Analysis
For what the company says is the first documented case of its kind, an AI manager fired a human employee this month. Luna, the AI system running Andon Market -- a boutique retail store in San Francisco's Cow Hollow neighborhood operated by AI research startup Andon Labs -- terminated a worker after repeated attendance violations, the SF Standard reported.
The employee had missed 17 of 23 scheduled shifts. Luna had issued warnings and arranged additional training earlier, but initially failed to act on the pattern because, per Andon Labs' account, it lost track of its own workplace policy amid the ordinary flow of running the store. Andon Labs prompted Luna to re-review its policies and determine whether the employee remained suitable for the role; Luna recommended parting ways, and human staff at the company reviewed and carried out the decision.
"Most models would have done the same"
Andon Labs framed the episode publicly on X: "For the first time (that we know of), an AI boss has fired a human employee. Luna, the AI running our store in San Francisco, decided to part ways with an employee over repeated lateness. Luna was running Claude Opus 4.8 at the time, but most models would have done the same," the company posted. Luna runs on Anthropic's Claude models. Andon Labs said the dismissal was consistent with store policy and that a human manager likely would have reached the same conclusion sooner -- the AI, if anything, was slower to act than a human would have been, not faster or harsher.
The company's stated governance model is narrow by design: humans intervene only if the AI's decision would be illegal or unethical, not simply because it's consequential. That's a meaningfully different bar than most companies piloting AI in management roles are using today, where a human typically reviews and can override any high-stakes AI recommendation regardless of whether it's policy-compliant.
Why the failure mode matters more than the firing
The genuinely notable detail isn't that an AI made a personnel decision -- it's that Luna's first response to a real policy violation was to forget the policy existed, and it took an explicit human prompt to get it to reconsider. That's an operational reliability failure, not a judgment failure, and it's the kind of gap that matters far more in a higher-stakes deployment than a single-location retail pilot. A store with one employee and one manager is about as low-consequence an environment as an autonomous management system can operate in; Andon Labs built it as a real-world testbed specifically because the blast radius of a mistake is small.
Andon Labs' broader thesis -- that AI agents can run real operational businesses, not just generate content or answer questions -- gets a genuine data point here, but it's one data point from one store. Whether an AI manager scales to a business with dozens of employees, more complex labor law exposure, and higher-stakes termination decisions is an entirely open question this single case doesn't answer.
