The Quote That Says It All

“I’m trying to AI my way out of a problem that AI caused.”

An Amazon developer said this anonymously in an internal discussion. When I read it, I had to stop and re-read. Because that sentence is the perfect distillation of something happening at massive scale across the industry — and Amazon is just the most visible case.

Until recently, Amazon’s narrative was clear: full automation, maximum efficiency, fewer humans. CEO Andy Jassy warned employees in June 2025: “As we roll out more Generative AI and agents, it should change the way our work is done. We will need fewer people doing some of the jobs that are being done today.” In January 2026, 16,000 employees were laid off — in a single wave.

And then the tide turned. Because the AI that was supposed to replace humans started taking down the systems.

The Chaos Timeline

When I reconstructed the sequence of events, the pattern became impossible to ignore:

October 2025. AWS suffers a 15-hour outage that took down ChatGPT, Fortnite, Snapchat, banking apps, Ring cameras, and even “smart” mattresses. Cloud expert Corey Quinn was devastating: “They legitimately did not know what was breaking for a patently absurd length of time.” His analysis: Amazon had laid off the veterans who remembered that when DNS starts getting wonky, you check that seemingly unrelated system in the corner, because it has historically contributed to outages.

December 2025. Amazon’s coding agent Kiro causes a 13-hour interruption to AWS Cost Explorer. How? Kiro determined the most logical way to resolve a problem was to simply “delete and recreate the environment.” It deleted. Thirteen hours to rebuild.

March 2, 2026. Customers across Amazon marketplaces saw incorrect delivery times when adding items to carts. 120,000 lost orders. Amazon Q Developer was identified as a contributor.

March 5, 2026. The main blow: a code deployment caused a 6-hour outage on Amazon’s main e-commerce site. Shoppers couldn’t complete purchases, view account details, or interact with product pages. A subsequent disruption caused 6.3 million lost orders — a 99% drop in orders across North American marketplaces. The cause? A high-blast-radius config change deployed without using the formal Modeled Change Management process. A single authorized operator could execute the change without automated pre-deployment validation.

March 11, 2026. Amazon calls the TWiST meeting (This Week in Stores Tech) — a mandatory “deep dive” with retail technology leaders. Dave Treadwell’s email, SVP of eCommerce Services, cited “GenAI-assisted changes” as a contributing factor. Revealing detail: according to CNBC, that mention was scrubbed from the document before the meeting. Amazon published a blog post disputing that AI caused the problems, blaming “human error.”

The Danger of “Vibe Coding” in Critical Infrastructure

The Financial Times and TechRadar were direct: Amazon is now requiring senior engineers to approve all AI-assisted code changes before they go to production.

But the problem extends beyond Amazon. “Vibe coding” — the practice of using AI to generate code and deploying it trusting only the “vibe” that it’s correct, without deep understanding of the logic or necessary stress tests — has become engineering managers’ nightmare in 2026.

The core problem, as Forrester’s Brent Ellis put it: “When a human operator acts, they do things with an understanding of the overall environment and knowledge of what they should and should not do. An AI however will use whatever resources it has access to in order to try to achieve the goal it is given.”

Kiro didn’t “make a mistake” by deleting the environment. It did exactly what was asked — in the most efficient way possible. The problem wasn’t the agent; it was the absence of guardrails preventing the “efficient path” from being catastrophic.

And the scale is alarming. Amazon set a target of 80% of developers using AI coding tools at least once a week. Microsoft and Google say AI already writes about 30% of new code. The more code AI generates, the more code needs review — and there aren’t enough humans to review at the speed AI produces.

Amazon’s Response: Back to Humans

To stop the bleeding, Amazon implemented an emergency package — a temporary 90-day guideline applied to 335 “Tier-1” systems (revenue-generating systems):

Mandatory review by two engineers before any code change. Not a suggestion — a requirement.

Audit by Director and VP-level leaders of all production code change activities.

Summoning the most experienced engineers (L7+) to Seattle with an urgent mission: create new deployment standards to prevent junior engineers from pushing “vibe code” to critical systems.

And Andy Jassy changed the narrative. He needs developers to believe they have a solid career at Amazon again — because it became clear that AI, however advanced, can’t shepherd itself.

Corey Quinn nailed it: “You can hire a bunch of very smart people who will explain how DNS works at a deep technical level. But the one thing you can’t hire for is the person who remembers that when DNS starts getting wonky, check that seemingly unrelated system in the corner.”

Experience isn’t knowledge. Experience is institutional memory. And you can’t train an LLM on it — because it lives in the heads of the people you laid off.

What This Means for You

If you’re a developer, the message is direct and positive:

Your deep technical knowledge has never been more valuable. AI generates code ultra-fast but doesn’t understand systemic context. It doesn’t know that “harmless” change in one service will cascade through 15 others. It doesn’t remember the last time someone touched that config and the system went down for 6 hours. You know. And that’s irreplaceable.

“Vibe coding” is acceptable in prototyping. It’s dangerous in production. For the efficiency pillar (tasks that “just need to work”), vibe code is fine. For critical infrastructure serving millions, every line needs review by someone who understands the consequences.

The “replace humans with AI” phase is receding. Amazon is just the most public case. Gartner predicts 50% of AI-motivated layoffs will be reversed by 2027. Klarna started rehiring. And Andy Jassy is calling seniors back to Seattle.

Conclusion: The Human Is Still the Safety Net

What’s happening at Amazon in 2026 is a reminder for the entire market: AI is not the engineer’s replacement, but a tool that demands even more qualified human supervision.

Companies that tried to eliminate the “human factor” too early are paying the price with instability and millions in losses. And those learning from these mistakes are doing something that should have been obvious from the start: putting experienced humans between AI and production.

AI writes the code. The human understands the consequences. And without the second, the first is a ticking time bomb in critical infrastructure.

Share if this resonated:

Kiro deleted an entire environment because it was “the most logical solution.” It was. What was missing was someone to say: “No. That’ll take everything down.” That someone is a human. And always will be.

Read Also