Software that decides its own next step
AI stopped writing and started acting. An agent is a model with a goal, tools, and permission to loop - software that decides its own next step. Wire one into real systems and it can move money, change records, and email your customers.
And it's non-deterministic. Every program you've ever trusted behaved the same way every time - test once, trust forever. An agent runs the same task twice and takes two different paths. The trust model software was built on is dead. Nothing replaced it.