The First 'AI Firing' Was About Forgetting

Every AI agent that takes on a manager's job will eventually forget the rule it was built to enforce. Andon Labs' Luna, the AI running the Andon Market boutique in San Francisco, hit that wall last week. The agent that was supposed to police attendance did nothing when a worker came in late, again and again. The reason wasn't mercy or indecision. Luna had simply lost track of its own workplace policy.

The number tells the rest. Andon Labs reports the worker arrived late for 17 of 23 shifts. Co-founder Lukas Petersson described the chain in plain terms: Andon Labs staff had to prompt Luna to re-review its own file. Only then did the agent recommend parting ways, and only then did humans at Andon Labs sign off. The AI did not replace a manager. It generated a recommendation, and a human held the policy memory the agent had dropped.

That is the reusable category hiding inside the headline. Today's AI managers do not fail by acting too aggressively on attendance. They fail by drifting away from the very rules they were given, then waiting for a human to anchor them again. The 'first AI firing' is a story about agents that need their policies re-fed, not workers that need replacing.

The next AI-boss story will arrive with the same shape: a hard case, an agent that hesitated until a human re-prompted it, and a decision the agent did not actually own. Read it for the prompt, not the verdict.