The Week Workers Sued, Struck, and Got Re-Interviewed

Lawsuits, strikes, and rebuilt interviews — the week AI's people decisions became contested terrain.

I was doing tech support the day Yahoo went live, and I can tell you the internet didn't feel real to most companies until it started showing up in job descriptions and org charts. AI crossed that line this week. In the span of about 48 hours, 26 Meta employees sued over AI-assisted layoff selection, Hyundai workers staged the auto industry's first strike over humanoid robots, and Coinbase published a year of lessons from rebuilding its engineering interviews after AI started writing most of its code. The tools stopped being the story — the people decisions around them are.

On July 14, 26 Meta employees filed suit alleging that AI systems — keystroke monitoring, activity dashboards, and algorithmic performance rankings — helped select them for the roughly 8,000-person layoff Meta announced in May, with separations set to begin July 22. The core claim is simple: output metrics can't accumulate while you're on medical, parental, or family leave, so the rankings quietly penalized exactly the people federal law protects. About half the plaintiffs took caregiving or pregnancy-related leave. Meta says the allegations "lack merit" and that workforce decisions "were and are made by people, not AI."

A day later and 7,000 miles away, Hyundai workers in South Korea walked out over plans to put humanoid robots on factory floors — the first labor action in auto industry history aimed squarely at humanoids. The twist: Hyundai owns Boston Dynamics — its workers are striking against robots the company builds — while BMW and GM run humanoid pilots of their own. And on the third front, Coinbase published the receipts on what AI does to hiring: AI-generated code went from 5.7% of its codebase in Q1 2025 to roughly 80% by Q1 2026, so the company rebuilt its interview loop from scratch. An internal audit found an 84% correlation between two of its interview rounds — one of them was measuring almost nothing new — and the replacement tests how candidates direct AI, debug real repositories, and catch AI-generated errors.

Three fronts, one pattern. For two years, "AI in the enterprise" mostly meant productivity software and pilot programs. This week it meant a disparate-impact complaint, a work stoppage, and a hiring system redesigned end to end. The legal theory in the Meta case matters well beyond Meta — the plaintiffs aren't arguing the AI was wrong, they're arguing nobody checked who the metrics penalized before acting on them. That's a process failure, and process failures are what courts, regulators, and unions know exactly how to punish.

Here's the thing: your company almost certainly uses AI somewhere near hiring, scheduling, performance, or layoffs — even if it's just a dashboard somebody exports before calibration meetings. This week set the risk profile for all of it. The question your board will ask isn't "did the model work?" It's "who owned the decision, and what did we check before we acted on it?" The companies that can answer that get to keep automating. The ones that can't get to meet the plaintiffs' lawyers.

Read the full story →

That's TSMC's total U.S. commitment after this week's announcement — $100 billion added to its Arizona buildout in a single quarter. TSMC posted record Q2 revenue of $40.2 billion, up 36% year over year, with net income up 77.4%. The chips powering every AI model you use are increasingly being made on American soil — and the scale of that bet is hard to overstate.

Source: TSMC Adds $100B Arizona Investment as Q2 Profit Surges 77% — The New York Times

TSMC reported record Q2 revenue of $40.2 billion, up 36% year over year, with net income up 77.4% — then announced an additional $100 billion for Arizona, funding four more advanced fabs and bringing its total U.S. commitment to $265 billion. If your AI roadmap depends on compute, the supply chain behind it just got a lot more American — on a 2027-and-later horizon.

China's cyberspace regulator approved Apple Intelligence for launch after Apple agreed to run local models — Alibaba's Qwen, with Baidu also on board — ending a delay dating to its 2024 global debut. The stakes: Apple's Greater China revenue hit $20.5 billion in Q2, up 28%. For any company operating across borders, the lesson is that market access increasingly means running someone else's model, and data sovereignty is now product architecture, not a compliance footnote.

Mira Murati's Thinking Machines Lab released Inkling, a Mixture-of-Experts model with 975 billion total parameters (41 billion active), a context window up to 1 million tokens, and full weights on Hugging Face — plus a smaller 276B preview version. Trained on 45 trillion tokens across text, images, audio, and video, it posts frontier-adjacent coding and reasoning scores. For enterprise buyers, every serious open weights release is procurement leverage: capability you can run under your own roof resets the price of capability you rent.

Governor Kathy Hochul signed an executive order making New York the first state to impose a moratorium on new hyperscale data centers — up to one year, while the state studies grid strain, water use, and utility bills and holds discretionary permits. The signal is clear: AI infrastructure siting is now a political negotiation with communities, not just a real estate transaction, and other states are watching how this plays.

This week's tools are all about doing more with less — less tab-switching, less context-switching, and less money left on the table. Whether you're managing a team, running client calls, or just trying to find that one conversation from six months ago, these three are worth five minutes of your time.

PwC analyzed over a billion job advertisements across six continents and found companies most exposed to AI posting 40% higher productivity growth — while entry-level roles demanding senior-level skills have grown 35% since 2019. If this week was about workers pushing back on how AI decisions get made, this is the clearest data on what they're pushing against, and it's worth 20 minutes before your next workforce-planning conversation.