What is AI Agent?
AI Agent: An AI system that pursues a goal by deciding its own next steps, calling tools, reading results, and repeating until the task is done or it hands off.
An AI agent is an LLM placed in a loop with tools and a goal. Instead of answering once, it decides what to do next (look up an order, search a document, draft an email), calls a tool, reads the result, and decides again. It keeps going until it reaches the goal, hits a limit, or needs a human. The model provides judgment; the tools provide access to real systems.
What separates an agent from a scripted automation is that the path is not fully fixed in advance. That flexibility handles messy, varied inputs, but it also introduces new failure modes: wrong tool choices, loops, runaway costs, and actions taken on bad assumptions. Production agents need clear permissions, step and spend limits, evals, logging of every decision, and approval gates on irreversible actions.
Why it matters when you build or buy AI
"Agent" is the most overused word in AI sales. Many problems are better solved with a fixed agentic workflow where the model only makes decisions at specific points. Use a free-roaming agent when inputs are genuinely unpredictable and the cost of a wrong step is low or caught by review. Whatever you buy, ask what the agent is allowed to do, what it cannot do, and how you will see what it did.
Example
A support agent reads a "where is my order" email, looks up the order in Shopify, checks the carrier tracking API, and replies with the delivery estimate, escalating to a human if the parcel is marked lost.
Related terms
More in Agents & Automation
Book 20 minutes, or send a brief
A former CTO replies. You leave with a written plan and a fixed estimate.
Prefer LinkedIn? Connect with Harish.