also: agentic AIalso: autonomous agent
What is AI Agent?
Last updated · 1 min read
An LLM answers; an agent acts. Give it a goal ("find me three flights under $400 and draft the approval email") and it loops: plan → call a tool → read the result → adjust → repeat until done.
The catch is compounding: each step has some error rate, and a 20-step chain multiplies those misses — which is why reliable agents are built from narrow, well-scoped tasks plus human checkpoints, not vague "do my job" mandates. Agents plus RAG and tool access is where most of the industry's energy is going.