Workers are training data now.

Train, track, replace. ClickUp’s 22% workforce cut on May 21 took out roughly 290 of 1,300 employees and left the company running about 3,000 internal AI agents—a 3:1 agent-to-employee ratio. CEO Zeb Evans calls the new structure a “100x org” and is introducing million-dollar salary bands for the survivors. The company was last valued at $4 billion in 2021. A Gartner survey now finds 80% of firms using autonomous AI have already cut jobs. The agent-to-employee ratio is becoming the new headcount metric.

The workers being replaced are also the training set. Meta’s MCI surveillance program—revived this week by leaked April 30 all-hands audio—captures keystrokes, clicks, screenshots, and browsing across Gmail, GChat, Metamate, and VS Code on every work laptop. The leak landed the same week Meta announced 8,000 job cuts, roughly 10% of its workforce, against $125-$145 billion in 2026 AI capex. More than 1,000 employees have signed a petition to halt the program; European staff are exempt under GDPR. Surveillance is now upstream of the layoff.

While today’s agents reshape labor, the next tier is moving closer to the door. Anthropic’s Claude Mythos preview briefly surfaced inside public Claude Code on May 25 before disappearing. Mythos has stronger code reasoning than Opus 4.7 and can exploit zero-days across operating systems and browsers. Inside Project Glasswing’s restricted tier, roughly 50 partners have already flagged 10,000+ high or critical vulnerabilities in its first month. Anthropic had previously said no public release was likely soon. The sighting suggests safeguards for wider distribution are getting close.