Agentic AI
An AI system able to plan and execute autonomous sequences of actions to achieve a goal, beyond mere generation.
Agentic AI refers to a composite system whose central engine is a large language model but which differs from passive generative AI by its ability to plan and execute autonomous sequences of actions to achieve a goal. Where a classic generative model answers a query then stops, the agent breaks a goal down into sub-tasks, uses external tools such as a browser, a terminal or an API, analyzes the results of its actions and self-corrects without intermediate human intervention. This shift from a machine that generates to a machine that acts is a major economic break: the algorithm can now replace the human in coordination and execution functions, not merely assist. For the risk manager, autonomy changes the nature of the error: faulty reasoning becomes an act, a signed contractual commitment, a completed transaction, exfiltrated data. It is precisely this ability to act autonomously, coupled with the model's opacity and its vulnerability to manipulation, that makes algorithmic execution hard to frame legally and to insure through traditional mechanisms.
Textbook case: an agent tasked with auditing insurance policies navigates a document base on its own, extracts the relevant clauses, drafts a report and sends it, chaining these steps without human validation at each transition.
agentic AI, AI agents, autonomous agents
Follow the research
One letter a week. First, the week's underwriting note, when one is published: a dated, argued decision, signed by Paul Bernardski. Then what the observatory published on cyber risk, political risk and insurance. No account, no tracking.
No spam, one-click unsubscribe, no data ever sold.