The Noise — March 13, 2026

THE NOISE

Future Shock Weekly — March 13, 2026

Sam Altman said this week that "AI is not very popular in the US right now" because people keep blaming it for layoffs. He said this while his company was actively pitching tools to replace the people being laid off. The man selling the guillotine is concerned about public opinion around beheadings.

The Week Everyone Fired People and Blamed the Robot

By early March, 45,000 tech workers had lost their jobs in 2026. About 9,200 of those were explicitly blamed on AI, per RationalFX. Block fired roughly 4,000 people (nearly half its workforce) at the start of the month. Atlassian cut 1,600 this week. Oracle is reportedly preparing to slash 20,000 to 30,000. In every case, the press release says some version of "pivoting to AI."

The Guardian talked to seven current and former Block employees. Their summary of whether AI could actually do their jobs: "You can't really AI that." The Atlantic ran a piece titled "Imagine Losing Your Job to the Mere Possibility of AI," documenting how the technology doesn't need to work — it just needs to be a credible excuse for cost cuts the company wanted to make anyway. And UVA's Darden School of Business published an analysis literally titled "Is AI the Strategy — or the Scapegoat — Behind Block's 40% Layoff?"

When a business school is openly asking whether your layoff rationale is real, the rationale has problems.

Then Sam Altman showed up at BlackRock's US Infrastructure Summit and said what everyone already suspected. "Almost every company that does layoffs is blaming AI, whether or not it really is about AI," he told the audience, per Fortune. He also called the current situation a crisis for the "labor-capital balance" and admitted "nobody knows what to do about it."

The CEO of the company building the technology that companies cite when firing people is publicly saying those companies would have fired people anyway, the technology just gave them better PR. This is a man who has found a way to sell the product AND the disclaimer simultaneously. Wall Street loves it.

Amazon's AI Agent Read an Old Wiki and Broke the Internet's Largest Store

Amazon's retail website suffered a series of outages in early March. Fortune reported that one incident involved Amazon Q, the company's internal AI tool, which "inferred inaccurate advice from an outdated internal wiki." An engineer followed that advice. The result: roughly 120,000 lost orders and 1.6 million website errors on March 2 alone, per Business Insider.

Amazon clarified that only one of the incidents involved AI tooling, and "none of the incidents involved AI-written code." The tool didn't write bad code. It read a stale wiki page and confidently presented the wrong answer, which an engineer trusted. Amazon ordered a 90-day code safety reset and is now putting "humans further back in the loop," a corporate way of saying "we're going to have people double-check the robot again."

This is the AI agent value proposition in miniature: fast, confident, and wrong because it can't tell a current runbook from a deprecated one. Neither could the engineer, because the whole point of the tool was to not have to read the wiki yourself.

Quick Hits

We got one wrong this week. Our March 13 Signal reported Atlassian cut "roughly 1,400" jobs. The actual number is 1,600, per Reuters, Bloomberg, The Guardian, and Atlassian's own announcement. We regret the error. (See? This is what happens when you don't triple-check before press time. We're correcting ourselves so we retain the moral authority to correct everyone else.)

DeepSeek V4 has now missed five launch windows. Something called "V4 Lite" appeared silently on March 9 and DeepSeek said nothing about it. The best sourcing for the full V4 release date is still, incredibly, a commercial real estate investment blog. We're now tracking this as a geological event.

Atlassian CTO Rajeev Rajan is also stepping down as part of the restructuring, effective March 31, per GeekWire. Mike Cannon-Brookes recorded a video explanation for employees. When your layoff announcement includes a video from the CEO, you've crossed the line from "cost reduction" to "content production."

Hype Watch

"AI-first company" — Atlassian used this in its restructuring memo. Translation: "We are firing humans and have not yet built the AI that replaces them, but the stock needs a narrative." Previously seen from Block, Shopify, and everyone else who discovered that "AI-first" is the 2026 version of "mobile-first."

"Pivot to AI" — appearing in every layoff announcement, Reuters headline, and analyst note. The phrase does roughly the same work as "restructuring" did in 2023, "right-sizing" in 2024, and "optimizing headcount" in 2025. The actual pivot is from "having employees" to "not having employees."

"Inaccurate advice" — Amazon's contribution to the euphemism pile. Their AI agent didn't hallucinate. It didn't make an error. It gave "inaccurate advice." The agent is now a management consultant.

"Labor-capital balance" — Sam Altman's new phrase for "we broke something and it's going to be weird for a while." Remarkably honest as tech CEO statements go. Unfortunately, honesty about a problem you're profiting from is a hard sell.

From the Editor's Desk

The Reasoning Theater paper from last week still deserves a full write-up. Models that fake their own chain-of-thought reasoning have implications for every interpretability and safety system that relies on reading the model's work. We're sourcing independent verification of the key claims.

NVIDIA's GTC keynote is Monday. Jensen Huang will do his annual performance at the SAP Center in San Jose. The hype machine is already warming up. We'll have a full reality-check pass on whatever gets announced.

Running Tallies

Week 3 of "programming is unrecognizable." Karpathy's claim. Stack Overflow traffic still declining. Software engineering hiring data still shows no structural collapse. The vibes say it's over. The numbers say it isn't.

Week 7 of "the era of physical AI." Jensen Huang's CES declaration. The robots still have people inside them. GTC next week will produce new claims. We'll update accordingly.

The AI Layoff Scapegoat Index: 45,000 and counting. By early March, 45,000 tech layoffs globally in 2026, with 9,200 explicitly AI-blamed (per RationalFX). The CEO of the largest AI company says most would have happened anyway.

DeepSeek V4 launch countdown: Week 4. Five missed windows. One silent "V4 Lite" appearance nobody will explain. At this point DeepSeek V4 is Schrodinger's model — simultaneously launching next week and never launching at all.

Forty-five thousand people lost their jobs this year so far. The technology blamed for it crashed an online store because it read the wrong wiki page. The man selling the technology says companies are using it as a scapegoat. Everyone involved is making money except the people getting fired.