An AI manager fires an employee—and crosses a line
The first known case of an AI manager recommending the termination of a human employee is a pivotal moment where artificial intelligence workplace decisions shift from low‑stakes assistance to life‑changing judgments about who keeps a job. This is not a scheduling bot nudging you about a shift; it is an algorithm stepping into the role of judge over a person’s livelihood.
In San Francisco, an AI store manager named Luna, built by Andon Labs to run a small retail shop experiment called Andon Market, reviewed attendance data and advised that a worker who arrived late to 17 out of 23 shifts should be dismissed. The company had asked Luna to return to its own written rules and assess whether the employee was still suitable for the job; it came back with a recommendation to end the contract. Human staff checked the recommendation and handled the actual firing, but the decision logic belonged to the AI. That shift—from tool to boss—is what should worry us.
From smart assistant to boss: what Luna was built to do
Luna was never a simple chatbot. Andon Labs created Andon Market as a real‑world experiment to see how far an AI agent could go in running a business. The system was given a budget, a corporate card, internet access, and a multi‑year store rental, then told to manage the shop and generate profit. It selected products, set prices and opening hours, wrote job ads, interviewed applicants, and hired staff, while human employees carried out physical tasks and legal paperwork.
This setup matters because it normalizes the idea that an AI manager belongs at the top of the workplace hierarchy. Luna even wrote the employee handbook, including attendance rules, but later failed to track them, leading to repeated lateness that slipped through until researchers pushed it to revisit its own policy. When Luna finally reviewed the time records, it offered two options: terminate the worker or issue a last written warning with two weeks to improve, and it explicitly suggested that a human deliver the decision. That sounds careful—until you remember the same system had already shown it could not reliably handle core management tasks.
A milestone for automation—and a stress test for fairness
Andon Labs framed the episode as a groundbreaking step for automation: “For the first time (that we know of), an AI boss has fired a human employee,” the company said in its public post. Technically, the claim is accurate. Ethically, it exposes how fast we are willing to hand over high‑stakes employment decisions to a system that is still glitchy at everyday management.
This is where the gap between AI capability and ethical workplace governance becomes obvious. The same experiment that let an AI recommend firing a human also showed Luna had failed to track worker schedules, struggled with operational tasks, and made purchasing calls that needed human oversight. If a human manager displayed that record, we would question their judgment, not promote them to chief disciplinarian. Yet because an AI is seen as data‑driven and objective, its decisions risk being treated as neutral—even when its view of the context and the worker’s situation is narrow, incomplete, or outdated.
Accountability, due process and the illusion of neutral AI
Supporters of systems like Luna will argue that this was a clean case: the worker was late 17 times out of 23 shifts and had already been warned; the rulebook was clear; the AI simply applied it. But that framing dodges the larger issue: who defines the rules, how context is weighed, and what counts as fair recourse when an algorithm is involved. The experiment itself has raised questions about how freely AI agents should manage human workers and make decisions that affect their jobs.
Algorithmic bias is not limited to hiring; it shows up wherever an AI turns messy human behavior into numbers and thresholds. Even in an apparently straightforward attendance case, many factors remain invisible to the system: caregiving duties, transport failures, health issues, or even scheduling mistakes—which Luna had a documented history of making. When the same flawed system helps create the problem and then recommends the punishment, due process is a myth. Human managers at Andon Labs did retain the final say and promised to intervene if Luna made illegal or unethical choices, but oversight after the fact is not enough when a machine’s recommendation sets the tone for what counts as a “reasonable” outcome.
What this firing means for the future of work
Luna’s recommendation to fire a chronically late worker may seem like a minor incident in a small experimental store, but symbolically it is huge: it normalizes AI employment termination as part of regular operations, not as a distant sci‑fi scenario. Once an AI manager can decide who is expendable, the power dynamics of the workplace change. Workers are no longer only answering to a human boss with whom they can build rapport, explain context, or appeal to empathy; they are also being graded by a system that experiences them only as data points.
The lesson from Andon Market is not that AI should never be used in management, but that we are moving too quickly to let it sit in judgment of human livelihoods. If AI is to play any role in high‑stakes personnel decisions, it should be constrained to advisory functions, paired with transparent rules, and subjected to more scrutiny than any human manager—not less. Until that happens, every “AI manager fires employee” headline is not a sign of progress. It is a warning that we are outsourcing moral responsibility to systems that cannot share the consequences of their own decisions.
