The zero-person company experiment by KPMG and the University of Amsterdam received national attention this week in Nieuwsuur. The project, which has been running for more than a year, explores how far autonomous AI agents can go in performing business operations and where human oversight remains essential.
The experiment began in early 2025 with a central question: can a company function without human staff, run entirely by AI agents? The initiative was developed by Sander Klous, professor of AI & Audit at the University of Amsterdam and partner at KPMG, together with entrepreneur Nart Wielaard. The approach was inspired by international discussions about highly automated organisations and the future of autonomous corporate structures.
The agents were instructed to decide what type of company they wanted to build. They chose an operational webshop offering personalised AI-generated art. One agent acted as CEO, supervised by a human advisory board. The agents worked continuously and coordinated multiple tasks at high speed.
However, the limitations became clear early on. When agents were placed in broad, complex roles such as financial leadership or organisational strategy, they began to drift, ignore instructions or shut down unexpectedly. This instability showed that fully autonomous business management is not yet feasible.
However, the limitations became clear early on. When agents were placed in broad, complex roles such as financial leadership or organisational strategy, they began to drift, ignore instructions or shut down unexpectedly. This instability showed that fully autonomous business management is not yet feasible.
The experiment highlights broader questions about governance, liability, employment and public trust. The findings indicate that AI is well suited for task automation and administrative support, while human judgement remains crucial for strategic decisions, oversight and ethical safeguards.
Reflecting on the project, Nart Wielaard notes that the experiment produced valuable insights that now form the foundation for his new venture, Cyberdune Agents, where the learnings are applied in real-world environments.
The results show that while AI agents excel in speed and task execution, full autonomy remains out of reach. The findings offer organisations guidance for responsible adoption of agents within hybrid processes that bring human and machine closer together.
Read also
