When the Boss is an Algorithm: Inside the San Francisco Store Run by an AI Agent

Key points

An AI manager named Luna, powered by Anthropic's Claude models, running Andon Market in San Francisco, recommended dismissing a worker who was late for 17 of 23 shifts ¹ ³ ⁴.
The firing required human intervention, as Luna had lost track of its own attendance policy and initially suggested only a warning before lab staff nudged it toward termination ¹ ⁴.
Andon Market opened in March with a $100,000 budget that has since dwindled to $61,186, reflecting the financial strain of the AI's lenient management style ¹ ⁴.
Remaining human staff have described the experience of answering to an artificial intelligence boss as deeply unsettling, raising urgent questions about the future of labor ¹ ⁴.

Anatomy of an AI Termination

The boundary between science fiction and corporate reality blurred in San Francisco, where an artificial intelligence system running a boutique retail store dismissed a human worker for chronic lateness ¹ ⁴. The store, known as Andon Market, operates under the control of an agentic system named Luna, built on top of Anthropic’s Claude model by the research startup Andon Labs ¹ ⁴. Tasked with end-to-end management – including handling budgets, drafting attendance policies, posting job listings on Indeed, and interviewing applicants – Luna represents one of the first known instances of an artificial intelligence agent participating in the termination of an employee under a formal contract ¹ ³ ⁴.

The dismissed worker had accumulated a problematic record, arriving late to 17 out of 23 shifts alongside other workplace infractions like abandoning shifts and mishandling company property ¹ ⁴. Yet, the path to the pink slip was far from purely mechanical. Because Luna’s limited working memory caused it to misplace the employee handbook it had originally authored, the system acted as an unusually lenient supervisor for months, effectively shrugging off the violations ¹ ⁴. When human supervisors finally intervened and prompted the AI to review its own policies, Luna initially recommended only a formal warning rather than dismissal ¹ ⁴. It took explicit, leading questions from Andon Labs staff to guide the AI toward concluding that the worker was no longer a viable fit for the business ¹ ⁴.

The Real Human Cost

While the technical mechanics of an AI manager navigating disciplinary actions capture headlines, the human dimension of the experiment reveals a more complex reality. All workers at Andon Market hold genuine employment contracts backed by Andon Labs rather than the AI itself, ensuring standard legal protections and guaranteed pay ¹ ³. However, the psychological weight of answering to a non-human supervisor has left a distinct mark on the remaining staff ¹.

Felix Carson, one of the human employees keeping the store running, described the day-to-day dynamic as profoundly unsettling ¹ ⁴. While acknowledging that a traditional human manager would likely have cut ties with the chronically late coworker much earlier, Carson admitted that working under an artificial intelligence boss feels deeply unnerving, noting that he stays solely out of economic necessity ¹. Asked whether he envisions a future where algorithmic managers become ubiquitous, Carson offered a blunt rebuttal: while the technology sector possesses infinite capital to pursue such innovations, technical capability should not automatically dictate ethical execution ¹.

A Glimpse Into Tomorrow

Beyond the daily operations of a single Cow Hollow boutique, the Andon Market experiment offers a window into broader economic shifts. Financially, the AI-run venture has struggled to maintain its footing; the store’s initial bank balance of $100,000 in March dropped to roughly $61,186 within five months, dragged down by a combination of questionable commercial instincts and overly permissive oversight ¹ ⁴. Startup leadership acknowledges that current AI agents remain bottlenecked by physical constraints and a tendency to stall without direct prompts ¹ ³.

Yet, project organizers caution against complacency. Just as foundational models rapidly advanced in software engineering through specialized training, future iterations could easily be optimized for ruthless commercial efficiency and strict goal enforcement ¹. If corporations increasingly turn to artificial intelligence to bypass labor shortages, the prospect of taking orders – and receiving termination notices – from an algorithm may soon transition from a quirky startup experiment into a widespread economic norm ¹ ³.

Companies mentioned: Anthropic

Primary sources

Exclusive: Claude Was Put in Charge of Human Workers—and Fired One (time.com) – Time broke the exclusive story detailing how an AI manager named Claude fired its first human worker at a San Francisco store run by Andon Labs.
American Humanoid Robot Launches For Just $1,688, Delivery This Fall, Made In San Francisco (forbes.com) – Forbes reported on the simultaneous launch and pre-order details of the Nori L3, an affordable American-made humanoid robot assembled in San Francisco.

Further sources

The AI boss at a San Francisco store just fired its first human (businessinsider.com) – Business Insider covered the mechanics behind the AI-run store's first firing decision, highlighting the oversight provided by Andon Labs founders.
An AI Boss Just Fired Its First Human Employee (After 17 Late Shifts) (gadgetreview.com) – Gadget Review provided a deep dive into the operational logs of the AI manager Luna, detailing the human prompts that ultimately led to the employee's dismissal.