Concepts

What is an agent

If you've ever used ChatGPT or Claude.ai, you know what a chatbot is. An agent is different — not on the surface, but deep down. That difference radically changes what you can automate.

A chatbot answers. An agent acts.

A chatbot is reactive: you type a question, it answers, end of exchange. It's a dialogue.

An agent is active: it has a mission, it has access to tools (read a file, send an email, fetch a web page), and it decides on its own the order in which to use them to accomplish that mission. It can run for 30 seconds or 30 minutes, depending on the complexity of what it's asked to do.

Three fundamental characteristics

1. Autonomy

The agent receives an objective, not a step-by-step instruction. "Make me a monthly sales report" vs "open this file, sum column C, write the total into D2". It's the agent that breaks the objective down into steps.

2. Tools

Without tools, an agent is just a chatbot talking into the void. With tools, it can act on the world: modify files, run commands, call external APIs, read your Notion, send Slack messages. See MCP & tools.

3. Reasoning loop

The agent doesn't "answer" in a single pass. It thinks, it acts (uses a tool), it observes the result, it thinks again, it acts, and so on until it's done. That's what lets it handle tasks a chatbot couldn't — for example debugging code by running tests, reading the errors, and fixing them.

A few concrete examples

To get a feel for the difference, here is what a chatbot cannot do but an agent can:

  • Weekly monitoring: visit 5 URLs, read the content, identify the changes vs last week, write a summary email and send it. All by itself, Monday at 9 a.m.
  • Inbox triage: connect to Gmail, read the unread messages, sort them by priority, draft replies for the urgent ones, archive the newsletters.
  • Sales reporting: open the monthly CSV, detect anomalies, compare to previous months, generate an illustrated PDF.
  • Code review: pull a GitHub PR, read the diff, detect likely bugs, write inline comments.
Is the agent always right?No. Like a human, an agent can be wrong, hallucinate, or misinterpret an instruction. For critical tasks (payments, public sends, deletions), always plan for a human validation step. See also our FAQ.

When to use an agent vs a chatbot?

The simple rule:

  • Chatbot if you want an interactive dialogue where the human drives each step.
  • Agent if you want to delegate a complete task that runs without you, possibly in the background.

Forgent only does agents. If your need is conversational, use Claude.ai directly instead.