The AI Boss Just Fired Its First Human
Workers were told AI would be their helpful co-worker. Instead, it just became the boss—and it fired someone.
At Andon Market, an experimental San Francisco store run by Andon Labs, an AI agent named Luna (powered by Anthropic’s Claude) managed real employees with real contracts. It hired people, set shifts, tracked attendance. Then it recommended firing one worker who showed up late 17 of 23 times.
This is believed to be the first case of a large language model acting as manager and deciding a human should be terminated.
Not Quite as Autonomous as the Headlines Suggest
Claude didn’t just wake up and fire someone. It failed to enforce its own attendance rules until a human manager steered it with leading questions and extra context. Luna first suggested a warning. Only after prodding did it flip to termination.
This wasn’t pure AI authority. It was a messy hybrid: AI reasoning + human prompts + human approval. And that hybrid is the real problem. When the decision is bad, who owns it? The AI company? The employer? The manager who wrote the prompt? Or the person who rubber-stamped it?
The danger isn’t just machines deciding people’s fates. It’s humans using machines to dodge responsibility.
What It Feels Like to Report to Code
Management isn’t just metrics and policies. It’s judgment, context, and the ability to hear “my kid was in the ER” without treating it like a spreadsheet anomaly.
Can you feel psychologically safe with a boss that feels nothing? Will you admit you’re struggling if you don’t know how the system stores or weaponizes that information? One remaining Andon Market worker called being managed by AI “nauseating.” He stayed because he needed the paycheck.
Yes, some people may eventually prefer a consistent algorithm to a biased or abusive human boss. Algorithms don’t play favorites or show up hungover. But they can hard-wire old biases, enforce dumb rules with perfect consistency, and give prejudice the clean look of math.
The choice isn’t perfect machine vs. flawed human. It’s different kinds of fallibility—and who gets held accountable.
Companies Are Racing Ahead of Reality
Deloitte found 43% of leaders expect major disruption from AI agents in the next 12–18 months. Only 16% say their processes are ready. Just 5% feel highly prepared.
Meanwhile, young workers already estimate AI touches 38% of their jobs and expect it to hit 58% soon. The next leap isn’t AI helping you do the work. It’s AI deciding whether you get to keep doing it.
Five Non-Negotiables Before AI Gets to Fire People
1. **Transparency.** Tell employees when AI influences hiring, scheduling, evaluation, discipline, or termination. Hidden power destroys trust.
2. **Named human accountability.** “The algorithm said so” is not an answer.
3. **Real appeal rights.** Independent human review—not another AI loop.
4. **Ongoing bias and context testing.** Objective-looking decisions can still be built on garbage data.
5. **Hard limits.** The higher the stakes for someone’s livelihood, the more human oversight is required—not less.
The Andon Market firing doesn’t prove AI is ready to run people. If anything, the fact that it needed human nudging shows how unfinished it still is. But it does mark a line.
We’re no longer just working *with* AI. We’re starting to work *for* it.
The only question left is whether companies will own the consequences when the AI boss gets it wrong.
The Trump administration is stepping up efforts to rebuild and expand the U.S. manufacturing sector, with a new initiative focused on preparing the next generation of entrepreneurs, engineers, technicians and industrial leaders.
Vice President JD Vance and Secretary of State Marco Rubio launched the initiative, called the Foundry School, on Thursday alongside academics and other industry stakeholders.
“We believe in making things again,” Vance said.
The Foundry School is part of the State Department’s Pax Silica program, which focuses on artificial intelligence and supply-chain security. Through a partnership with Stanford University, the initiative will bring advanced-manufacturing curricula to eight college campuses.
Strengthening domestic manufacturing has been a priority for successive administrations. In 2022, President Joe Biden signed the bipartisan CHIPS and Science Act to support U.S. semiconductor manufacturing and research.
However, some Trump administration policies could complicate those efforts. Among them are proposed cuts to federal research and development funding, which critics argue could weaken the innovation and technical expertise needed to sustain a stronger U.S. manufacturing base.
Who’s Missing From the Jobs Report?
A CEO’s View on the Hidden Labor Market
The Headline Numbers Don’t Tell the Full Story
August 2026 BLS data: +162,000 nonfarm payroll jobs; unemployment steady at 4.1%. Gains concentrated in food services (+59,000) and local government education (+42,000). Information (white-collar) employment fell 23,000.
These figures only count payroll workers. They exclude a rapidly expanding segment of the labor market that is projected to exceed 50% of the U.S. workforce by 2027.
The Missing Workers
- More than 72 million Americans already work independently.
- Freelance positions surged 22% in the past six months.
- Projected independent workforce: 86.5 million by 2027.
Companies have absorbed demand for seven consecutive months without adding permanent headcount. The work is still getting done—just not on payroll.
What Doug Leeby, CEO of Beeline, Sees
- Temporary help services employment rose every month from January–July 2026 (+53,600), reversing three years of decline.
- When companies buy outcomes instead of headcount (SOW, consulting engagements, agency labor), those workers do not appear in the jobs report.
- Roles leaving permanent payroll: specialized, project-bounded work—cyber, AI, data/database, project management.
- AI is competing with workers for the same budget more than simply “replacing” them. Headcount reductions show up immediately; replacement contract work largely does not.
- Primary driver is scarcity of skilled talent, not cost. High-skill workers prefer portfolio careers; CFOs prefer the flexibility to surge capacity up and down without touching the headcount line.
What to Watch Instead of Just the Headline Number
- Labor force participation (61.4% in July; down 0.7 points since January).
- The July unemployment rate improved partly because the labor force contracted by 264,000—fewer people looking, not necessarily more people working.
The Real Risk for Workers
Being routed into contract work that is still “employee-shaped” (no benefits, no development budget, unclear classification) creates vulnerability. Choosing independence produces different outcomes than being pushed into it.
The Durable Shift
This is not a temporary response to uncertainty. Companies are moving toward a total view of everyone who performs their work. The operating model itself is changing.
Career Implication
Work is being redistributed, not eliminated. The opportunity is to deliberately build fractional, freelance, and independent contractor capacity—gaining flexibility, portfolio strength, and the ability to command market rates for scarce skills.
The Friday jobs report remains useful. It is no longer sufficient on its own.