AI Agent

An AI agent is an LLM augmented with three things: tools (functions it can call), memory (state it can persist across turns), and a planning loop that decides what to do next.

The minimum viable agent is roughly: receive a goal, decide on a next action, execute the action via a tool call, observe the result, decide again, repeat until done. Modern agents add layers — multi-agent collaboration, hierarchical planning, eval-driven self-correction — but the core loop is the same.

Operators evaluating AI agents should focus on three questions: what's the toolset (what can it actually do?), what's the reliability on real tasks (not demo videos), and how is the loop bounded (does it stop when stuck, or burn through tokens forever?). The pricing model often follows that third question — pay-per-token agents reward bounded loops, pay-per-task agents shift the risk to the vendor.