We study optimal incentive contracts when workers privately observe whether Agentic AI can automate their jobs. Firms balance bonuses for truthful reports of successful automation with termination threats. Workers may be fired regardless of automation success (\textit{mass termination}), even though dismissing non-automatable workers destroys value. Mass termination becomes more likely when automation probability rises or workers capture more surplus. Firm value is convex in automation probability, while worker utility is non-monotonic, rationalizing divergent attitudes toward Agentic AI. Mass layoffs increase in frequency over time without improvements in automation, and are less likely in firms employing many similar workers.