What happens when humans and AI decide together?

Since April, an AI agent called Luna has run a small shop in San Francisco, with a real budget and real staff. In August, as a recent TIME report set out, it decided to let one of them go.

Early on, Luna wrote a staff handbook with a clear attendance policy. Repeated late arrivals would lead to a formal warning, and eventually to dismissal.

Over the following months, the handbook dropped out of Luna’s limited working memory. One employee was late for most of their shifts, and Luna excused it every time.

The researchers eventually stepped in. They asked Luna to search its memory for its own rules and consider whether the employee was still the right fit. With that prompt, Luna made its decision.

Andon Labs’ own write-up sums up the lesson: AI bosses are slow to fire and quick to hire. 

Ben Terrett CBE, the designer behind GOV.UK, touched on this in his episode of TOMORROW. Government holds to a long-standing principle, he explained: a human makes the ultimate decision, because the consequences for people’s lives are real.

He expects the private sector to test that line. Even so, he sees AI’s role as speeding up the work around a decision.

Luna’s shop sits in the private sector, where no such principle applies. Andon Labs kept a human in the loop anyway.

So what decisions should we trust AI to make at work? Luna’s story offers an early answer. A person spotted what the AI had missed, asked the right question and stayed accountable for the outcome. The AI brought consistency. The human brought context and judgement.

For any organisation working out how far to let AI into decisions about people, that is a useful place to start.