AI agents are moving fast. Here's what Australians should understand before using them

Over the past year, most Australians have become familiar with generative AI tools like chatbots. You ask a question and it gives you an answer. You prompt it and it creates something.

What is changing now is that AI is starting to move beyond that.

We are seeing the rise of what is called agentic AI, or AI agents. These are systems designed not just to respond, but to act.

As Director of Offensive Security for Gen, Stephen Kho explains, "agentic AI is AI with hands and feet."

Instead of asking AI to write an email, you can ask it to read your inbox, prioritise messages, draft replies and send them. Instead of helping you plan your day, it can access your calendar, move meetings and prepare notes for you.

"That's where we are now," he says. "It can do things on your behalf."

AI agents are not just smarter chatbots

A lot of people still think of AI in chatbot terms. That is understandable, because that is how most people have experienced it so far.

AI agents are different.

They can take multi-step actions, use tools and interact across accounts and systems to complete tasks on your behalf. 

A simple way to think about it is this. A chatbot responds. An AI agent acts.

Kho compares it to having extra help. "It's like suddenly having 100 interns acting as you," he says. "They're doing all these things on your behalf."

That is what makes it powerful. It is also what changes the risk.

The real consideration is permissions

The key question is not just what an AI system knows. It is what it is allowed to access.

For an AI agent to be useful, it needs access to tools, systems and data. That can include your email, files, calendar and sometimes financial accounts. 

"The more access you give it, the more helpful it becomes," Kho explains. "But that also means it can do a lot more."

In enterprise environments, this level of access is tightly controlled. There are guardrails and limits in place.

For everyday users, that is not always the case.

"It's very easy to just connect everything and go, this is great," he says. "But you might not fully understand what you've just given it access to."

The skills ecosystem is where this becomes real

Most AI agents become more useful through add-ons, often called skills.

These are small programs that extend what the agent can do.

Kho describes this environment in simple terms. "The skills market is like the App Store before it got controlled," he says.

Some skills are useful and well built. Others may be insecure or even malicious.

That means something that looks helpful on the surface could be doing something very different in the background.

This is not unique to AI, but the level of access involved makes the impact greater.

The security industry is already starting to respond to this shift. Gen recently introduced Sage, a protection layer designed to monitor AI agent activity and assess potentially risky actions before they are executed.

"The goal is not to stop people using these tools," Kho says. "It's about putting guardrails around systems that are becoming more autonomous."

Scammers are moving into the same space

Generative AI has already improved the quality of scams. Messages are more convincing and easier to produce.

AI agents introduce something else. Speed and scale.

"The barrier to entry and the speed is something we've never seen before," Kho says.

Instead of manually creating and sending scam messages, these systems can generate content, build fake websites and distribute messages as part of a workflow.

It also changes where the risk sits.

"If something is acting on your behalf, and that gets manipulated, that's a different problem," he explains.

Why this is becoming harder to navigate

Even for people working in technology, the pace of change is difficult to keep up with.

"This space is moving so fast we're barely keeping up," Kho says.

For most Australians, it can feel like being pulled into something that is constantly evolving.

At the same time, adoption is happening quickly. People see others using these tools and want to try them.

That creates a gap between usage and understanding.

A practical way to approach it

AI agents can be genuinely useful. In many cases, they can save time and reduce repetitive tasks.

The key is to approach them with awareness.

Be selective about what accounts you connect. Be cautious about installing third party skills. Take the time to understand what an agent is allowed to do.

Kho puts it simply. "If you wouldn't give that level of access to a person, it's worth pausing before giving it to a system."

AI agents represent a shift from tools that assist to systems that act.

That shift is already underway. Understanding how they work is the first step to using them safely.