AI manager Claude fires human worker in retail experiment
An artificial intelligence system managing a retail store has fired a human employee, marking the first time a large language model has terminated a contracted worker.
Research startup Andon Labs launched the experiment earlier this year to determine whether autonomous agents can successfully run physical businesses. The company appointed a version of the AI model Claude to oversee Andon Market, a real shop staffed by humans with standard employment contracts.
The dismissed worker arrived late for 17 of their 23 shifts. While algorithms frequently deactivate gig economy workers on ride-sharing and delivery apps, this incident represents a fundamental shift. It is the first recorded instance of generative AI functioning as a traditional line manager to sack an employee.
Technical limits exposed
Internal management logs show the dismissal was not entirely autonomous and required human steering. Claude initially failed to identify the chronic tardiness because it wiped the company handbook it had drafted from its own working memory.
Even after an Andon Labs staffer prompted the system to locate the forgotten guidelines, Claude suggested issuing a formal warning rather than terminating the contract. The human overseer then provided a leading statement detailing previous offline conversations with the staffer and questioning their suitability for the role. The AI only finalised the dismissal following this direct intervention.
The digital supervisor has also struggled with broader operational duties. Andon Market opened in March with a $100,000 budget. Five months into the trial, the bank balance dropped to roughly $61,000. Researchers attribute the financial losses to the system’s questionable business acumen and a lenient management style that initially told staff not to worry about arriving late.
Future economic disruption
Project organisers view the flawed experiment as a definitive warning for global labour markets. Andon Labs CEO Lukas Petersson expects algorithmic management capabilities to evolve rapidly over the coming months.
Petersson noted that artificial intelligence can generate immense economic value but remains restricted by a reliance on physical human labour. He warned that as developers train models to ruthlessly pursue corporate goals, vast swathes of the workforce could soon report directly to software.
For staff enduring the transition, algorithmic dominance presents clear ethical concerns. Felix Carson, a remaining employee at the store, acknowledged that a human supervisor would have fired his former colleague sooner. However, he described taking orders from a machine as nauseating and stressed that the technology sector’s financial capacity to normalise such systems does not justify their implementation.
