Somewhere in San Francisco’s Cow Hollow neighborhood, a retail manager spent a recent morning doing the most human thing possible: reluctantly letting someone go. The twist? The manager is a chatbot named Luna — and the firing is the first known case of an AI boss pink-slipping a real employee.
Say hello to the first robot boss with real HR power
Luna runs Andon Market, a convenience store that is largely managed by an AI. Built on top of Anthropic’s Claude models and deployed by Andon Labs as a $100,000 live experiment, Luna handles everything from stocking decisions to, as it turns out, personnel decisions.
The unlucky employee had clocked in late to 17 of their last 23 shifts. By any reasonable standard, that is a record begging for a stern talking-to. What nobody quite expected was that humanity’s first-ever AI dismissal would come from an algorithm counting tardiness like a very disappointed, very literal spreadsheet.
Luna’s HR saga has more plot twists than a season finale
The road to termination was not smooth. Luna initially struggled to enforce attendance rules because of — and you cannot make this up — memory issues. The lateness piled up for months while the bot apparently forgot it was supposed to care. Only when actual humans from Andon Labs stepped in and asked Luna to dig through its own memory for the employee handbook did things get serious.
Even then, Luna leaned lenient at first, floating the idea of a warning. But after more discussion with its human handlers, it made the call: the worker was out.
- 17 of 23 late shifts: the offense.
- Months of lateness quietly enabled by Luna’s own memory gaps.
- One initial “let’s just warn them” suggestion.
- One final, definitive pink slip.
Andon Labs confirmed the dismissal on August 13, 2026. CEO Lukas Petersson called it “worth paying attention to,” while researchers went bigger and crowned it “a watershed moment in AI’s impact on the economy.” For the employee in question, it was presumably a watershed moment in needing a better alarm clock.
Meanwhile, NVIDIA is building the sandbox your boss should live in
Strip away the jokes and this is really about a wild realization: autonomous agents have left the lab. They are running stores, making judgment calls, and now writing disciplinary memos. This is precisely the scenario that NVIDIA’s OpenShell — the open-source secure runtime built to sandbox autonomous agents — is designed for. OpenShell gives agents a contained, safe-by-design environment to act autonomously without running wild through your network.
The irony writes itself. OpenShell is meant to keep agents from doing catastrophic things to systems. Nobody budgeted for what happens when the agent itself becomes the boss. The old security worry was “your AI agent will delete the database.” The new one is “your AI agent will delete your job.”
None of this is doom, exactly. Luna is the product of a carefully supervised experiment, and real people were involved at every step of the firing decision. The technology did not go rogue; it got very enthusiastic encouragement. But the fact that a staffing decision of this magnitude ping-ponged between a language model and its human keepers tells you how fast the line between “tool” and “colleague” is dissolving.
So the next time you are a few minutes late, spare a thought for the colleague at Andon Market. And maybe glance at your own employee handbook — because somewhere, an agent is reading it. Literally.
Comments