Amazon warehouse managers bypass AI staffing tools during pilot tests
As the world’s largest company by revenue scales automation, leaked communications show managers disabling systems they deemed unable to account for worker capabilities and real-world nuances.
Amazon warehouse managers intervened to disable or bypass new AI staffing enforcement systems during early pilot tests, citing the software's inability to account for individual worker capabilities and real-world nuances. Internal documents and communications reveal that human managers regularly ignored the software's suggestions, found loopholes around its requests, or shut down certain components altogether when they believed they knew better.
Minutes after the new AI staffing enforcement was switched on at one facility, a manager reportedly begged an engineer to shut it down. Leaked Slack messages from the pilot phase show staff requesting the system be turned off due to operational hindrances, with one employee comparing the physical differences between workers to highlight the software's failure to consider individual strengths.
Internal memos stated that offering best practices for managers was "insufficient" and that manual overrides were eroding the system's effectiveness. One memo noted that without system-enforced guardrails, manual overrides and habits erode even the best science. As a result, a primary objective identified for the current year was implementing "hard" enforcement to prevent these manual overrides.
An Amazon spokesperson told Moneywise that leaked files depict an "incomplete picture" and that the anecdotal observations came from a "handful of managers" during the initial adoption period. The spokesperson added that these observations do not reflect how the system operates today and were used to inform adoption going forward.
Amazon maintains that the initial testing phase is incomplete and that the technology aims to improve staffing consistency based on real-time needs. The company states the technology is designed to augment human judgment rather than replace it, aiming to free up manager time for supporting teams. Future plans for AI automation will be contingent upon each facility's unique needs, with adjustments made before scaling.
