Agentic is the word you'll hear everywhere now, and it just means the AI doesn't stop at one reply. It plans, uses tools, and chains steps together — browse a web page, run a bit of code in a sandbox (a locked-down test area), fill in a form, call an API (the way apps talk to each other), then ping you when it's finished.
Don't picture a software team. Picture a small business owner. An agent might draft your deposit-policy email, pull the figure off your price-list doc, and queue a reminder to chase it — three steps in a row, not one block of text you then have to act on yourself. That's the shape of what's being marketed hard in 2026.
A grounded example you can poke at: Gertrude here on the site. She's a bounded demo — deliberately not unlimited ChatGPT. What she shows you is the personality and capability a custom assistant can have when someone actually builds one. Members learn to build their own — their scope, their rules.
Why agents matter for your job specifically: they don't just suggest the email, they can draft it, look up the detail, and queue the action. That's exactly why desk-bound work is in the crosshairs — which is the whole of Lesson 5, coming up.
And the honest limit, because I won't hype it: agents fail. They loop, they mis-click, and they burn real money in API calls while doing it. The slick demos hide that; anything running in the real world needs guardrails and a human approving the important bits. The course teaches exactly that for your own builds later — it's not an afterthought.
Continue — how search is changing.