The gap between what AI can technically do and what companies are actually willing to think through before deploying it has become the defining tension of the week. From Amazon employees gaming internal AI leaderboards into uselessness to ClickUp cutting 22 percent of its workforce based on AI agents that executives don't fully understand, the pattern is clear: speed of adoption is outpacing quality of decision-making. Aaron Levie's term "AI psychosis" captures something real. The people signing off on AI-driven layoffs are demonstrably not the people who understand the jobs being eliminated. Meanwhile, Cognition's Scott Wu insists that Devin, his AI coding agent, isn't meant to replace human programmers, yet the market is already pricing in replacement scenarios. The disconnect between what builders say their tools do and what executives are using them to justify has become a liability.
Training data acquisition is normalizing surveillance as a transaction. A startup now offers free home cleaning in exchange for the right to record it all for robot training. This is not presented as a privacy trade-off to debate; it's presented as a deal. The economics are straightforward: human labor is cheap, but labeled video of human labor is valuable. Separately, Google's Gemini Spark was given access to a user's emails, documents, and calendar to plan a birthday party, and it still failed at the basic task. The infrastructure for invasive data access is being built and deployed faster than the systems using it are proving competent. The mismatch matters because the surveillance is permanent and the competence is contingent.
Layoffs framed as AI optimization are now coordinated enough to register as a labor moment. Wikipedia editors are organizing a strike over Wikimedia layoffs. Chinese courts have begun enforcing rules that bar AI-justified terminations. A UK think tank backed by the Trades Union Congress called for workers to have real say in AI rollout decisions. Amazon employees gamed their internal AI token leaderboard so aggressively that management killed the program. None of these movements are connected, yet they all happened this week. The signal is that workers are no longer accepting "AI made us do it" as a closing argument. Simultaneously, Anthropic closed a $65 billion Series H round while funding for other companies slowed. Capital is concentrating, but labor resistance is spreading.
Sloane Duvall