Glossary
AI agent
An AI agent doesn't just answer. It takes actions for you: browsing sites, filling forms, running multi-step tasks with limited supervision. More powerful than a chatbot, and worth supervising precisely because it acts.
A regular AI assistant is a brilliant advisor: it tells you how to compare ferry prices. An agent goes and does it: opens the booking sites, checks dates, fills the forms, comes back with options. You give a goal; it plans and executes the steps.
Think of the difference between a recipe and a cook. By 2026 agents book tables, do deep research across dozens of sites, manage email triage, and operate a computer like a careful intern.
“Careful intern” is exactly the right level of trust. Agents inherit AI’s weaknesses, like misreading a page or picking the wrong button, except now the mistake is an action, not just a wrong sentence. Sensible rules: let agents act freely on low-stakes tasks, require your confirmation before anything involving money, accounts, or sending messages, and be deliberate about what data and logins you hand them.
Where you’ll meet this
ChatGPT’s agent mode, Claude’s computer-use and agent features, Gemini’s agentic tools, plus browser agents like Comet. Before connecting an agent to your email or files, think about what it will see along the way. Our paste checker logic applies doubly here.