Generative AI makes things. Agentic AI does things by itself.
That’s the whole note I scribbled at Exponential Day 2026 (in Catalan, and with a typo worthy of a photographer: “argentica”). A bit more meat:
- Generative AI creates content (text, code, images) when you prompt it. It’s reactive: you ask, it answers, and a human looks at the output before anything happens with it.
- Agentic AI gets a goal, not a prompt. It plans the steps, uses tools (search, email, databases, other programs), remembers what it has done, checks the results and keeps going until the goal is met, on its own, within the permissions it has.
- They’re not rivals, they’re layers: an agent usually has generative models inside, with planning, tools and checking wrapped around them.
At the event, Albert Esplugas framed it as going from using AI to working with systems that collaborate with you (his talk: De la IA generativa a la IA agèntica). And on jobs, the same line Daniel Susskind gave in the keynote: AI doesn’t replace roles or people, it replaces tasks.
Chatbot vs intern, more or less. The intern you still have to supervise.