From Prompts to Production: The Systems Agenda for Agentic AI
In this talk, I will discuss our recent work on benchmarking, evaluation, and deployment of multi-agent LLM systems, and use it to outline a broader research agenda for agentic AI as a systems discipline - where progress depends not only on better models, but on principled infrastructure for observability, reproducibility, safe experimentation, and scalable execution.
Overview
Agentic AI is rapidly evolving from prompt-based assistants into autonomous systems that plan, call tools, coordinate multiple agents, and interact with live environments. This shift turns “AI behavior” into a systems problem: executions become stochastic, architectures heterogeneous, and failures difficult to reproduce, diagnose, and compare. I will conclude with a forward-looking vision of what an AI-native software and network stack might look like if agents become a dominant interface to computation and information.
Presenters
Brief Biography
Marco Canini is a Professor of Computer Science at KAUST. He obtained his Ph.D. in computer science and engineering in 2009 from the University of Genoa, Italy, after spending the last year of his degree as a visiting student at the University of Cambridge, U.K.
He holds a Laurea Degree with Honors in Computer Science and Engineering from the University of Genoa. He was a postdoctoral researcher at the École polytechnique fédérale de Lausanne (EPFL), Switzerland, from 2009 to 2012. He then worked as a senior research scientist at Deutsche Telekom's Innovation Labs and the Technical University of Berlin, Germany, for one year.
Before joining KAUST, Canini was an assistant professor of computer science at the Université catholique de Louvain, Belgium. He has also held industry positions with Intel, Microsoft, and Google.
