Agent
Receives a goal, decides on actions, uses tools, observes, and iterates. A chatbot only responds.
In one sentence
AI system that runs in an autonomous loop: receives a goal, decides on actions, uses tools, observes results, and iterates until it completes or escalates. Unlike a chatbot, which only responds.
Before
Goal received
What it does
Thinks, uses tools, observes
After
Result delivered or escalated
Example
Ask a chatbot to "write Ana's letter" and it writes using whatever is in front of it. An agent opens the field report, notices the school attendance grade is missing, checks the records, writes it, verifies the text against the source, and escalates to a human because the child has moved to another city.
The final text may even look similar. What changes is the number of decisions the agent made on its own.
In practice
- Start with a script you run manually. Automate the trigger only after the result stabilizes.
- Give it the minimum set of tools needed to solve the task. One extra tool is one more surface for error.
- Log from the very first run: input, output, tools called, cost.
- Set the iteration ceiling before running it. Without a ceiling, a confused agent turns into a high bill.
How to make it tangible
A running service: model, system prompt, tools, and iteration ceiling. It starts as a script you run manually.
Translated from Portuguese with AI assistance.