AI agent
A model given tools, permission and a loop, so it can act.
Until now a wrong answer cost you the time to read it. An agent can send the email, move the record, run the script or spend the money. The capability is real, and what matters is the change in cost: being wrong stops costing attention and starts costing consequences.
It pays on high-volume, low-stakes, reversible jobs: tagging, routing, drafting, formatting. It does not pay where an error compounds silently, touches money or identity with nobody watching, or cannot be reconstructed. An agent that took forty steps to a wrong result is only fixable if it logged the forty steps.
Find your way around
Learn it in the course, copy it from a playbook, see it work in a case study.