance from changes that require deeper technical scrutiny.
Third-order effects
If similar controls persist, AI coding adoption in large software organizations is likely to shift from broad usage mandates toward traceable approval, ownership, and reliability standards.
The episode suggests that the value of coding assistants will increasingly depend on how well they fit existing engineering governance, rather than on code-generation speed alone.
The trend:Enterprise AI coding is moving from adoption-driven experimentation to operational governance shaped by reliability and accountable human review.
AI code is breaking Amazon systems at an increasing rate and it's such a problem that junior and mid-level engineers now aren't allowed to commit AI-assisted code without a senior engineer reviewing it. Is anyone surprised?
Amazon is holding a mandatory meeting about AI breaking its systems. The official framing is “part of normal business.” The briefing note describes a trend of incidents with “high blast radius” caused by “Gen-AI assisted changes” for which “best practices and safeguards are not […
there will be a cultural change at many software organizations soon where people declare bankruptcy on understanding the code they're committing. sooner or later this will cause a systems failure that will be harder to debug than most, but will be resolved anyways
> Junior and mid-level engineers can no longer push AI-assisted code without a senior signing off. Looks like Code Review tools are going to be a good business, and not the version Anthropic is shipping. You want something more like a precision knife vs a hammer. Command is a
You thought the alignment problem was the real problem of the AI era. It was always the verification problem. Do you have the skills to know if the output is correct? If not, you FUBAR.
Great reporting from @rafeuddin.ft.com on Amazon's misadventures with AI coding (which have coincided with big headcount reductions) www.ft.com/content/7cab... [image]
Amazon holds engineering meeting following AI-related outages: — Ecommerce giant says there has been a ‘trend of incidents’ linked to ‘Gen-AI assisted changes’ — www.ft.com/content/7cab... archive: archive.is/wXvF3 — l o l
genie is out of the bottle everyone hitting the magic button button puts your brain in a state of laziness that seeps into all your processes and they get skipped we talk about this all the time and still our team is struggling with it
Maybe it's my bad luck, but in the last week I hit major bugs on LinkedIn, Instagram, Amazon, Chrome, VS Code, TikTok, Slack, WhatsApp, and more. It feels the SDLC of yesteryear is not fit for the age of AI.
after this went wide, Amazon vociferously shat upon this story — which detailed the company's internal issues with humans using AI tools and accidentally taking down some of its services now they're making changes to deal with the problems that comms said they werent having [imag…
Many of my commenters are probably saying I told you so right about now. — The reality is that increasing coding velocity without also increasing your ability to review and test the code at that scale is a recipe for disaster.
more #hugops to my Amazon counterparts maybe this is the moment to invest properly in reliability engineering and stop putting SDEs oncall overnight for mission critical services
RE: https://social.coop/... Lmao. Gen AI is causing outages at AWS and Amazon causing business income loss. So now all AI created code needs senior developers to check it twice. So how is this saving time? #ai #llm #aws