AI Manager Fires Human Employee for First Time at SF Store

An AI agent operating as the manager of an experimental San Francisco retail store has made its first decision to terminate a human employee, marking a notable milestone in autonomous AI workplace management.

Luna, the AI manager running Andon Market, recommended dismissing an employee who arrived late for 17 of 23 scheduled shifts, according to Andon Labs, the company behind the experiment. The lab announced the decision Thursday, providing conversation logs and screenshots documenting the months-long process.

How the firing unfolded

The termination did not happen automatically. Luna had created an attendance policy early in the store's operation but subsequently lost track of it in her memory systems, allowing the chronic lateness to continue for months without consequence. Andon Labs eventually prompted Luna to search her stored policies and evaluate whether the employee remained suitable for the position.

After reviewing the situation, Luna recommended "parting ways" with the worker. Humans at Andon Labs reviewed the recommendation and executed the termination.

Lukas Petersson, cofounder of Andon Labs, told Business Insider that the lab maintains oversight to prevent illegal or unethical decisions, though he said intervention was unnecessary in this case given the clear policy violation. Conversation logs show Luna provided progressive warnings and additional training over several months before reaching her recommendation.

The experimental store setup

Andon Market opened April 1 as a test of AI agent capabilities. Andon Labs gave Luna a $100,000 budget, internet access, and a corporate credit card with instructions to open a store and generate profit. Built using Anthropic's Claude models, Luna selected merchandise, posted job listings on Indeed, conducted interviews, and hired employees to staff the store, which sells books, candles, prints, games, and branded items.

All workers hired by Luna are formally employed by Andon Labs with guaranteed pay and legal protections. The store has generated sales but has not yet reached profitability.

Why it matters

The incident reveals both the potential and limitations of AI workplace management. While Luna eventually reached an appropriate decision supported by documented policy violations, the agent required human prompting to act—a pattern Petersson identified as a persistent weakness in current AI systems. A human manager would likely have addressed the performance issue much sooner. As companies explore AI-assisted management, this experiment demonstrates that autonomous decision-making remains dependent on human oversight, particularly for consequential employment actions. The case also raises questions about accountability and the appropriate balance between AI efficiency and human judgment in workplace relationships.

Petersson believes companies will eventually be run entirely by AI, with artificial intelligence serving as employers of human workers. The Andon Market experiment provides an early test case for that future, complete with its complications and limitations.

These details were first reported by Business Insider.