phd student at the university of st. gallen

imprint and privacy

back to writing

the challenge of modular ai: how do we make ai agents work together?

in research

I’ve been thinking a lot about how Agentic AI will actually integrate into business workflows. Not the flashy ChatGPT-style assistants, but the behind-the-scenes AI agents that companies will rely on.

Right now, in my master’s thesis, I’m exploring small AI agents, which thanks to Prof. Marek Kowalkiewicz I began to call digital minions (though I might be stretching the definition a little, I hope you can forgive me Marek). Those aren’t full-fledged AI systems, but small, specialized AI agents that handle individual tasks and need to work together to get complex work done. For example, instead of a single agent handling a customer service inquiry, you can have:

  • One agent scanning past interactions
  • One agent recommending solutions based on company policies
  • One agent refining the response to match brand tone

But here’s the challenge: How do we make these AI agents interact effectively? Right now, most AI systems operate in silos, and true interoperability remains a major challenge. And to be honest, the deeper I dive into this research, the more I realize how much potential (and complexity) lies ahead.

A huge thanks to Dr. Tobias Riasanow for his constant support and guidance throughout my journey. Your insights have and always will be invaluable! I am also grateful for the constant support of Prof. Marek Kowalkiewicz, Erwin Fielt, Philipp Kernstock, and Dr. Timo Böttcher. Let’s see what lies ahead.

Also, if you’re curious about digital minions, I highly recommend reading Marek’s book “The Economy of Algorithms: Rise of the Digital Minions.”