What Are AI Agents?
An AI agent uses a language model as its "brain" to decide which actions to take, one step at a time, until a goal is reached.
Chatbot vs Automation vs Agent
| Chatbot | Fixed automation | AI agent | |
|---|---|---|---|
| What it does | Replies with text | Runs the same steps every time | Chooses its own steps |
| Example | "Explain RAG" | "Every night, email the sales report" | "Find why sales dropped and draft a report" |
| Handles surprises | Only in words | No, it breaks | Often, by trying another step |
| Risk | Low | Low, predictable | Higher, it acts in the real world |
The Parts of an Agent
- Model: the language model that thinks and decides.
- Instructions: the goal and rules (a system prompt).
- Tools: functions it can call: search, calculator, read file, send email, query a database.
- Loop: think, act, look at the result, repeat.
- Memory: what it has done so far, and sometimes facts saved for later.
A Tiny Agent-Like Program
This program is not intelligent yet, but it has the same shape: a goal, tools and a loop that stops when the goal is done.
Example
def check_stock(item):
stock = {"chargers": 4, "earbuds": 0}
return stock.get(item, 0)
def place_order(item, qty):
return f"Ordered {qty} {item} from supplier"
items = ["chargers", "earbuds"]
for item in items:
count = check_stock(item)
print(f"{item}: {count} in stock")
if count < 5:
print(" ->", place_order(item, 10))chargers: 4 in stock -> Ordered 10 chargers from supplier earbuds: 0 in stock -> Ordered 10 earbuds from supplier
Here the rule "order if fewer than 5" is fixed. In a real agent, the model would read the situation and decide, for example, "earbuds sell fast before Eid, order 20".
Real Examples of Agents
- Coding agents: read a project, change several files, run the tests and fix failures.
- Research agents: search many websites, read them and write a summary with sources.
- Browser agents: fill forms, compare prices and book things on websites.
- Customer support agents: look up an order in the database and start a return.
- Data agents: open a spreadsheet, clean it and make charts.
A Goal in Plain English
You give an agent a goal, not a list of steps:
Example
Goal: Prepare a one-page summary of our September sales.
You can use: read_sales_file, calculator, write_document.
Rules: Do not email anyone. Ask me before saving any file.The agent decides it should read the file, calculate totals and best sellers, then write the summary, and it asks before saving.
What Agents Are Not (Yet)
- They are not fully reliable. They can misunderstand, loop or take a wrong step.
- They are not always cheap: each step is another AI call.
- They should not be trusted with important actions (payments, deleting data) without a human check.
A good agent task is one where mistakes are easy to spot and undo, such as drafting, researching or organising, before you trust agents with bigger jobs.
Exercise
Exercise
Which part of an agent lets it act in the outside world, like searching or reading a file?