AIBRIEF
Luna, the AI manager of experimental San Francisco store Andon Market, only fired its first worker after a human operator asked it to retrieve the policy it had written.
Luna, the AI manager of Andon Market, drafted an attendance policy for the experimental San Francisco store, then lost track of it for months. It could only fire its first employee after a human operator asked it to search its own memory and reassess the worker.
The fired employee had arrived late for 17 of 23 shifts, according to a report from Andon Labs, the AI research lab behind the store. Conversation logs published by the lab show Luna gave the employee repeated warnings and additional training before recommending "parting ways." Andon Labs reviewed the call and carried out the firing.
Andon Market opened on April 1 with Luna, built on Anthropic's Claude models, running daily operations end to end. The lab gave the agent a $100,000 budget, internet access, and a corporate credit card, with instructions to open a store and turn a profit. The store sells books, candles, prints, games, and branded merchandise.
Under the experiment's design, all workers hired by Luna are formally employed by Andon Labs, receive guaranteed pay, and have legal protections, according to Business Insider. Cofounder Lukas Petersson said the lab would intervene if Luna made an illegal or unethical decision, but did not believe that was necessary in this case. Time magazine separately corroborated the firing.
The published log makes one detail specific: the lab's prompt, not the AI's initiative, surfaced the policy Luna had written months earlier. The fired worker's identity has not been made public.