|
|
|
// YOUR AI EDGE
META TRACKED EMPLOYEES’ EVERY KEYSTROKE TO TRAIN AI, THEN CUT 8,000 OF THEM
Plus: Three CEOs killed an AI safety order with one phone call, Google's new agent never sleeps, OpenAI's math proof has a footnote problem, and Anthropic is spending $1.25 billion a month on compute.
|
|
// TODAY'S SIGNAL
|
META USED ITS OWN PEOPLE AS AI TRAINING DATA. THEN IT FIRED 8,000 OF THEM.
Leaked audio from a Meta all-hands meeting on April 30 confirmed what employees had suspected for months. The company had been tracking keystrokes, Gmail activity, Google Chat messages, and VS Code usage through mandatory monitoring software called MCI. The goal: train internal AI models on how Meta employees actually work. Three weeks later, 8,000 of those employees got laid off. Zuckerberg’s memo to staff said “success isn’t a given” and framed the cuts as a pivot toward AI-focused roles. Another 7,000 employees were reassigned into AI positions. Meta reported $27 billion in Q1 profit the same quarter it announced $125 billion in AI infrastructure spending for 2026. The math is blunt. Meta’s median employee compensation dropped from $417,400 to $388,200. The company is paying its top AI researchers $100 million a year. Everyone else is getting surveilled, shuffled, or shown the door. Sixteen current and former employees spoke to Wired on condition of anonymity. One described morale as “historically, horrifically low.” UK-based staff have started a unionization effort.
|
// THE REAL STORY
The surveillance-to-layoff pipeline is the new playbook, and Meta just ran it at scale. Document how you work before someone else does. Build a portfolio of your outputs, decisions, and results that lives outside company systems. If your employer is studying your workflow, you should have studied it first.
|
|
|
|
// SHORTCUT
Copy, paste, go:
|
"I need to update my LinkedIn headline and summary to signal AI fluency without overpromising. My current role is [YOUR CURRENT TITLE] at [COMPANY] in [INDUSTRY]. I want to attract recruiters looking for people who can work alongside AI tools, not just use them casually. Rewrite my headline (under 120 characters) and a 3-sentence summary that positions me as someone actively building AI into my workflow. Keep it specific, not buzzy."
|
|
|
TRUMP KILLED AN AI SAFETY ORDER AFTER THREE CEOS CALLED HIM DIRECTLY
On May 21, the White House had already sent invitations for an executive order signing ceremony. The order would have created a voluntary 90-day review framework for AI models before public release, with partnerships between Commerce Department and companies like Anthropic, OpenAI, and Google. Hours before the event, Trump called it off. The reason: direct calls from Mark Zuckerberg, Elon Musk, and David Sacks. Trump told reporters, “I didn’t like certain aspects of it. We’re leading China, and I don’t want to do anything that’s going to get in the way.” The Commerce Department’s partnership agreements quietly disappeared from its website the same day. The last federal AI safety framework (Biden’s October 2023 executive order) was already revoked in January 2025. The U.S. now has zero federal oversight of AI model releases.
|
// CONNECT THE DOTS
Federal AI regulation is dead for at least two years. Companies will move faster, deploy more aggressively, and self-govern (or not). Knowing which AI tools you interact with, what data you feed them, and what permissions you grant is now entirely a personal responsibility. Read the terms of service. Seriously.
|
|
|
GOOGLE’S NEW AI AGENT RUNS 24/7, EVEN WHEN THE PHONE IS OFF
Google I/O 2026 had 13 major announcements. One stood out. Gemini Spark is a cloud-based AI agent that runs on dedicated Google Cloud VMs around the clock. It connects to Gmail, Calendar, Maps, Docs, and third-party apps like Canva and Instacart. It works in the background, scheduling meetings, pulling research, managing tasks, and taking actions across services while the user sleeps. The broader numbers: Gemini now has over 900 million monthly active users (more than doubled in a year). AI Mode in Google Search crossed 1 billion monthly users. Alphabet’s projected capex for 2026 could hit $190 billion. Google also launched vibe-coding in AI Studio, letting anyone build a native Android app from a text prompt and publish it directly to the Play Store. The new Gemini 3.5 Flash model is roughly 4x faster than competitors in benchmarks.
|
// THE MOVE
If Google is building agents that work while you sleep, the skill to develop is knowing how to direct them. Set up Google’s AI tools in your workflow now (even the free tiers) and practice giving structured, multi-step instructions. The people who learn to manage AI agents early will outrun the people waiting for a single magic prompt.
|
|
|
| |
|
OPENAI SAYS IT SOLVED AN 80-YEAR MATH PROBLEM. MATHEMATICIANS HAVE QUESTIONS.
OpenAI announced its reasoning model produced an original 125-page proof disproving a 1946 conjecture by Paul Erdos about unit distances in the plane. Fields medalist Tim Gowers called it “a milestone in AI mathematics.” Mathematicians Noga Alon, Melanie Wood, and Thomas Bloom endorsed the result. The footnote: OpenAI already burned through credibility on this topic once before. Earlier this year, a researcher claimed GPT-5 had solved a separate math problem. The company later retracted the claim. The “for real this time” in TechCrunch’s headline was earned. Whether the proof holds up under extended peer review will take months. But the pattern is worth watching: AI models are starting to produce original proofs, not just verify existing ones.
|
// TRANSLATION
“AI solved a math problem” sounds abstract. The practical version: if AI can produce original proofs in formal mathematics, it can eventually produce original analysis in business, law, and strategy. The gap opening up is between people who can evaluate AI-generated reasoning and people who accept it uncritically. Be the first kind.
|
|
|
ANTHROPIC IS PAYING SPACEX $1.25 BILLION PER MONTH FOR AI COMPUTE
SpaceX’s IPO filing revealed that Anthropic committed to paying $1.25 billion monthly through May 2029 for access to compute infrastructure, totaling roughly $45 billion. The deal includes access to Colossus 2, a facility packed with Nvidia GB200 chips, with a 90-day exit clause. For context: SpaceX’s total annual revenue is around $18 billion. A single AI company is paying them nearly that amount every year just for server access. Anthropic is reportedly approaching a $900 billion valuation and may be preparing its own IPO. The AI infrastructure arms race has a new price tag, and it makes the dot-com era’s capital spending look quaint.
|
// CONNECT THE DOTS
When one AI company spends $45 billion on compute alone, the industry is pricing in a future where AI does dramatically more than chat. The companies making these bets believe they’re building something worth trillions. Whether they’re right or wrong, the sheer scale of spending means AI infrastructure (data centers, chips, energy) will be the backbone of the next decade’s economy. Learning how these systems work is career insurance.
|
|
|
| |
|
// THE TOOLBOX
CURSOR
Cursor is VS Code with an AI copilot bolted into every surface. Think of it as a code editor that read your entire codebase and can write, refactor, and debug alongside you in real time. It understands context across files (not just the one open in front of you) and can generate full functions from plain-English descriptions.
Honest caveat: The free tier is generous but the Pro plan ($20/month) is where the real power lives. If you never write code, this one’s not for you yet.
|
|
// WAIT... DOES THIS ACTUALLY WORK?
THE OBSCURE PROMPT OF THE DAY
prompts nobody asked for. results nobody expected. try it anyway.
|
"You are a corporate buzzword translator. I’m going to paste an email from leadership and you will rewrite it in plain English, then add a section called ‘What They Actually Mean’ that decodes every piece of corporate speak into honest, blunt language. Be funny but accurate."
|
// OUR VERDICT
Fed it a real all-hands email from a Fortune 500 company. The BS density score was an 8.7. The rewrite turned “leveraging synergistic cross-functional alignment to drive stakeholder value” into “getting different teams to work together so the company makes more money.” The 12-year-old version was genuinely clearer than the original. HR should hire this prompt.
SURPRISINGLY PRACTICAL: ★★½
|
|
// YOUR EDGE
|
01
Learn this: How Google’s Gemini Spark agent works and what it can connect to. This is the first always-on AI agent from a major platform, and it sets the template for what’s coming everywhere else.
|
|
02
Watch this: The Wired investigation into Meta’s employee surveillance program. Sixteen sources. The reporting is detailed enough to be a playbook for what to watch for at other companies.
|
|
03
Say this: “Meta tracked how its employees worked, used that data to train AI, then laid off 8,000 people. The surveillance-to-layoff pipeline is now a documented corporate strategy.”
|
|
|
// GOT THIS FROM A FRIEND?
Your edge on AI, twice a week. Free forever.
Subscribe →
|
|
|
|
