So far, you’ve mostly been the person directing AI step by step.
You ask a question.
AI responds.
You decide what happens next.
An AI agent can take this idea further.
What Is an Agent?
Think about the difference between these two situations.
You say:
“Find me three hotels in Cape Town under R2,000 per night.”
AI gives you information.
You then decide what to do.
Now imagine saying:
“Find three suitable hotels in Cape Town under R2,000 per night, compare them against my requirements and prepare the best options for me to review.”
A more capable agent may be able to work through several of those steps itself.
Depending on the agent, it may be able to:
Search
Compare
Organise
Use other tools
Take certain actions
Return to you when it needs approval
The exact capabilities vary.
That’s important.
Think of an Agent as an Assistant
A useful analogy is a personal assistant.
You give the assistant a goal.
They work through the steps needed to achieve it.
They may use different tools along the way.
But there’s a difference.
An AI agent doesn’t automatically have human judgement or responsibility.
And you decide what it is allowed to do.
Start With the Job
This is the most important part.
Don’t start by asking:
“Where can I get an AI agent?”
Start with:
“What job do I want done?”
For example:
“I want help planning a five-day trip to Cape Town.”
Now break down what you’d actually want the system to do.
Perhaps:
Find flights
Find suitable accommodation
Compare car rentals
Keep my requirements in mind
Organise the options
If you want it to actually book something, that’s different.
It may need:
Access to booking services
Permission to enter information
Permission to make a purchase
Your approval before taking certain actions
Now you know what capabilities you’re looking for.
Does an Agent Already Exist?
You don’t necessarily have to build one.
Some AI platforms provide pre-built agents designed for particular jobs.
So you can search for something based on the job you need done.
But don’t simply look at the name.
Look at what the agent can actually do.
Ask:
What tools can it access?
What information can it use?
What actions can it take?
What can’t it do?
When does it need me?
That’s how you work out whether it really matches your needs.
What If Nothing Matches?
You have several choices.
An existing agent does the job.
Use it.
An existing agent is close.
See whether you can customise it.
Nothing suitable exists.
You may need a different AI approach, a custom-built agent, or decide that an agent isn’t worth using for that task.
The important thing is that you don’t start with:
“I need an agent.”
You start with:
“I have a job that needs doing.”
More Capability Means More Responsibility
An AI that writes an email draft is one thing.
An AI that can send emails is another.
An AI that can access your calendar is another.
An AI that can spend money is something else again.
The more an AI can do, the more carefully you should think about:
Permissions
Privacy
Approval
Checking
What happens if it gets something wrong?
Let’s Make It Practical
Choose one real job you’d like AI to help with.
Then work through this process:
1. Describe the job.
What do you actually want accomplished?
2. Identify the capabilities.
What would AI need to be able to do?
3. Look for an existing agent.
Search an agent marketplace or platform.
4. Inspect it.
Read what it can actually do.
5. Test it safely.
Start with something small and low-risk.
6. Decide.
Is it suitable?
Does it need customisation?
Or would another AI approach be better?
You’ve now moved beyond simply knowing what an AI agent is.
You’re learning how to decide whether one is actually useful to you.
What You’ve Learned
An AI agent can potentially work through several steps toward a goal with less step-by-step direction from you.
But the important skill isn’t simply knowing about agents.
It’s knowing how to go from:
Job → capabilities → find → inspect → test → decide
And as AI takes on more responsibility, another skill becomes increasingly important:
How do you know whether the work it produces is good enough?
That’s next.
Next Lesson