Episode Details

Dr. Venki Padmanabhan | The Long Game | March 2026

For The Long Game for All (thelonggameforall.substack.com)

∗ ∗ ∗

On Thursday, Amazon’s retail website locked shoppers out of checkout for six hours. Not a DDoS attack. Not a server fire in Northern Virginia. An AI agent followed advice from an outdated internal wiki, and the system that serves roughly 310 million active customers went dark in the middle of the afternoon.

The same week, Jensen Huang stood on a stage and told the world not to ask for ROI on AI. “Let a thousand flowers bloom,” he said, borrowing Mao’s most famous invitation to chaos.

I want to propose a different metaphor. Not a thousand flowers. One bloom. Specific, singular, already planted, already paid for—and, in Amazon’s case, already ripped out of the ground to make room for the machine that just failed.

∗ ∗ ∗

What Actually Happened

According to the Financial Times, an internal document prepared for Amazon’s emergency review identified “GenAI-assisted changes” as a contributing factor in a pattern of incidents stretching back to Q3 2025. Four high-severity failures hit the retail site in a single week. The internal document was then edited—the AI reference deleted—before the meeting took place.

Amazon’s public response was revealing in what it conceded while trying to minimize. The company said only one incident involved AI tools, and that the cause was “an engineer following inaccurate advice that an agent inferred from an outdated internal wiki.”

Read that sentence again. An agent inferred advice from an outdated wiki.

Someone used to keep that wiki current. Someone used to know, from years of working with the system, which pages were stale and which were gospel. Someone used to be the institutional memory that prevented a new engineer from following a three-year-old procedure on a live production environment.

That someone was already paid for. And Amazon cut them.

∗ ∗ ∗

The Pattern Nobody Wants to Name

Amazon laid off roughly 14,000 corporate workers in October 2025—mostly middle managers, the people who carry institutional knowledge in their heads the way a root system carries water. Another 16,000 followed in January. That’s on top of 27,000 cut between 2022 and 2023. CEO Andy Jassy wrote in an internal memo that the company would need “fewer employees thanks to AI-driven efficiency gains.” When the October cuts came, he reframed the rationale as “culture”—the company had grown too fast, needed to be “lean.”

(There’s that word again. Lean. The word manufacturing has been hiding behind for thirty years when it means: we cut the people and kept the machines.)

But a separate Amazon memo announcing the same layoffs cited the need to adapt to “transformative technology.” The real reason. The reason they always walk back. The reason that now has a six-hour checkout outage attached to it.

Here is the pattern: fire the people who know things, replace them with systems that don’t, then act surprised when the systems infer from an outdated wiki because nobody is left to update it. It is not an AI failure. It is a bloom failure. Amazon ripped the living intelligence out of its own soil and expected artificial intelligence to photosynthesize in the dark.

∗ ∗ ∗

I Have Seen This Before

People will read this essay and wonder what a manufacturing guy knows about AI deployment at a trillion-dollar tech company. Fair question. Here is my answer: the pattern is identical. I have watched it play out not in server rooms but on assembly lines—and the economics of the failure are just as damning when you actually do the math.

During my Walden Pond years—two years on the third shift at GM’s Lansing Delta Township plant, launching three SUVs simultaneously—I witnessed