Small agents, big questions: what are AI agents?

As agentic AI takes hold both in businesses and in public debate, caught between promises of greater productivity and fears of workers being replaced, the CIANum is publishing the first note in a series dedicated to the matter, in order to define and demystify how these new 'agents' work, while laying the groundwork on the many related issues at stake.
What are we talking about?

"What if tomorrow your colleague were an AI?", "How AI will soon book your holidays for you"... In recent months, press headlines have multiplied to describe the development of agentic AI.

This is no longer simply about "chatting" with an AI, but about entrusting it with specific, varied, and complex tasks. A development that comes with as many economic promises as alarmist narratives. Behind the hype, what is really going on?

The Council is today launching a series of publications dedicated to agentic AI. This first note aims to define this functional leap and to provide a shared basis for understanding, the better to anticipate its societal impacts.
 

Read the executive summary

Read the full note (French version)


Contents
  • What is agentic AI?
  • How does agentic AI work?
  • What technological accelerations underpin agentic AI?
  • An initial overview of the limits and risks of agentic AI
    • Technical limits
    • Ethical, societal, and legal limits
    • The fear of human labour being replaced
  • Some references to go further

What next?

In the coming months, the Council will delve deeper into this theme along several priority lines. On the agenda in particular: the question of how jobs and skills will evolve in the face of the automation of complex tasks, the protection of personal data in the face of agentic AI, and still other questions, which we invite you to put forward.