Nearly half of Australian workers are already using AI agents. Not AI assistants. Agents. Software that acts across multiple systems, makes decisions, sends things on your behalf, and keeps moving without waiting for you to click approve.

HP’s 2026 Work Relationship Index puts the number at 45%, ahead of the developed-market average. That’s not a pilot programme. That’s nearly every second person at a desk somewhere in Australia running automated multi-step tasks through AI, right now.

When one of those agents does something wrong, who owns it?

Brad Pulford, HP’s VP and MD for ANZ, put it plainly: “When someone uses AI to draft a document, that’s their own workflow. When agents handle multi-step tasks, it changes how work moves between teams, who signs off on what, and what a role is for.”

He’s right. And that shift is where most small businesses are flying completely blind.

An agent isn’t just a faster way to write emails. An agent can reschedule bookings, update records, send supplier quotes, move money between accounts, or push a file to a client. One instruction from an employee sets it going, and then it keeps moving through your systems and your teams without waiting for another nudge.

Now imagine it gets something wrong halfway through. Maybe it pulls the wrong client record. Maybe it sends a quote with last year’s pricing. Maybe it cancels the wrong booking and the confirmation email is already gone.

Who’s responsible?

The employee who set up the agent? The manager who approved using the tool? You, the owner, because it happened in your business? Or the vendor who sold you software that made a decision without asking?

Most small businesses have no answer. The KPMG AI Pulse from July 2026 shows 89% of Australian organisations expect AI to stay a top investment priority, with CEOs personally owning the strategy in 83% of cases. Yet knowledge workers are 19 percentage points behind IT decision-makers on agentic adoption. The tools are running ahead of the governance.

For big companies, that gap creates legal risk, compliance headaches, and board-level conversations. For small business owners, it creates something more immediate. A single agent mistake is very hard to undo and very hard to explain to a client.

You don’t need a committee to fix this. You need a decision, written down, shared with the people using these tools.

Specifically: which systems can an agent act on without human sign-off? What needs a person to approve before it moves? And if something goes wrong, whose name is on it?

Pulford again: “Security, governance and human judgment aren’t brakes on AI, they’re what lets you scale it with confidence.”

That’s worth printing out and sticking on the wall.

Agents are useful. Many small businesses are getting real work done faster because of them. But they work best with clear rules around what they can and can’t do alone. When something goes sideways, you need to know exactly who calls the client and what they say.

Set your accountability rules now. Before an agent does something you can’t undo.

More hot takes: Hot Takes →

📬 The SmallBizAI Brief

One practical AI tip for Australian small business. Every Tuesday. Free.

Join business owners getting smarter about AI every week.

Subscribe free →

Sources