Amazon owns up to needing more human oversight over AI code—unfortunately, it wants to do that with fewer people

Turns out relying on AI-assisted vibe coding or similarly generative tools for major tech infrastructure has some drawbacks—who could've foreseen this? Apparently, Amazon may have some regrets.

Earlier this week, the e-commerce team at Amazon arranged a "deep dive" meeting with a number of engineers to investigate what led to a series of outages. A briefing note for the meeting on Tuesday was seen by The Financial Times, and it describes a "trend of incidents" with a “high blast radius." Some but not all of the discussed incidents were tied to the use of AI coding tools; "Novel GenAI usage for which best practices and safeguards are not yet fully established," was listed as one contributing factor.

Though we don't know which incidents the group specifically discussed, internal communication suggests a broad scope. Dave Treadwell, the senior vice-president of Amazon's eCommerce Services team, is reported to have said to employees over email: "Folks, as you likely know, the availability of the site and related infrastructure has not been good recently."

Article continues below

The meeting follows last week's six-hour Amazon outage that was due to "a software code deployment" (via CNBC). Relatedly, multiple reports have placed the blame for two Amazon Web Services outages last year at the feet of an AI coding tool allowed to 'delete and recreate the environment' from scratch.

It's worth clarifying that AI coding tools aren't what caused the October 2025 outage that affected Fortnite, Roblox, Reddit, and many others—that was a Domain Name System (DNS) error. Amazon claims the two December 2025 outages were "extremely limited," with one affecting just a single service in parts of mainland China, and the other apparently not affecting anything AWS customers would see.

At any rate, Amazon isn't ditching AI coding tools; when these tools have been implicated in outages before, Amazon has been keen to frame it as an access issue rather than anything else. As such, the briefing note from this week's meeting reveals that junior and mid-level engineers at Amazon will now require a senior engineer to sign off on any AI-assisted changes.

While guardrails and human oversight make sense for any AI use, in this instance, it arguably just adds to the workload of a greatly reduced team; according to The Financial Times, Amazon has laid off more than 30,000 employees since October 2025. One longstanding employee paints a particularly damning picture, saying, "Day to day it just feels untenable. Our workload is increasing and the number of [problems] to deal with is just piling up. Some managers know this is the case, but executives just keep pointing to some bigger AI picture.”