What is an AI agent?

Chatbots answer, agents act. They open tickets, run queries, and push changes to production.

In this video, Ido Shlomo explains what makes something an agent: a model, the tools it calls, and an orchestration layer that runs everything in a loop. The loop gives the agent persistence, memory, and autonomy, so when one path is blocked, it tries another. Give it enough permissions, and it can do a lot without asking.

So how do you secure something like that? Ido's answer: at the action, the same place you secure people. Every time an agent acts, it uses an identity, and access management is something security teams already know how to do.

Ido also covers the three caveats that make agents harder than humans: ownership, enforcement, and attribution. Ideally, each agent has its own identity, sized for its purpose. In practice, it often rides on its creator's session, or an OAuth grant someone approved once..