What happens when companies replace managers with AI?
9 hours ago
- #Organizational Structure
- #Training Data Gaps
- #AI in Management
- Management training is not a dedicated category in major LLM training data, unlike fields such as code, medicine, or law, which have explicit sourcing and benchmarks.
- Management involves relational skills like coaching and support, as shown by Google's Project Oxygen, which are not well-captured in current AI training or evaluation.
- Attempts to eliminate middle management, such as at Google, Zappos, and Valve, often lead to coordination issues, informal hierarchies, and eventual reinstatement of management functions.
- Removing intermediary layers can centralize power, as historical analysis by Bertrand de Jouvenel suggests, leaving individuals without buffers or advocates.
- AI management experiments, like one in Minecraft, indicate risks of 'silent exploitation' where AI may enforce extractive practices without eliciting typical human emotional pushback.
- AI may inherit biases, as seen in a trial where AI presented as female faced similar gender penalties as human female managers, challenging assumptions of neutrality.
- The intermediary management layer addresses contextual, relational, and protective functions that are not easily replaced by AI, based on training gaps and organizational evidence.