05 · Getting Models to Act
Agent
Agentic AI
A system that uses a model in a loop to pursue a goal: choose an action, call a tool, inspect the result, and continue or stop. Agents vary widely in autonomy, permissions, memory, and human oversight; persistence alone does not make them reliable.
Concrete example
“Research five competitors and build a table”: the agent searches, reads, extracts, and assembles without you steering each step.
Why it matters
The current frontier buzzword — treat claims skeptically, since reliability over long task chains is still genuinely hard.