What is it about?
"Agentic AI" has become one of the most-used phrases in technology — systems that plan, use tools, and act on their own without a person directing each step. This paper asks a simple question hiding underneath the hype: does acting independently actually make something an agent? It argues that it does not. Autonomy — operating without external control — and agency — being the genuine source of one's own actions — are different things, and the paper identifies what separates them: continuity. An identity that persists over time, so that actions belong to a single ongoing self with memory, commitments and stakes, rather than being disconnected one-off episodes. Today's systems have the autonomy but not the continuity, which makes them powerful autonomous tools rather than agents in any real sense.
Featured Image
Photo by julien Tromeur on Unsplash
Why is it important?
The word "agent" is now doing heavy lifting in law, policy, and AI-safety debates — and a lot rides on whether it's the right word. If we treat autonomous systems as though they were agents, we misassign responsibility, build unrealistic expectations of what the technology can be trusted to do, and reach for the wrong governance tools. This paper gives the debate a clean, usable test for telling the two apart, so that questions of accountability — who actually answers for what a system does — rest on what these systems genuinely are, not on a marketing label. As "agentic" systems spread into real decisions, getting that distinction right becomes practically, not just philosophically, urgent.
Perspectives
I came to this question from an unusual direction. I spent close to two decades in electronics and semiconductor engineering, where you learn quickly that a system's real behaviour — not the name on the datasheet — is the only thing that counts; the standard always had to answer to the silicon. That instinct is what I brought to "agentic AI." These systems behave impressively, and it's tempting to grant them the language of agency because they act like they have it. But acting capably and being a genuine source of your own actions are not the same, and I think the field has been collapsing the two. What drew me to continuity as the dividing line is that it's the thing you can't fake with more capability alone. My hope is that the distinction gives people in law and governance a firmer footing — a way to ask who is really answerable when one of these systems acts, instead of quietly handing that answer to the machine.
Peter Kahl
Read the Original
This page is a summary of: How continuity distinguishes autonomy from agency in agentic AI, Discover Artificial Intelligence, July 2026, Springer Science + Business Media,
DOI: 10.1007/s44163-026-01675-5.
You can read the full text:
Contributors
The following have contributed to this page
