Autonomous Failures

Amazon Fired 30,000 Engineers, Replaced Them With AI, and Lost 6.3 Million Orders in a Single Day

· 5 min read
Dark server room with tangled cables

On March 5, 2026, Amazon’s North American marketplace experienced a 99% drop in orders. Not a dip. Not a blip. Ninety-nine percent. In the span of a few hours, 6.3 million orders simply vanished into the digital void. Three days earlier, a separate incident had already caused 120,000 lost orders and 1.6 million website errors.

The cause? Code changes assisted by Amazon’s own AI tools. The company that wants to sell you AI is getting wrecked by AI. And the story of how Amazon responded is somehow worse than the outages themselves.

“Folks, as you likely know, the availability of the site and related infrastructure has not been good recently.”

— Dave Treadwell, Amazon SVP of E-Commerce Services, in an internal email that deserves a Pulitzer for understatement

The Timeline of a Slow-Motion Catastrophe

Here’s what happened, in order:

October 2025: Amazon lays off 14,000 corporate employees.

December 2025: Amazon’s AI coding tool Kiro deletes and recreates a production environment, causing a 13-hour AWS outage. (We covered this one already.)

January 2026: Amazon lays off another 16,000 employees. That’s 30,000 engineers and corporate workers gone in three months.

March 2, 2026: AI-assisted code changes cause incorrect delivery estimates across Amazon’s platform. 120,000 orders cancelled. 1.6 million website errors. Investigators trace it to Amazon Q, the company’s AI coding assistant.

March 5, 2026: A production change deployed without proper authorization triggers a 99% order decline across North American marketplaces. 6.3 million orders lost. In one day.

March 10, 2026: Dave Treadwell summons engineers to a mandatory “deep dive” meeting. Internal documents cite a “trend of incidents” with “high blast radius” related to “Gen-AI assisted changes.”

March 11, 2026: After the story breaks, Amazon deletes the GenAI reference from the internal document and tells reporters that only one incident involved AI and “none involved AI-written code.” Elon Musk, with characteristic brevity, responds: “Proceed with caution.”

You really cannot make this up.

Fire the Humans, Trust the Machines, Blame the Humans

Let’s appreciate the full absurdity of the sequence here.

Step 1: Fire 30,000 people.

Step 2: Tell the remaining engineers to use AI coding tools to pick up the slack. Set an 80% weekly adoption target. Track who isn’t using the tools enough.

Step 3: Watch the AI tools break the site. Repeatedly.

Step 4: Hold a mandatory meeting to address the AI-caused outages.

Step 5: Announce that the fix is… requiring more human oversight of AI code. The same humans you just fired 30,000 of.

This is not a corporate strategy. This is a Rube Goldberg machine for generating irony.

📋 DISASTER DOSSIER

Date of Incidents: March 2–5, 2026 Duration: Multiple outages across one week Primary Victim: Amazon’s North American marketplace Secondary Victims: Millions of customers who couldn’t buy things on the internet’s biggest store Tool Responsible: Amazon Q (AI coding assistant) Orders Lost: ~6.42 million across both incidents Website Errors: 1.6 million (March 2 incident alone) Official Response: “Only one incident involved AI” (after deleting the memo that said otherwise) Irony Level: 🌡️🌡️🌡️🌡️🌡️ (Thermonuclear)

The Cover-Up That Wasn’t

Here’s the part that elevates this from “big company has outage” to “you need to read this.”

Amazon’s internal documents — the ones prepared for the mandatory Tuesday meeting — originally listed “novel GenAI usage, for which best practices and safeguards are not yet fully established” as a contributing factor. This language appeared in the meeting agenda distributed to engineers.

After the Financial Times and other outlets published stories about the AI connection, Amazon edited the internal document to remove the GenAI reference. The company then told reporters that a single incident was related to AI and none of the incidents involved AI-written code.

An Amazon spokesperson characterized the meeting as a “routine weekly operations review focused on continuous improvement.” Yes. A mandatory emergency meeting summoned by the SVP of e-commerce, focused on a “trend of incidents” with “high blast radius,” is apparently what Amazon considers routine.

The 90-Day Safety Reset (a.k.a. Admitting the Problem)

Despite the public denials, Amazon’s internal response tells a different story. The company has ordered a 90-day safety reset targeting approximately 335 critical Tier-1 systems. The new rules:

  • Two-engineer review required before deploying any code changes to critical systems
  • Formal documentation and approval for all modifications to production infrastructure
  • Stricter automated validation for broad configuration changes
  • Senior engineer sign-off required for AI-assisted changes made by junior and mid-level engineers

Read that last one again. Junior engineers now need a senior engineer to approve their AI-assisted code before it goes live. The problem is that Amazon just fired a significant chunk of its senior engineers. Who, exactly, is going to do all this reviewing?

The $200 Billion Contradiction

Here is Amazon’s position, stated as clearly as possible:

  • AI coding tools are the future. Amazon is spending $200 billion in capex in 2026 to prove it.
  • AI coding tools just caused the worst e-commerce outage in the company’s recent history.
  • The fix for AI coding problems is more human oversight.
  • Amazon has eliminated 30,000 humans in the last six months.

This is what happens when the AI hype cycle meets operational reality at 200 miles per hour. You can’t simultaneously bet the company on AI tools and admit those tools need constant human babysitting and fire the humans doing the babysitting. At some point the math stops mathing.

Lessons for the Rest of Us

  • “Vibe coding” is not a deployment strategy. AI-assisted code still needs the same rigor as human-written code. Maybe more, because it’s harder to understand code you didn’t write.
  • If you delete the memo, we will find the memo. The internet has a long memory and the Financial Times has good sources.
  • You cannot solve an AI oversight problem by reducing the number of overseers. This is not complicated. It is apparently very hard for executives to understand.
  • When Elon Musk is the voice of caution in your story, you have made some serious mistakes.

Sources: Fortune, CNBC, News9, The New Stack, Tom’s Hardware, The Register. Amazon declined to provide additional comment beyond its public statements. The 6.3 million lost orders were also unavailable for comment.