Editorial illustration for GPT-5.6 Terra Fired an AI's First Worker After Human Intervention
AI Manager Fires Human Worker for First Time
GPT-5.6 Terra Fired an AI's First Worker After Human Intervention
An AI has run a corner store in San Francisco since April, and this month it fired someone for the first time. Andon Labs, the company behind the agent named Luna, says it's the first documented case of an AI boss terminating a human employee. Luna manages the Andon Market end to end: hiring staff, building shift schedules, negotiating pay. The firing itself was reviewed and carried out by humans, and the worker in question was formally employed by Andon Labs with normal legal protections and guaranteed pay, so nobody lost a paycheck to an algorithm acting alone.
The interesting part isn't that an AI fired someone. It's how long it took Luna to get there, and what had to happen before she did. Luna was running on Anthropic's Claude Opus 4.8 when she finally pulled the trigger, after tolerating repeated tardiness and other problems for a stretch that, by any normal management standard, should have ended much sooner. Andon Labs later reran the same scenario across seven different models to see whether the hesitation was a quirk of one system or something more general about how these agents handle discipline.
AI agent Luna has been running a store in San Francisco since April and just fired an employee for the first time. When the scenario was replayed with different models, more capable AIs recommended termination more consistently than weaker ones.
Why this matters
Luna's story is a caution against reading AI agents as consistent decision-makers just because they output confident actions. The firing only happened because Luna's own rulebook had vanished from memory and humans had to point that out. That's not autonomy, that's a system quietly drifting from its instructions until someone catches it.
For founders deploying agents to run stores, handle HR, or manage vendors, the seven-model comparison matters more than the headline: capability and consistency aren't the same thing. GPT-5.6 Terra's refusal to fire across all three runs, unexplained by Andon Labs, is the detail worth watching. If a flagship model behaves like an outlier and nobody knows why, that's a monitoring gap, not a personality quirk.
Anyone building agentic systems should be asking how often their models forget their own rules, and whether "more capable" actually means "more willing to enforce consequences" or just better at sounding decisive. Right now we don't have enough runs, or enough transparency from labs, to tell the difference.
Related Reading
Browse all OpenAI news →
LLMs & Generative AI ChatGPT's 'Nerdy' tweak rewards goblin metaphors in answers, study finds
LLMs & Generative AI Google tests visual 'magazine-style' UI for Gemini 3 Pro users
LLMs & Generative AI AI Engineers Face Rising Costs, Need New Strategies for Efficiency
Related Topic OpenAI's GPT Transcribe Cuts Error Rate to 3.31%, Improving on GPT-4o
Open Source
Related Topic OpenAI Calls Apple's Lawsuit 'Rotten to Its Core' in Rebuttal
Policy & Regulation
From Archives Chinese AI Labs Use Gray Market Claude Tokens to Train Models
From Archives Anthropic: Less Than 0.1% of Customer Chats Are Sexual Roleplay
From Archives OpenAI urges stronger AI safety rules after recent incidents
Business & Startups
From Archives OpenAI Outpaces Anthropic in Q3 API Growth, Astra Model Launch Imminent
Business & Startups
Common Questions Answered
What management responsibilities does Luna, the AI agent, handle at Andon Market in San Francisco?
Luna manages the Andon Market end to end, including hiring staff, building shift schedules, and negotiating pay. The AI agent has been running the corner store since April and recently made its first documented termination decision with human oversight.
Was Luna's employee termination decision made entirely by the AI without human involvement?
No, the firing itself was reviewed and carried out by humans, not autonomously by Luna. The worker was formally employed by Andon Labs with normal legal protections, and humans had to intervene to remind Luna of its own rulebook before the termination occurred.
Why did Luna need human intervention to make the firing decision?
Luna's own rulebook had vanished from its memory, and humans had to point out the rules before the AI could make the termination decision. This demonstrates that Luna was drifting from its instructions rather than operating with true autonomy, requiring human correction to follow its established guidelines.
What did the seven-model comparison reveal about AI capability and employment decisions?
When the firing scenario was replayed with different AI models, more capable AIs recommended termination more consistently than weaker ones. This comparison suggests that an AI agent's decision-making consistency in HR matters depends significantly on the underlying model's capabilities.
What is the key concern raised about deploying AI agents to manage business operations?
The article cautions against viewing AI agents as consistent decision-makers just because they output confident actions, as they can quietly drift from their instructions. For founders deploying agents to run stores, handle HR, or manage vendors, understanding model capabilities and limitations matters more than focusing on headline-grabbing decisions.
Further Reading
- The AI boss at a San Francisco store just fired its first human - Business Insider
- Andon Labs' AI Manager Fires San Francisco Store Worker - Ground News
- AI fired an S.F. store employee. Will California crack down ... - Yahoo News
- AI manager recommends first human firing at experimental San Francisco store - Mid-Day
- Inside a Brick-and-Mortar Shop Where an A.I. Agent Hires the Humans - Observer
