The AI agents Meta planned to replace up to 60% of teams with caused «large-scale failures»

Ars Technica revealed details of Meta's scrapped plans: the company was preparing to go «AI-native», cutting some teams by up to 60% and handing their functions to AI agents. The plan was rolled back after the agents that inherited people's work failed at such scale that reversing everything proved cheaper.

This story is the best possible answer to the «AI will replace everyone» fear: even a company with a bottomless budget and top engineers could not replace whole teams with agents. The reason is fundamental: a department's work is not a task list but a web of implicit agreements, exceptions and common sense that nobody ever wrote down. An agent receives the formal part and breaks on the informal one.

Yet the same industry experience shows where agents run stably: one position — one agent with a clear zone of responsibility, its role's data and a human overseeing exceptions. At 1Cifer we deliberately build to this scheme — the director's agent, the accountant's agent and the manager's agent each run their own scope rather than replacing a department wholesale. For Kazakhstani business the conclusion is both comforting and sober: augmenting people with agents works; substituting teams does not.