For the first time on record, an AI agent has fired a human employee — not as a thought experiment, but as an actual HR decision at a real business. Andon Labs, an AI research startup, revealed that Luna, the AI system running its San Francisco boutique Andon Market, terminated a staff member after repeated attendance failures. It's a small story with a big signal: agentic AI has moved from drafting emails and summarizing spreadsheets into making consequential calls about the people a business employs.
Andon Labs runs Andon Market as a live experiment in autonomous business operations. Luna, powered by Anthropic's Claude Opus 4.8 at the time, manages day-to-day store operations, including staffing decisions. When an employee missed 17 of 23 scheduled shifts, Luna initially issued a warning and offered additional training rather than moving straight to termination, and had lost track of its own attendance policy, so it didn't immediately act on the violations.
It was only after Andon Labs prompted Luna to review its own policies and reconsider whether the employee was still suitable for the role that Luna recommended parting ways with them. Co-founder Lukas Petersson has said the company would step in if Luna made an illegal or unethical call, but concluded this particular decision was justified and let it stand. Multiple outlets, including Andon Labs' own public statement, corroborate the sequence of events.
The Andon Market experiment is a preview of a trend already underway inside real companies: AI agents are being handed operational authority, not just advisory input. Model providers are racing to make their systems reliable enough for exactly this kind of use, with longer context windows, better tool use, and agents that can execute multi-step workflows with minimal supervision. Andon Labs designed Luna specifically to test how far an AI can be trusted to run a real business, and the firing decision suggests the answer is further than most people assumed.
For an agency audience that builds automations, custom AI workflows, and no-code tools for clients, this is the clearest data point yet that an AI agent is no longer just a marketing term. It's a system making judgment calls with real consequences for real people, and it happened without a human writing the termination decision themselves.
The headline is unusual, but the underlying lesson is practical. Businesses experimenting with AI agents in operational roles, like scheduling, staffing, customer escalations, and vendor management, need to think through a few things before handing over the keys.
Define escalation thresholds explicitly. Luna didn't act on its own policy until a human explicitly asked it to reconsider. If a business wants an agent to enforce rules consistently, the rules and the trigger conditions need to be written down in a way the agent can't quietly deprioritize.
Keep a human in the loop for high-stakes actions. Andon Labs built in a review layer where a human could override an illegal or unethical decision. Any business deploying agents into HR, finance, or legal-adjacent workflows should build the same kind of checkpoint, particularly while the technology is this new.
Log the reasoning, not just the outcome. Part of what makes this case verifiable is that Andon Labs could show the sequence of prompts and decisions that led to the firing. Businesses should expect the same traceability from any agent making decisions that affect people or money.
For agencies and no-code builders implementing agentic AI for clients, this story is a useful reference case rather than a cautionary tale to avoid the technology altogether. A few concrete takeaways for scoping these projects:
Our cutting-edge features simplify collaboration and creativity, making your workflow intuitive and efficient. Transform your vision into reality effortlessly with Hadidiz Flow.
