Agent runtime

AI Agent

A language model given tools, context, and a goal — so it can take actions and check results across multiple steps, instead of producing one answer and stopping.

Operating principle

Production AI is a system: context, tools, permissions, traces, evals, and feedback loops around the model.

What it is

An AI agent is a language model put in a loop: it receives a goal, looks at the context, decides on an action (call a tool, ask a question, write a result), observes what happened, and decides again. A chatbot answers; an agent works. The word covers everything from a single automated step to a long-running process with memory and approvals.

Why it matters

Agents are where AI stops being a writing aid and starts being labor: triaging tickets, drafting replies, updating records, monitoring systems. That shift also raises the bar — an agent's mistakes are actions, so production agents need permissions, approval gates, traces, and evals around them.

How it works

The loop is: assemble context, present tools, let the model choose, execute the choice, feed the result back, repeat until done or escalated. The engineering lives in what surrounds the model — the agent harness, the tool registry, the runtime that makes the loop restartable and observable.

Related resources